A live first-impressions review of GPT-5.6 SOL after roughly a month of hands-on testing, evaluating it as a tool for knowledge work. # Channel Identity Observations ## Channel / Organization Identity - **Name: "Every"** — the host identifies himself as "host Dan Schipper, co-founder, CEO Every" (transcript opening), later directing viewers to "every.0" for the published review. - **Team roster presented on camera:** - **Dan Schipper** — co-founder, CEO - **Ariel** — "my sister Ariel... Ariel leads operating room activity in Every" - **Katie Parrot** — recurring contributor, "star of the hour"; wrote the written version of the review ("We have a written version that was written Katie") - **Mike Taylor** — "manager of the technology consulting department at Every" - **Product portfolio self-described by the host:** - "We are Spiral – agent for writing texts" - "Kora – agent for working with electronic mail" - "Sparkle – program for organizing files" - "Monologue – program for transformation speech to text" - Plus camps, courses, and enterprise consulting ("we cooperate with large companies, to help them implement artificial intelligence") ## Claimed Scope and Positioning The channel makes an explicit, repeated self-description (stated at least three times in the transcript): > "We are the only subscription that you need to stay on the edge of AI... You can imagine us a little like a laboratory for advanced research [into] the future of work with knowledge." Demonstrated scope claims include: - **Early model access**: "We get models before they leave [public release]" and "when the models... we are holding for you vibe checks, as in a live broadcast." - **Multi-domain testing**: "we take these models, we test them on writing, coding, design, marketing, PowerPoint... everything that you can imagine, and we use this for our real work." - **A published review pipeline**: "Today we published our review of the GT 5.6 Soul, our vibe check... We have a whole team of people who protested [tested] this." ## Intended Audience - **Knowledge workers, not engineers**: The core framing is AI for non-technical professionals — "the first time when... if you are not [a] technical man, and you... are fascinated by AI, but maybe you don't program... you are starting to understand this." - **Subscription subscribers and prospective ones**: Direct address to viewers ("If you haven't yet subscribed... it's worth signing up"), plus interactive elements (reading viewer questions from "Anom" and "Harold 33," live comments). - **Business/enterprise stakeholders**: A guest from OpenAI mentions recurring "calls regarding deployment with some of our corporate customers," and Kyle's finance-automation demo targets operational/business users. ## Observable Presentation Patterns - **Live "vibe check" format**: A live-streamed group reaction to a same-day model release, with sunglasses as a running visual gag ("Soul means sun... that's why we wear sunglasses"). - **Guest-heavy format**: Multiple OpenAI guests appear ("Dom... lead developer," "Kyle Cobber," "Claire Waugh, founder of Chat PRD"), giving the channel a scene/insider-access character. - **First-person workplace use cases as evidence**: Ariel's 46-CSV benchmark, Dan's food-journal and Facebook Marketplace agents, Kyle's month-end close automation — the channel's stated methodology is "real vibe checks from real people who constantly use these." - **Casual, family-flavored tone**: Sibling hosts, calling their mother during the stream, banter about pronunciation of "Soul/Sol." ## Caveats - The transcript appears to be a rough machine translation; wording of quotations may differ from the original audio. - The claim of being "the only subscription you need" is a creator marketing claim, not an independently verifiable fact. - All observations derive from a single video and should not be generalized to the full channel without corroborating material. # Delivery Observation: GPT-5.6 SOL Vibe Check Livestream > **Note on evidence quality:** The transcript appears machine-transcribed and partially translated (mixed English/Ukrainian passages, garbled product names like "ChatBT," "Codecs/Codedex," "Soul/Solu"). Quotations below reflect the transcript as given. This degrades quote precision but the *patterns* remain observable. ## Format & Structure - **Panel rotation with host-directed handoffs.** Dan consistently cues each speaker with a near-identical formula: *"Could you please tell us a little bit about what you are for you use it"* and later *"share their own, uh, general impressions from 5.6, Please."* - **Recurring promotional interstitials.** The subscription pitch repeats nearly verbatim at least three times (~[76], ~[764], ~[1910]): *"this is the only one subscription that you needed to stay on the edge of AI"* and *"you can imagine us a little like a laboratory advanced research future of work with knowledge."* - **Structured segment plan, loosely followed.** Dan announces coverage areas ("coding, in we have writing... almost everything") and does move through them — overview → personal use cases → coding → merge discussion → guest demos — though transitions are conversational rather than rigid. - **Live audience interaction.** He reads viewer questions aloud and names them: *"we had very interesting and important question from Anom, who said: 'What the hell is this?'"* and *"I see question from Harold 33."* ## Teaching Habits - **Concrete personal demos over abstractions.** Nearly every claim is backed by a specific artifact: Ariel's *"46 different CSV files"* benchmark; Dan's food-journal agent that checks Monologue voice notes and the photo app; Kyle's month-end close automation that shrank *"almost 5 working days to five hours."* - **Named frameworks the team repeats across speakers.** The phrase *"I don't do work alone; I am building a system that performs work"* appears in Dan's segment and is echoed in concept by Kyle ("codify this as... skills") and Claire (*"create systems, who do the work, but don't do it independently"*). This looks like a deliberate house vocabulary. - **Explicit scoring/quantification.** Dan gives numbers when evaluating: *"he got 56 out of 100. Fable got 90 out of 100"*; Kyle describes a learning curve: *"from the first time it can be about 70%... in a month or two, it approximately 90, that is, 95."* - **Head-to-head comparison as a teaching device.** Dan and Claire both show same-prompt outputs side by side: *"I asked GPT 5.5 to do scheme... and then I asked Fable to do the same... this is also GBT image model, but you can just another level."* ## Tone & Register - **Casual, self-deprecating banter.** Dan jokes about his own forgetfulness (*"Ден забуває їсти з дитинства"*), the sunglasses bit ("Soul means sun... that's why we wear sunglasses"), and calling his mom mid-stream: *"I'm on the stream, Mom."* - **Sibling teasing as recurring texture.** Ariel and Dan exchange jabs: *"Sorry, I'm a bit confused... when it comes to things reaches mathematics... This is a family. This is genetic thing."* - **Enthusiastic superlatives from the host.** *"gold standard for working with knowledge"*, *"Fable—this is something else. It's like tactical nuclear weapon"*, *"This, according to essentially... it's too exciting."* Hedging coexists with hype — Dan corrects his own benchmark: *"I think this score actually not representative... if I am going to choose number... it will be something like 70 or 80."* ## Recurring Phrases & Vocabulary - **"Vibe check" / "vibration checks"** — the team's signature term, used at least six times. - **"Lost in the weeds"** — used independently by Ariel and Dan about Fable: *"it's like a little lost in weeds."* - **"Warm paper tones"** — repeated by Dan and Claire: *"she likes warm paper tones"*; *"I simply I like warm ones. paper overtones."* - **"Contagious"** — Ariel: *"his love for CodeEx absolutely contagious"*; Mike: *"your contagious enthusiasm regarding CodeEx."* - **"Daily driver" (translated as "daily car/tool")** — Mike: *"now he's mine daily car"*; Claire: *"this is completely mine everyday tool."* ## Interaction Patterns Between Speakers - **Active affirmation.** Speakers validate each other mid-stream: *"I agree. I am completely I agree"* (Dan after Katie); *"I had the same thing"* (Dan after Mike); *"I just said that I support this"* (Katie). - **Host tees up guest credentials.** Before Mike speaks, Dan frames him: *"people should know that you historically were heavy user Cloud."* Before Kyle: *"we have a very special guest."* - **Playful disagreement is voiced directly.** On Fable's design: *"First, we will have to to quarrel because I think that... creative and designer things Fable simply amazing"* (Dan to Claire). ## Uncertainty & Caveats - **Single-video evidence.** All patterns here come from one session; stability across other videos cannot be assessed. - **Translation noise.** Some "repetition" (e.g., mangled product names) may be transcription artifacts rather than the speaker's actual usage. - **Sponsorship dynamics.** OpenAI employees appear as guests; the uniformly positive framing of 5.6 relative to competitors is observable but its cause is not — no inference is made here. - Some patterns (e.g., "contagious") appear only twice; flagged as thin evidence. # Video Observation Record: GPT-5.6 "Sol" Vibe Check Livestream ## Source Overview - **Video:** "GPT-5.6 SOL: THE GOLD STANDARD FOR KNOWLEDGE WORK" (Every production) - **Format:** Live "vibe check" of OpenAI's GPT-5.6 model, branded "Sol/Soul" (sun-themed; hosts wear sunglasses as a running joke) - **Caveat:** The transcript is a rough, apparently machine-translated rendering with frequent garbling; names of models, tools, and people (e.g., "Fable," "CodeEx," "Opus," "Enthropic") are recorded as they appear in the source and may be distortions. Confidence in quotes is accordingly limited. - **Speakers identified from transcript:** - **Dan Schipper** — host, co-founder/CEO of Every - **Ariel** — Dan's sister, leads operations at Every - **Katie Parrot** — Every team member ("star of the hour") - **Mike Taylor** — head of technical consulting at Every - **Dom** — OpenAI guest (introduced as a lead developer) - **Kyle Cobber** — OpenAI guest (finance context) - **Claire Waugh** — founder of Chat PRD, host of "How I Became an AI" (garbled), final guest --- ## Claims and Positions by Speaker ### Dan Schipper **Core claim:** The combination of GPT-5.6 with ChatGPT CodeEx is "the gold standard for knowledge work." - Reasoning stated: the combination offers "power, productivity, amenities of use and speed." - He positions **Fable** differently: "like a tactical nuclear weapon... used for big tasks," and says he uses 5.6 as "an incredible subagent for Fable." - Describes OpenAI's design philosophy claim: instead of "creating the maximum smart model that is so smart it's really hard to use," OpenAI created "something fast, really productive and convenient." - Notes the **app merge**: ChatGPT and CodeEx desktop apps merged into one app with two sections — "ChatGPT Work" and "ChatGPT CodeEx." - Shows personal automation examples: a CodeEx-driven food journal built from Monologue voice notes and his photo app; a Facebook Marketplace listing agent. - Benchmarks: reports 5.6 scored ~56/100 on a coding test where Fable scored ~90/100, but he **qualifies** this as "not representative" and says his gut impression would be 70–80; attributes the low score to 5.6 not deleting/simplifying enough code. - **Contradiction/tension visible:** Dan says Fable is "generally a better programmer, but... less convenient in use." ### Ariel **Core claim:** 5.6 showed a large practical jump over 5.5 for operations work; accuracy and multitasking matter most for her role. - Benchmark anecdote: consolidating **46 CSV files** into one analysis. - 5.5: "huge failure" — unusable result. - CodeEx with 5.6: found the right email, reviewed all data, asked smart clarifying questions, produced something that "passed the sniff test for accuracy." - Fable: "got stuck in the weeds," claimed a connector didn't work, wanted files downloaded to disk — though its summary tab was "useful and interesting." - Says CodeEx became her "main work surface" and calls AI a "brain prosthesis for mathematics." ### Katie Parrot **Core claims:** - "I love 5.6." Describes distress when access was temporarily lost (references a government-related access interruption, and considering Opus/Sonnet alternatives) and relief at its return. - Praises it as "very ergonomic... for quick decisions, quick execution, iterations, revisions, many options." - Notes it handles mid-course steering well — taking extra direction "but remaining focused on the bigger task." - **Contrast with Anthropic models:** she finds them "prone to excessive fixation," making the product less convenient; 5.6 keeps her "in motion." - Reports making ~24 drafts during the vibe-check writing process, noting "review is so cheap" with the model. - Notes the bug: after updating, the CodeEx app icon disappeared due to the merge. - Later flags enthusiasm for the local browser extension and cookie/importer features enabling Google Docs work. ### Mike Taylor **Core claims:** - Moved aggressively from Fable to Sol as his daily driver; Fable is now ~20% of his use (previously ~80%). - Two things won him over: (1) the in-app browser for working on shared documents, (2) the "router" feature for stream-to-stream messaging, solving an orchestration problem he'd spent significant time on ("even passed a two-day intensive course in Gastown"). - **Qualification:** he still thinks "there's something special about Fable" and calls it one of the smartest models; he jokes it behaves like his editors — telling him he squeezes too many ideas into one post. - Describes his "eval pack" practice: asking the model to generate an HTML page with before/after code blocks and evidence after complex tasks. - **Design critique:** Sol's design (warm paper tones) is now everywhere and "looks like slop" from overuse — though he admits he likes it; says design improved from 5.5 to 5.6 but "there's still a lot of work to be done" and notes Fable still produces higher-variance, sometimes better creative design. - PowerPoint demo: 5.6 produced branded output with custom SVG elements; notes it was "the first time an OpenAI model ignored me and did its thing, and it's better." ### Dom (OpenAI) **Claims about the launch:** - 5.6 is "an extremely new Frontier model, available to everyone," used internally for weeks across finance, recruiting, and sales. - The merge message for developers: "this is the same CodeEx that you know and love... We just make it better." - Positions non-coding use: engineers also do documents, research, and source-checking — tasks that previously required copying between apps. - New features cited: CodeEx documentation rebuilt in real HTML from screenshots tied to the codebase; a "visualization" capability producing interactive demonstrations (he showed it rebuilding interactive elements by checking video footage via the Chrome extension); a Chrome extension chat with local filesystem access; "Sol Ultra" for "your most ambitious tasks." - **Positioning claim:** the model is more autonomous — it can run 5–6 tasks in parallel on your computer and "naturally checks its own work." ### Kyle Cobber (OpenAI, finance) **Core claims:** - Calls it "an operating system for knowledge work"; he is "not an engineer" but has effectively worked as one since February by integrating it into workflows. - Detailed example: monthly financial close automation previously took ~5 person-days, now ~5 hours — **with qualifications:** "it's not ideal" and requires iteration; you can't build it once, you "teach" it over weeks. - **Trust and hallucination discussion (in response to audience question):** - Acknowledges two persistent problems: hallucinations ("significantly decreased but still occurs") and, more commonly, **omissions** ("maybe it just didn't view this file or didn't receive this email"). - His method: don't expect perfection on first pass; encode corrections as skills (e.g., an "audit skill" that checks discrepancies between memos, slides, and spreadsheets before sending). - Estimates a growth curve: ~70% quality on first use rising to ~90–95% after a month or two of skill-building. - Advice to knowledge workers: don't try to automate everything at once; pick the most painful process, codify it via skills, and "recursively improve your skills." ### Claire Waugh **Core claims:** - "I love 5.6" — values **efficiency over everything else, perhaps even pure intelligence**; her tests focus on whether the work actually gets done. - **Direct disagreement with Dan on design/creativity:** she finds **Fable takes fewer creative risks** while 5.6 is "a little more original" — the reverse of Dan's and Mike's framing. She showed head-to-head image comparisons where she preferred Fable's outputs in some cases ("Fable is amazing" at creative/design things) while crediting 5.6 for being the only model willing to deviate. - On writing: "I think 5.6 is a good writer, but... I still want to talk only to Sonnet" as a personal agent. - **Framing claim:** "5.6 seems optimized for the human interface of the model, while Fable is optimized for the model-to-model interface." - On the broader race: both companies' models are "very powerful, very intelligent"; she sees OpenAI optimizing for non-technical/medium-technical users and the other camp for deterministic, technical use — "almost completely opposed" strategies. --- ## Recurring Themes and Points of Disagreement - **Usability vs. raw intelligence:** Dan, Katie, and Mike praise 5.6's speed and ergonomics; Dan and Kyle explicitly concede Fable remains the stronger pure programmer. - **Creative/design split:** Claire credits 5.6 with more creative risk-taking; Mike credits Fable with higher-ceiling creative variance — an unresolved disagreement. - **The agent-cyclic workflow:** multiple speakers (Dan, Ariel, Kyle, Dom) independently describe the shift to "building a system that does the work" — delegation, skill-building, and review loops — as the significant change for non-technical knowledge workers. - **Persistent limitations acknowledged:** over-rewriting code (Dan's benchmark), design monotony ("slop," Mike), hallucinations and omissions (Kyle), and background-mode interruptions of active work (Katie/Dan exchange). ## Unresolved Tensions Preserved - Whether 5.6 or Fable is more "creative" remains contested between Claire and Dan/Mike. - Dan's coding benchmark score (56/100) is presented but self-disqualified as unrepresentative — never reconciled. - Claude/Sonnet retains stated loyalty from both Katie (writing "vibe checks") and Claire (personal agent chat), despite overall preference for 5.6. # 视频来源观察 **视频标题：** GPT-5.6 SOL: THE GOLD STANDARD FOR KNOWLEDGE WORK **视频 ID：** 01a0f15a-6652-73ce-a6f7-4930fd36dc0e ## 转录质量说明 - 转录内容存在严重的乱码或机器翻译痕迹（ 英语、乌克兰语片段混杂，多处语句碎片化）。所有发言者和引语仅能归因到现有明确可辨识的范围内。 - 未提供策划的元数据（仅提供了标题和转录内容）。 ## 发言者和角色 ### Every 主办方 - **Dan Shipper** — 主持人；Every（Every Inc.）的联合创始人兼 CEO。开场介绍，主导流程，分享对模型的初步印象。 - **Ariel** — Dan 的姐姐；根据 Dan 的介绍，负责 Every 的“运营活动”。讨论了将模型用于运营和数据分析工作。 - **Katie Parrott** — 被描述为 “star of the hour”；据称撰写了已发布的关于该模型的书面回顾。分享了作家的视角。 - **Mike Taylor** — 根据主持人的介绍，担任 Every 技术咨询主管。讨论了编码工作流和设计示例。 - **Austin** — 被提及（非发言）为 Every 的开发负责人。 - **Kate** — 被提及（非发言）为 Every 的编辑，她构建了一个应用程序。 ### OpenAI 嘉宾 - **Roman** — 被介绍为 OpenAI 的 "head of DevRel"（开发者关系）。讨论了 ChatGPT/Codex 的合并、文档和功能。 - **Dom** — 被介绍为 OpenAI 的 "lead developer"（在此转录中名字拼写不清，可能为 "Dom"）。讨论了模型发布和功能。 - **Kyle Cobber**（拼写根据转录，未经验证） — 被介绍为 OpenAI 的财务专家。展示了内部的财务自动化工作流。 - **Claire Waugh**（拼写根据转录，未经验证） — 被介绍为 **ChatPRD 的创始人**，以及播客 *"How I Became an AI"* 的主持人；被描述为外部特别嘉宾，而非 OpenAI 员工。 ## 引用及第三方声明 - **第三方模型/产品引用：** GPT-5.6 ("Sol"/"Soul")、GPT-5.5、Fable、Opus 4.8、Sonnet 5、Claude Code、Codex、ChatGPT、Dia 浏览器、Chrome 扩展、Google Docs/Slides/Sheets、Excel、Slack、Apple Photos、Facebook Marketplace、Monologue、Spiral、Kora、Sparkle（Every 应用）、ChatPRD。 - **第三方声明：** - 提及 Howard Lutnick 与暂时禁止访问该模型有关（发言人声称，未验证）。 - 一位名为 "Anom" 的观众提交了实时问题；观众 "Harold 33" 就 Codex 的可用性进行了提问。 - Claire 引用了个人的基准测试/对比；Mike 引用了编码测试的“56/100 vs 90/100”分数（非正式基准，由发言人自行描述）。 - **Every 的书面输出：** 提到了由 Katie Parrott 撰写的 GPT-5.6 书面 Vibe Check 回顾，发布在 Every 网站上。 ## 推广部分（标记） - **反复的 Every 订阅推广：** 将 "all is one" 订阅描述为“保持在先进 AI 前沿所需”；将 Every 描述为“知识工作实验室”。 - **Every 产品推广：** Spiral（写作代理）、Kora（电子邮件代理）、Sparkle（文件组织）、Monologue（语音转文字），以及为订阅者提供的营地/课程和企业服务。 - **“无耻广告”自我标记：** OpenAI 嘉宾为 Codex 指南进行了推广；从声明本身的文本中明确标记为推广内容。 - **推广链接/目的地：** every.to（网站）、learn.gptt.com（OpenAI 文档）。 ## 可用性与访问说明 - 嘉宾加入了有限的直播环节；主持人在结束前指出与 OpenAI 嘉宾的时间紧张。 - 转录中未说明评论可用性。 - 转录内容在 Claire Waugh 的结束语处中断（"Claire, if people want—"）；未捕捉到完整的结尾。 And we are here. Welcome to our GPT 5.6 test Soul Soul Vibe. We don't know. but it's about the sun. That's why we are in sunscreen glasses. I am yours. host Dan Schipper, uh- e, co-founder, CEO EveryY. This is my first time here. with his sister Ariel. Greetings. Greetings. Ariel leads operating room activity in Every. And in We, as always, have Katie Parrot, star of the hour. Katie, how are you? IN I'm fine. I, You know, you better call. Solu, this is what I I repeat to myself again and again. Hmm, maybe this is it. should have gotten to checking vibes, but uh, that didn't happen . That would be a good phrase. Uh, I have a daily knowledge tasks, Better call Saul. Um, uh, and are we actually we say Sol or Soul? How is this? pronounced? Hmm, ah now you're forcing me to feel very I'm embarrassed, too. It seems that when you you pay attention to this, I will like it. ...I...I...I can't to tell you. I am afraid. Let's just say soul. I will say soul. Good. To us better call it that soul. Invite someone from Open Eye team so that we talked about our pronunciation and others more interesting things. So, if you are here, you should know that we... "all is one" subscription that you needed to to stay on advanced AI". We we create ideas, applications and we teach. As for ideas, so every day we publish something about how we we use AI, as we use him in our work, how we use him in our lives. We get models, before they leave . Today we published our review of the GT 5.6 Soul, our Vibe check. Look at this. It you can say that we are thinking about this model, because on in the background Vibe checks are something like sports Porsche. It I really mean it. I think. Hmm, that's good, that's very nice model. Um, we missed her, until she was gone. We, uh- uh, Howard Lutnick, uh, banned us from her use at all not for long. Not us personally, simply in general. Uh, and then we returned it and, uh, very happy that she returned. Very glad she came out today. We have full check the vibrations of everyone...Well, just go to every.0. This will be on main page. IN We have a coverage test. We have coding, in we have writing, we have there is almost everything that is possible to expect. And we we are going to spend , uh, next hour- two, discussing our experience with this model. Uh, we have there will be some people from OpenAI. We will have a novel. from OpenAI, head of Devril. IN there are a few more of us with OpenAI, who will come and will come out, and we are here to tell you about this model. So, uh, I I'll just start with a small overview, at least my thoughts . And then everyone will share his/hers thoughts. But in general, I think version 5.6 is in Chat GPT, CodeEx and Chat GBT works, that is, they merged ChatBT CodeEx, they merged CodeEx...I I can't even do this. say. They placed CodeEx in ChatBT, but then ChatT in CodeEx, hmm ... this is too exciting, too exciting, uh, therefore, uh, obviously, in they were separate branches , they had a desktop ChatBT app and desktop application CodeEx, and they are them merged, so now there is one desktop an app called ChatBT called CHBT, and it has two sections: CHBT Work and CHBT Codecs. This, according to essentially, just codecs from working side. Yes, let me just to show you it. With working side they, uh, hid all the code, and that's it. Ahem, so, I am doing this now I will open. You can see if you see my screen. Um, I I will zoom in. If I'll zoom in, Katie, Do you see? Yes. Good. So we have chatbt codeex, chatbt work for developers who, I I think they don't do it. work. We need to practice a little on this occasion. Ahem, if you are from an open space and look, I have some. notes. Hmm, but in anyway, uh, Here's how it works. I think i'll move on to the merger a little later, because I think that this really interesting topic. Uh, it's a bit difficult but I think that this combination of chat GBT codeex and GPG 5.6 is for me gold standard for working with knowledge. This is the place where I spend all my time, except those cases when I I use Fable, to which we will get to later. I I use it. for coding, I I use it. for writing, um, I I use it. for many different things. And I think that this , uh, combination power, productivity, amenities use and speed, what does its great for joint work. Fable— this is something else. It's like tactical nuclear weapon that used for big tasks. Uh, I use...I actually I use this. 5.6— this is incredible subagent for Fable, about which we also Let's talk. Uh, but I just think that OpenAI really worked over this model, and they have this interesting focus, instead of create the maximum a smart model that so smart that it's really hard use. Let's create something fast, really productive and convenient for any work that you can to want to do So, this is mine, uh, mine current feeling with about this. Let's move on to Ariel. Could you please tell us a little bit about what you are for you use it, What are you using it for? Do you use? What are you generally love in models, and then what Do you like which one? What do you think? What are you Do you use it? This is yours. daily incentive? For why are you doing it Do you use? Yes, completely. So, mine role in the operating room Everyone's activities means that I benevolent lord when it comes down to such things as company expenses. I I watch a lot. reports, I think about problems with staff, I think about tools, and what for me really important in work, this accuracy and, in in principle, very good multitasking. So, I, Dan, was the first. who influenced my life personally. You know, his love for CodeEx absolutely contagious. CodeEx really became my main work surface, and I noticed such a big the difference from 5.5 to 5.6, six, in all these tactical tests that I am looking. This forced me to do a lot more, faster, and I initially had more confidence in its accuracy. On Sorry, I'm a bit confused. skillful when it comes to things reaches mathematics, especially to the mental mathematics. This is a family. This is genetic thing. Yes. It's a genetic thing. We tried very hard. Hmm, but AI became for me huge discovery, and I thought: "I am now "I can do it." This is our brain prosthesis. for mathematics. Hmm, and that's why, hmm, I really needed a model that would accurately performed calculations, as well as explained how they are being performed. For example, I spent almost benchmark at 5.5, 5.6, 6 and Fable, where do I it was necessary collect 46 different CSV- files. It was something. like huge dump data. Hmm, me too it was necessary to collect them into one extensive analysis and to create from it model that actually would have some different nuances. Hmm, and hmm, I first tried do this with 5.5, and this was huge failure. I received something that was absolutely unsuitable for use, and, um, someone sent me by email 46 CSV files. And, um, I just started with that, what Codex said: "I have there is an email from this person with 46 CSV- files, find it and bring it." And he replied: "I can't to do this." On In fact, he just doesn't could. He asked me, Who is this letter from? although I already said it, and then asked me indicate the subject of the letter. Hmm, when I tried this is from 5.6.6, it really is found the right one email, reviewed all data I think about the breadth and quality. Hmm, and then gave me a few really smart questions about how I I want this job. was completed. Hmm, and all that I it was necessary to do, this answer once, and then he created something that was real useful and passed sniff test for accuracy and precision. Hmm, I also tried that. exactly from Fable. And Fable stuck, as if he said that the connector is not works. He wanted to I downloaded 46 CSV- files to disk. Hmm, and it's kind of interesting that this is a little more got stuck in the weeds. Hmm, it is proposed summary tab, um, which I found useful and interesting. Hmm, but overall, I think, um, 5.6 six helped I am very fast. to achieve something very useful from minimal intervention and in large quantities sane. And this, I think, is really good expression of some important things in 5.6, which are different from Fable, um, and especially 5.6 in the ChatBT Codeex application, which is a combination of Codeex and ChatBT. If you just joined, they all combined, um, they made a combination. Ahem , so one thing that, I think, Ariel, you you point out, this is what it is this way of working, which became very common for programmers for last year with artificial intelligence, and namely: "I don't do work alone; I am building a system that performs work", um, if you recently heard about cycles. That's what it is. cycle. It's like you install rails or a track, which the agent works, and agents who perform work, and you are viewing results and over time improve them. Therefore, the task you done, where are you delegated him to a long process, collecting a bunch of CSV- files, is this one cyclical working the process that begins to seep into working with knowledge. This is starting to happen. part of the work with knowledge, and I think that in time this will be all more common phenomenon. And I think that 5.6 six in the Chadbt Codex, Chadbt's work is, uh, the first time when is it really possible, if you are not technical man, and you have some acquaintances with artificial intelligence, and you are fascinated by AI, but maybe you don't do you program from the side or something like that. So I I think you are starting. understand this because it is powerful enough, quite accurate and easy enough use. It feature of Fable, right? Yes? Fable in, uh, in the same scenarios for you. Um, it seems a little gets lost in the weeds, and um, um, it's like a little lost in weeds. It seems speaks his/her own language own, small in private language, and then it is in cloud application for desktop, which I understand because of him. history and how it became what it is. It's, um, just a little bit. more difficult to understand permission structures and what it can do, and what is not, and what I do. I I believe when it says, what can't this do thing and all that. AND I think Codex is something like CHBT Codeex ch, I will have to just hit me hammer on the head, if I say Codeex. Um, uh , it's hard to do that get used to. Um, but this, um, um, that's a lot productivity in package, which, in my opinion, opinion, maybe use everyone who really perfectly. Yes. I think one of things that are important to me really important, this that I am moving quickly and I do many different workflows simultaneously. People write me something on like: "I need funds for X, Y and Z", and I also have problems arise with the entrance, and also me I'm trying to work. over some big ones, more important strategic tasks and simultaneously execute multi-threaded task. Therefore I do not I want to think about " this is chat, this is collaboration, this is code, this is whatever "? I just want to it worked, and that's it What does working in chat give? me, and I think that this advantage 5.6 compared with previous ones models. So I want, I want bypass all these issues . I want Katie joined. I want, for Mike to join. But before we do that we will do, we had very interesting and important question from Anom, who said: "What the hell is this?" like this?" Honestly, I often asked myself this is when I woke up in the morning. So, thank you for that. asked a question. This is our direct line. broadcast 5.6 Soul, uh. Soul means sun for those who don't follow by us. And that's why I wear sunglasses . Bright. Sunny. Uh-huh , we are the only one subscription that you needed to stay on the edge of AI . Can you imagine us? a bit like a laboratory advanced research future of work with knowledge. We let's take these models, we test them on writing, coding, design, marketing, PowerPoint, creation presentations, everything that can you imagine , and we use this for our real work. And when the models we are leaving we are holding for you vibration checks, as in a straight line broadcast that we we do here. We have there is also a written the version that was written Katie, who is fantastic, with breakdown from the whole our team. We have a whole team of people, who protested this and told you, for What is it good for? why is this not good for what are we we use this in our everyday life. Uh, differs from conventional tests that , uh, it's doubtful whether they apply or not. These are, uh, real vibration checks from real people, which, uh, constantly use these in pieces. Okay, here it is. This is, damn it, like this . Uh, I'm doing my best. I can say that. Katie Parrot, share their own, uh, general impressions from 5.6, Please. I love 5.6. I was one of those people, who really had difficulties in the Delta, where we are not for long lost access under time of little government bypass. Uh-huh , until that moment, when for a minute I started to feel very gloomy and as if was exemplary homeless, because wanted to return to Opus 48, because it too slow. Sonnet is just amazing, and we wrote about this in our Sonnet 5 review v five last week, so I'm just very worried, or not left for me models? But then 56 returned, and I I remembered why I I love. Uh, what? approximately corresponds because of what we are talking about still talking. It quickly. So, you know, I very iterative a person like mine main working process, that is writing, and for tips in general. Hmm, it's fast and manageable. For example, the number the time I spend to manage GPT in chat or GPT models, not because they do something is wrong, but because I think about something extra that I would like to know what affects work. Hmm, and how GPT56 is capable of this pick up and to work with this, fix, but remain focused on a bigger task. It what really bothers me impressed, this is what it is very ergonomic convenient model for quick decisions, quick execution, iterations, revisions many options. Hmm, one point that, in my opinion, it was It is interesting to note that what I used 56 for creating a draft vibration checks. for about 66 work. Um, and how about some of the you, and you and Soul together, 24 drafts. Um, yes. Wow. Yes. We made 24 чернетки. Um, This is partly due to moments about division into delegates and співпрацівників. I I realized what I want. more on this working with artificial виділили на Fable. But missed. So I made wholesale, you know, red draft review vibrations, and he passed and applied it. But then some of for example, there are large, whole changes; some of them are smaller, more purposeful changes. But in this the whole point of the model is because, uh, in any case in any case, review so cheap. You you can get, for example, from the point time, you can just quickly get a quick problem solving, if you want to see options if you are not arranges the direction, which one is taken, this very easy to fix, um, what is special to me? annoying in anthropic models. I found that they are prone to excessive fixation; they are just excessive are considering things, I I think so, that , um, makes the actual product less convenient for use. AND this is what I want to emphasize that what Dan said, namely what appears to be OpenAI puts in a lot of effort efforts to create models with which I want to spend time and work, so that the model is great. He fast, adaptive, inventive, he finds, has a good sense, both he and Codedex application is great experience work. All ecosystem now with Chat GPT and Chat GPT Codedex, all ecosystem—this just for fun; it just a nice place to do the job. As Ariel said, I not only spend most of his time in Codedex, not in desktop application Claw the Claw, I'm spending most of his even more time, than in the browser. Hmm, even more than with friends and family, isn't that right? Yes, I will. volunteer. It, certainly true for me. Um, so Katie, first than I invite Mike, I I think that people should know that you historically were heavy user Cloud. So I would like to so that you could just say this is just quick told about it themselves . Yes. So, my current, something on like a "bug-" "bear", which in I have something I'm talking about. I begin to think about and what brings me back to mine feelings, this, hmm...u I have this feeling, that we stop at how good working models noticed this in I used to give them language. Моделі Chad GPT та much more less prone to instructions, and they just didn't sound like human as антропні. Hmm, I would said that from 55 and 56 I began to notice turn. Not it is necessary that professionals are found stronger than the first once. This is what you you will see in our benchmarks, this is what Saul actually showed the worst results, when we just asked him take several tests at once. But, but, he is much better perceives the opposite connection, and I can just say no yes on the first try, and much faster make a second one passage, that it is simple, like, supports me in motion and gives impetus, Do you understand? I am not I'm fixated on something, that's so far away from the reality that I I don't know how to move. next, what am I last with? sometimes I encounter anthropic models. I agree. I am completely I agree. Okay, sir. Taylor, manager department technological consulting on Every Welcome. Well, tell us. about your rating models. Yes, I think Katie has radio. Uh, I, uh, this ridiculously. I was talking about this is the other day. Uh, uh, me always said that when I write, the main feedback, which I receive from editors, consists of because I'm trying to squeeze in too much ideas, technical ideas in one post, and, uh, me need slow down and, uh, explain these technicalities concepts. And now I I find out what Fable does This is with me. So, although I love Fable as one of the smartest models, I actually quite aggressively moved from him to Soul, and now he's mine daily car. So I still I use Fable for some tasks choice where I know that this will be enough difficult, but, uh, me just not even I think about the task, which I am going to solve. I just I say, "Hey, do this, do it". And it was several different things, who pushed me before this. Uh, me was very enthusiastic user wired communication before, and I know that you were skeptical to codecs yet, maybe two weeks about three weeks ago so, right? I hesitated. I hesitated, and, uh, and you know your contagious enthusiasm regarding CodeEx, but, uh, there were two things that, uh , conquered a little me before this model, uh, when I tested CodeEx. One of them was a browser in application. So possibility download document Google or something else and work on this document together–this, well, extremely useful. Um, and I guess... they are potentially change it in this one too topic. But, well, but, well, this thing with router, where one stream can to send message to another. This seems like it. obvious, simple function, but it orchestration problem . I wasted a lot of time. to find out, how best to do software factories provision and orchestration, and, like, I even passed. two-day intensive course in Gastown. And, and, I, like, experimented with all these different things, and now I not at all anymore I think about factories. software provision or orchestration. Me just like it this is, simply put: " Hey Codex, update. this topic". Hey, Codex, update this topic. Set a goal for yourself this and tell me, when it's ready, you know, and it's just like that as if it seems natural. So I I think it was two, Well, bricks in the wall. Uh, and then, and then, when we had Soul Access, I thought, “Oh, OK. This is enough. steeply". And uh, and this too already the level of the past Opus. I still think that There's something special about Fable, and, you know, it will be very interesting to see. THERE ARE- uh, it's something in between Fable and Opus, but, uh, but, but, uh, I would used Fable less, much less . I would say that now Fable—that's about 20% my use, whereas, uh, you know, when we tested this previously, it was more than 80% and 20% of codecs. Yes. Yes. I had the same thing, and I have the same thing with this, as if Fable were saying in a language I don't know I understand, uh, you know, old CodeEx models did it, like 5.22, for example, it was very it looks like he just saying something, I don't understand. Hmm, and that's what I'm working on. this is creating a standard for uh...Um, and, uh, and, and yes, that means that although he really is use it for big things, where you think: me just need something, erase this task, which is not always is happening. Soul is not just better much better designer. Hmm, me too designer you know what I mean. because I think Soul much worse. Uh, I even think that Fable. Give us a call a few examples, and I . Yes. So let me I just to share yours screen. Uh, okay. під назвою eval packs. And I засновника Pi. He I don't understand what artificial intelligence зробив для мене. So I пояснює мені це. And this great, isn't it? easier to understand. It just a template, but log. We don't see this. Wait. Now we see it. Now we see it. Start showing. Yes. Yes. So, um, well, when I am trying understand the concept, which I don't really I understand, uh, he's essentially creates for me HTML page, and there is different tabs. There will be written "to" and " after "uh" with blocks code, usually yes? Well , um, and, uh, you know, this as if to tell me that, what, uh, what, what changed, uh, and what evidence that, uh, we changed this thing, right? So, it's called eval pack, and that's just me I ask for it now. do eval pack at the end something complicated. That's great, Mike. Me this is necessary. And this...You sent by one of them. I thought, "This is actually very good ". So this is for me I like it. Yes. Yes. I actually revealed the source code this skill. I...I...I I will share it. But yes, but, uh, this looks great, truth? This is very beautiful. for the eyes. Light understand. This is what I had to...that's what mine the soul did it, truth? This is what in I was there before. It what Claude did. AND it was something on like, you know, so-so lousy style, like Tailwind. What is Claude? What is Clad? It was Opus, but, um , I haven't tested much Fable on the subject designer things, but, um, yes, it was looks like what I have I took it, um, because I much more I like the design with Soul. I asked him. redo this, and here it is How does it look now? Yes. So I would say that they definitely updated design from 5.5 to 56. And one of the things you notice, just talking to team about this, um, the thing they tried to improve . And what will you notice about how they did it thinking: when you to do, he decides, what concept is this cause? Do I want to? to cause something to happen like a forest, you know, gardener forest theme, or beach, beach theme , що завгодно? And then he is trying this to embody. What do I think? quite consistent in this model, so this is what that she likes warm paper tones. And you you can see the warm paper tone on in the background. So these things, but it's a little it is foreseen that, on less successful. One of things you are talking about, Mike, you talk a lot. about cloud models, this what they have high dispersion. Hmm, results for creative things, such as design, I I think sometimes they are better. AND I'll give you an example. in this stream, this is what that I believe that the 56th the release was the first available for knowledge workers, . Е-е, щоб більше a system that performs work, not on itself work. And here I am asked GPT 5.5 to do scheme, and that's what in The result was: you know, she at least readable , and uses GBT image model, yes, at least she is readable, and then I asked Fable to do the same, and this is Fable, and this is the same query, the same model image, this is also GBT image model, but you can just another level. Хм, і, е- uh, and yes, I think that OpenX they improve, but there is still a lot of work to do need to be done over design. Yes. So maybe I would, uh, reworked it to ...I like it CodeEx design, you know, I I mean, I simply I like warm ones. paper overtones. Yes. I like it when the paper is warm I'm scratching an itch. This is funny. I asked. similar for reports and others things, for example, site, and it looks almost identical to, used your skills. looks like slop. So, Yes. Можливо, мені це just like it me too I like it. you know, it's something Yes. Yes. Це новий ШІ. Yes. Хм, у мене є ще ШІ. So I'll show you. this. Е, знаєте, я не PowerPoint presentation. THERE ARE, I don't know if you already you can see it. THERE ARE, you can see it. see. Yes. Steeply. Отже, ось що, е- е, ось що зробив Soul. Ahem , for those who don't know Yes. So, this one simple. Simply just, uh, convert this to PowerPoint presentation, to present for example, like this, uh, this не схожа на це. Uh, від сценарію. Uh, done. She looks good. She looks great. She looks better. ніж те, що у нас є. AND that's what singularity. Therefore, hmm, when I first saw one of AI models, er, with a different design, usually they just are trying and are trying implement our design. Hmm, and, uh, this was Opus 48. Uh, I think that this was the first time, when she just ignored me and did her thing. And this the first time when OpenAI model ignored me and did its thing, and this better. Uh, and, uh, intervals are enough good. Uh, it's like it covers all key topics, looks professional, I think. And she uses the same brand colors and different, but she developed this new background, and then she developed all these custom SVG elements, and this looks pretty branded for us. Yes. This looks very good. Yes. So, I think that it was interesting for me. Uh, and, uh, maybe it's just perfectly use image model for the design of things, but, uh, yes, I want to research this a bit further. Yes. So, uh, if you noticed that I taking pictures, uh, for now you were talking, uh, this because I have talk to our Mom, uh, now. We we usually call Mom on Thursdays, uh , just to check, but, live broadcast. So, then I just took a picture of myself and Ariel. Аріель— my sister. Yes. Yes. So I just thought: "I'm on the stream, Mom." THERE ARE- Well, okay. If you are here, What you should know This is first of all, drink. water. Е-е, беріть, е-е, Everyone, take yours. water bottles. Hooray. THERE ARE- your agent. In these develops like this subscription that you needed to to stay on advanced AI. You can imagine us a little like advanced laboratory for working with knowledge. Uh, we we play with everyone new models. We, uh -eh, we tell you, What are they good for, and inspections. We have live broadcast. IN daily. Uh, where? is there a website? Right here. You can see what we have here Hmm, we have a big subscribers, which, uh, coverage. Е-е, тест what are we we use this модель щодня? For яке я розповім далі. IN є купа інших речей. AND things for you. Hmm, every time we we create, we we create ideas, applications and conduct teaching. What when it comes to ideas, we we are hosting a vibe- checks like this. Hmm, we even talk about things like maximization efficiency or token maximization , uh, and we give you practical ideas about how to, uh, fight with such things. We have there are also a bunch of programs, which we create on their own. IN We are Spiral–agent for writing texts. IN We are Kora–agent for working with electronic by mail. We have Sparkle– program for organizing files . We have Monologue– program for transformation speech to text. IN We have a bunch of other programs. We also conduct teaching. Apart from this live broadcast, we we hold camps and courses for subscribers. And also we cooperate with large companies, to help them to implement artificial intelligence. If you haven't yet subscribed, then on the limits of your possibilities worth signing up for. If you are subscribed, then you are extremely cool, me too I like that you here. Okay, so now I want some. to talk about coding. I see question from Harold 33. Harold, I think the 56th will come out. in CodeEx during the day. I I also think that he will be included with program update and GBT's work in chat. This is called chatbt work and chatbt codeex. So follow updates. I just , uh, someone from openi told me that this is called chatbt work, and not chatty codex. So just remember this . Uh, and if you see this or not, leave it comment, we Let's talk about it. Uh-huh to talk about encoding by with this help models. And before I I will do it, I would like to remove these sunglasses fed up. Uh, okay. модель кодування. Uh, in my experience. Uh, no. need quality рівня Fable. This is also субагент для Fable. Therefore , if you, um, trying to force Good job. is called engineer. He asks вібейбовим кодом. Um, how good is he це якось смішно. He , he got 56 out of 100. Fable got 90 out of 100. Um, and, uh, this is a big, big difference, I would say. I I don't think there's a difference, and I just see how people say that they just received in-app updates Codex. Um, if you, um, looking for updates, just click " update" or " update", and you it you will receive. Hmm, how am I? already said, 56 out of 100 for 56. Um, 90, 91 out of 100 for Fable. Fable just crushed their. I think this score actually not representative Ahem, I have to say this because this is the score he got got it, but, hmm, if I am going to choose. number, leaving only from impressions, it will be something like 70 or 80. Um, I think we, probably need view this benchmark for this a specific thing. The reason why he received the lowest score, 56, is that he did not delete or simplified enough code. He rewrote code base, but left a bunch of code, hmm, which was still standing, and which probably should would be to delete. And the code , the new code that he wrote, was, hmm, much larger and less conceptually clear and simple, ніж Fable. Uh, yes, that's it. one of the differences, which you will find. I I think that a lot people who use this in produced for creation software software, discovered that it a little перезаписується. I I also think that it is pages, for example, while writing a web -site, but not in others reasons. Uh, this is it. atmosphere for coding. I I use this as subagent for Fable and random project. Uh, one of the things that I understand, this is what I недостатньо їм. Someone порядку?» And I дуже погано». Uh, no wonder. Ден забуває їсти з дитинства. Ahem, Треба пити воду. Uh, oops. Oh, just pointing it out це на екран. Uh, yes. well, so one of the things, better. So, uh, huh I can create really simple you want to call it. I What. Е-е, поки я до нього не звикну. Uh, computer. He працює постійно. So create a small addition. Бачите, у нас Of course, for you need a warm one paper, of course. Uh, and basically that he does, it's a monologue, which is our app for conversion speech to text. It just let me to make voices notes, and then in I have an app for photos, and every time, when my computer open, Codex is simple checks the monologue and my app for photos, and if he finds a photo of food or a note about food, he simply creates food journal. He is looking for macros, and then simply reports To me, that's what it is. AND one of the things that in this is so cool, this what, um, an app for never take photos had: "I ate marmalade worms of the past night." Application for photos, hmm, I got carried away... by the last one Sometimes I started admire sweets. I don't know. Why. Hmm, I never was a fan sweets. I was when I was little. When I was little, I was a fan. sweets. We have here fact checking. But the last, probably, 10 or so actually wasn't. IN anyway, I was ...So what really interesting, that's what application for Therefore you cannot create applications in your photos Apple. But now you you can, because you have The computer has codecs. It's very easy. створювати речі. AND photos and just, food diary. And then I'm going for example, I have трекер сну. So he, sleep, and then I one of those глюкометрів. So, in anyway, I I'm going to do everything. connect. This will be knowing how program. Um, so, I recommend, uh, CodeEx things. I also have let's see if he тут є, чи ні. I have Marketplace, er, searching захочу купити. And he Facebook Marketplace. So, here it is. was. І він працює в Do you want me to do this? did? Then just doing photo, he finds in his photo app the thing I want sell. He identifies her online from my email and from image search, writes a description, and then communicates with people, and they come take her away. So this is just cool for that kind of thing. AND so I think that's it in a certain way initiates this new way thinking about work with knowledge. Um, and this, and the main thing is that it really powerful coding tool with which quite quickly , easy to communicate and work, so you don't need to be specialist to to know how to do it use. Well, in Does anyone have any? feedback on this? Me just constantly things that amaze come to your mind when you use models because it simply fascinating a look at how your mind is working. What is it, uh, tell me Tell me what? , tell me so that I could have done better describe? I I mean, this is the same, the same the root principle that , in my opinion, is that as you just started to do artificial intellect, I thought that it makes so much sense for Dan. It's simple. as if you have this bottomless need understand yourself and your strength, know everything about yourself, and therefore you constantly looking new ways to explore the depths The Absolute. Finally, I found something that just happy exploration of the depths, like me, and these are codecs. Dan, Dan, how are you? do you think...uh...I recently thought about this is, uh, about JSpace, where you can see that the model thinks, and, uh, that while she formulates its respond. And, uh, me too saw that there was one thing where they said, uh, when do you do typographical error, quite often model thinks: "like an idiot." Really? Yes. I am not I know, have you seen it? Yes. And, uh, now I I think that Codex thinks of us when we we do these random things, you know? I think the Codex is like to this guy need help. It for the fifth time now we say it. My CEX is very successively sarcastic about concepts in your thoughts . Oh yes, you gave it to me. sent. Very ridiculously. Yes. That's right. And whether Do you know why? Is there it was written explanation why? So I asked, and it was essentially said that it it was similar to adaptation to my style and models thinking. I don't know, when to change situation. Yes. I don't know. , or is it flattering me, or stunned. Hmm, but yes, I like it. Hmm, uh, Katie, for you. Reactions. Reactions to... Well, first of all, I can confirm that now I have a job. Chatbt app. I clicked "update", and ...the Codeex application disappeared, although it is possible was to leave the icon Codex. So this is what you're talking about. you know, Let's talk. Let's talk about merge per second . Therefore, they united applications. There is something real here. interesting. Chatbt is out, It seems that in 2022, or around 2021 year, 2022, and he defined the era of the first AI waves. When you thinking about communication with AI, you first think about Chatd. Hmm, for the last one year of AI, especially in coding, was able to become much more agent, er, colleague where you can delegate work, and she actually performs it, and it's not easy step-by-step chat. AND Anthropic with Claude Coad was one of the first people, uh, one of the first companies that reacted to this seriously and did this is successful. And I think , which is historically, uh, what it was really interesting, is that if you company and created something as successful as CHBT, it was very difficult move to new paradigm, new era, and, uh, what we see in AI, this is what big companies actually are able to do it. Uh, and in particular Codedex was so-so separated product and model line. This is a team product developers , model, several researchers who work independently, just cooking something brand new, so far CHP team was engaged supporting this app with 800 millions of active users. Hmm, and this It worked. For example, Codedex became such that we, the greater part of the team, switched from Cloud Code to Codeex, because it really is good. But then this creates this crazy problem, when you think, "Well, now we have two things : we have ChatGBT, then in we have Codedex, and how is it explain to people? And there is three, um, actually, big, hmm, wait. Well, so we have guests. Um, so there are three big ones. user groups, for which you need to contact one and by the same application that is, um, ordinary consumers, tablet developers AI, and then, for example , business users , all in one app . And this is really very It's hard to fit all of these in. things together. Therefore I I think they are are trying to do with this association, yes is to squeeze them into one application, but save him in such a way that regardless of whether who are you, you were not upset and could use it as you would expect. I I think this is mine. kind of unresolved questions in AI, namely what we found out how to disturb oneself as a company. Hmm, but as soon as we we don't know how product. And I think that this is their attempt. I I think Enthropic has that the problem itself. Um, and remains to be seen which are the best practices. And, uh, to two of us joined very special people from OpenAI. House and Roman. Welcome. Hey, how are you? Greetings . Good. How are you doing? Uh, Please, working at OpenAI. Yes. Hi, I'm Dom. I Sweet. lead developer . Uh, but yes, Tell us how йдуть справи. So, this is there was a slightly different launch models because it ще до її виходу. Uh, today she came out, package. Я казав про saying that this робота з чіпами. Therefore, як це називати. Uh, So tell us. a little about it. Yes. No, this is extremely the new Frontier model, доступна всім. We hard worked to for everyone. So this is use. AND during the last several weeks many people for development software for everything that вже багато тижнів. We company team daily from finance to recruiting and sales. All tried use this product to create with it amazing things. AND intellect for coding. We can talk more with knowledge. AND to make sure that For example, if you not really backstage. And that's why we agent. This gives much more opportunities. You can you connect to all my friends tools, services that use every day. AND now you have, how are you seen in a straight line broadcasts, possibility to perform any work on your own computer with Chad GPT. Surprisingly. And if I Codeex user and I I'm trying to think, How am I now? use this addition? How about me? this is to think and what differences? When I need to switch to the "Work" tab, and when to the Codeex tab? Yes. Yes. This is great. question. Honestly , for users and Codex developers this the same codec that you know and love, isn't that right? We just let's make it better. Now he has separate, allocated space in the Chat app GPT. But if the codec you suitable, you can just stay there for all your work. And in fact, we, as engineers, have must be performed coding tasks , isn't it? For example, documents, we have to to explain one's have to follow various sources. So працює в Codedex. I think research, for example, regarding functions that I must create", but then I have to copy , що дивно, чи не так? AND deep I am ready for conversation. So, this interesting, because now all these tasks and yourself. I also think What not worth too much to worry about forged. You can task, simply to switch quickly interface switch between them. But you don't informational vice versa. I used, informational technical work this week over codecs. Exactly the same. Shameless advertising. IN we have a guide to codecs for informational technical work, which one, if you're curious use codecs for tasks, not related to coding, you can review. He quite good, if I can do that say. Yes. And this will work. without connection to networks, and this will work natural, like everything that By the way, you wrote, that's fantastic uh this will work natural for Chipity users, because they are all the same connectors, the same plugins and everything else will be there. And by the way, we released weekly in Thursday during the last 10 weeks or about that. So today is not is an exception even for developers who use CodeEx. IN new features, of course, such as Soul Ultra to get for yours the most ambitious tasks. But we have there are also new things, code editing, directly in CodeEx, So, if you're interested, we are just getting started to work for developers. I like it. One of the things that we were talking about, I think , there is such a very, hmm... very relevant now hmm...instead of to do work on your own, you performs it; and this, this агент, більш-менш. AND one of the things that I using this about that, is workflow available for you can actually delegate tasks to force him about this. Хм, схоже, що There is no work. many. Змусити цикл much more leverage. with your electronic here and there, but by email for me. Ем, але мені internally? Чи у вас є examples of how this changed your way work? Because I I really think that's it. one of the things that we talked, and I It's a shame to say this, but I think Fable generally better programmer, but he less convenient in use. This is less convenient model. And I I think it's version 56. still very powerful, but she is much more convenient for use to someone who is not technical specialist, could use it for delegated work in a brand new way. So me I wonder which one you have. experience. Yes, I think so. the most interesting thing about the 56th version is that she, naturally, more, I would called it independent. Especially if you have computer, and Chrome extension just doing his job work and does what necessary, right? So, if you use computer, extension Chrome, even for knowledge work, there is many things where she can simply execute everything from the beginning to end and check your work. AND, maybe you think that it may take a little more time for task completion, but that's because before as long as you come back, it's already will be fulfilled. Um, I I noticed it myself, when he performed five or six tasks in parallel, because, you know, it was just doing work on such equals what I thought:" I don't at all. need this not even necessary to think about it, especially if you, for example, not encountered a pill artificial intelligence, як усі ми. You don't need to think about Yes? Це просто щось на kind of a model that naturally checks his work and makes sure that everything Maybe it's not example of an employee knowledge as such, but I I will show you. demonstration, Please. Give me look where it is мої вікна. Excuse me, I we also updated documentation. We moved CodeEx documentation with новий learn.gptt.com. And we приємних моментів. AND we did this here, many of these are not screenshots. We took a screenshot of CodeEx, and he just tied its with the codebase and rebuilt it all in real HTML. Wow. And for me it is диким минулої ночі. I under the name " visualization "in, um, an application that works regardless of whether Do you work in chat? on CodeEx or even in Internet. You can to ask him/her visualize concepts for you, and we allowed me show it on the blog; actually, we have a few examples of how it looks. This is really cool; He gives you such interactive demonstrations where you you can play with different things to visualize ideas. I gave Soul a draft of this blog post, and I I will show you the task. in a second. But actually he was able to create everything for me interactive elements from the video. He actually checked video using Chrome extension, found out everything took screenshots different frames, and then figured out how actually work interactions that we we are trying here to show. And that's why he was able to completely rebuild this, checking the footage video. It's easy for me. amazed. And, obviously, I even didn't see this. It perfectly. Але ми б не Yes? Тому що це було б too much effort or too much many turns. But that the model This is so cool. look at this and You think, "I don't know, used visualization". I I think one thing, understand, that's what when models become able to work becomes a bottleneck understand what it is just done? Or I am a real mentalist. is happening? So ability what if you think a huge breakthrough. Absolutely. And that's why I Yes? We are very important. You want to create through research for people. That's why we a lot of time on has an open source code, using everything to work during to check reliably importantly. And by the way, rule, decided on at this stage. Lot rather, and the fact that you can run agents for delegation of tasks, which can be quite vague, not even quite accurate, but the model understands the intention, starts to move on your own programs or Chrome tabs by with their own help cursors. You can continue working and check when it will be done. This is really great. see. I like it. I use this. constantly. I often I attend meetings, and my computer something does it by itself, and I'm like, "Oh, \[ \_\_ \]." Codex- this is Codex working. Hmm, So guys, I know, What have we run out of? time, and that you have very tight schedule. Hmm, Thank you very much for that. came. Something Finally, you want tell us before Will you go? I think there is one more. a small function that I highly recommend to familiarize yourself with: if in you are installed Chrome extension latest version, in you now have a chat on website. So you can open chat directly in Chrome. And this is incredible, because that he actually connects to your GPT program for chat, which means that in you have access to local file systems, and this allows you to do things like website browsing or Google Doc, and it can directly link to files, which are in your file system, and interact with them. Hmm, and it's simple incredible for such things, such as, for example, working with knowledge. Absolutely. And mine the farewell speech was would be like this: "You know, these models have already reached of this level opportunities that don't exist no impossible ideas. As well as give him yours , all the ideas you postponed, which ones are you never tried before, the most difficult mistakes , and, uh, we can't to wait for see what happens. Uh, because it really is wonderful model. She is mine. I like it. Thank you, who joined, guys. Thank you for Of course. I appreciate it. Thank you very much for that. invited us. Have a nice day. To meetings. Good. Team thoughts, reactions, comments. Feels like виклик, знаєте. Me need to find something, that constantly him breaks. What are you going to do? do? I mean therefore everything is possible, Mike. Yes. Grow up. Grow every 10 times. Um, I started something. an ambitious project, or on my scale, I would said. Гм, тобто, е-е, we, uh, we put set goals for yourself on the third quarter as a company, and, uh, one of the things, what I want to do is to collect some, for example, goals for tracking, uh, what not easy mathematically justified, uh, automatically extracted from different data sources that we we have all over companies, uh, but also make them digestible and actually telling a story for people who take participation in our weekly meetings . Uh, you know, there's a little bit art in order to: "Okay, we're not just create a table and we show it to people; there are- Well, we say, "Okay, these are the changes we contributed a week for a week", or "This is how we are moving towards this "goals", or, uh, "Here what does it mean to achieve it ". Uh, and, uh, I I'm actually working on it. this with Austin, and, uh, Austin is our leader. development department. Yes. And one of the things, what we do is that that we had a meeting on this occasion, and I asked Codedex simply extract transcript this. And, in principle, how can we create something like a package with a discount, for example, What is the plan of attack here? Then Codedex sent and Codedex from Austin reviewed it and sent me some editing, etc. But the ultimate goal is in that we we cooperate with our two CodeEx. We we are going to create such an HTML site with a bunch of various built-in automations that collect data from various sources. I like it. And that's what I would literally never not created before. And I would also not talked to by someone else in Slack. This is a whole new world. I talked to Kate, editor, and she created yesterday own app for tracking where everything is in editorial process work. And I just I thought that was it. Kate, we are working. together for three years or so something like that. And, uh, Kate, uh, Kate is interesting mine, because if Kate I know that this is not It's just amazing how people with artificial intelligent like me. Uh, will do, but only really interesting. Katie, ти хочеш додати? Uh, just one more time talked about everything functions and всередині Codedex. I am very local this type browser extension, that would be extraordinary massive, um, simple in order to to return everything to its original state , like I, like, loved, like me, me historically liked work, but, like, with this advantage agent, that really steeply. This is really interesting. I'm interested in this because, obviously i'm big professional in use built-in browser for such things. Um, and if you have built-in browser, and they just added that I like it. they added an importer for all your files cookies, so that you can get all login details and other things so that you don't had to deal with with all these keys access, etc. Hmm, availability internal browser means that you can, for example, if you are in Google Doc, take the whole context from your computer and put it there. Historically, it has been difficult to force agents use Google Docs. But I I think it's starting. change, because now they good enough to use you, they can use Google Docs. But will win. Це Chrome з a small plugin браузер агента? Hmm, I I personally give preference for built-in by default. it's something like whether you want at the very beginning? Hmm, that is, if you want to stay informed what you work with, you actively use AI as sparring partner, or wrong? So, for you it makes sense to have a browser in the app, and all this in одному й тому ж. Uh, and you do a lot rapid iterations. Uh, what am I really I'm trying to do вершині циклу. So, is it can he do this without me? Um, and I think so too people, er, chrome browser makes sense because simply Chrome. I don't care, I don't need it. to see what you are doing. you are good cope without me. depending on the task. for example, just need to for such things. That's right. Тоді як, як I work at Chatbt. What do I need? I like this, so this is what, um, even for, even for, working in a loop, I I think that, uh, if I I turn it on, and it uses Chrome or, in my case, DIA, uh- Well, while I'm at it I use it, then we Let's quarrel. But if he does it in in your browser application, then it's like sandbox for this one stream. So I don't need it, by the way. actually, no need worry about it. AND I have it now. all this is happening thing, where yes, often it runs in the background mode during using computer and not switches between applications. But sometimes it's something like: "Oh, I can't do this in in the background, so I'm just going to for example, Alt, Tab, go to yours, type, before this other application and, type, to start doing something stupid ". And I am like that, I in the middle of work here. That's so rude of you. side. Hmm, yes. occurs less frequently. But yes, I think it is makes sense. Yes, it's really good. remark. Me It seems that, hmm, maybe this a factor that we haven't yet he should be like this own computer in Chrome, and it doesn't need with him but yes, and then you can have just a window, and then you will receive the best of both worlds. Yes. Okay, if you here, then soon to more will join us one special guest from OpenAI. We already used this model for general vibration checks. We used its for coding, for writing texts, for design. We talked a lot about working with knowledge. to share yours thoughts. By the way, to things, by the way, by the way, considered that, on model? Я думаю, що те, because I don't think so we must but I think that this applies to all simply ingenuity and to take the right ones use. Hmm, in writing. I in some cases I like it. I like mine. yes, every time we we work according to the system semesters, hmm, in anyway, yes, just you know, I I think that's it for me. what felt like such a drastic change from 55, it's not only faster, but also closer to the correct one faster. Yeah, Dan, I don't know, Did you have a chance? to play with ultra mode. Hmm, this is on my list for check, but a few people were asked me like: "What Is this ultra? This is new model?" And, as far as I understand, it's more like something like extremely high efforts, but I don't know, is that accurate corresponds to the function Claude's ultra code work processes, dynamic workers processes. These are not dynamic workflows, this simply higher efforts. Um, and one of the metaphors, which I recently heard , which I really like I liked it, what is it? the difference between higher model level and higher level thinking? And this the difference between that to know more, and therefore, to try harder. Um, and you can very- try very hard, but if you don't know enough, you seem to will peak in some moment. Um, and if you know a lot , you are not very try as if you you won't be like that good as they would be, if there were many tried. Um, that's why I I think it's true. interesting. Yes, I am. actually used Ultra. Um, Ariel I can to say that I woke up in Slack from her question: "You used all our OpenAI tokens of the past ночі?» And not only everyone all OpenAI tokens for the entire organization. So all our limits worked, and I thought: "Oh." Um, I like that one. planner The food is out. wonderful. Yes. Um, the people at OpenAI, who I really thought that testing were free of charge. Um, to turn back a little perfectly. If not, it is That's fine too. Um, in special guest, um , Kyle Cobber of OpenAI. Kyle, welcome. . How are you doing? Good. Um, pleasantly to get acquainted, guys. I won't start with calculations to thanks...em, financial specialist. Tell us about it. your thoughts on this models. Yes. I would said that I am very operating system for working with knowledge , і це 100%правда. I still not an engineer programmer. Now, let's say, since February I I work as an engineer- programmer, simply integrating this into your ordinary workers processes, and, you know, it a very exciting day with Soul and combination codecs and checks. I feel that now everyone will probably feel it this workflow in codec styles, where, if you just not CodeEx as an employee intellectual work, and I'm joking I'll say, don't "slouch" employee intellectual work, you only need a few months to do almost everything. But is it really this is what we are we are engaged several months. Codex— все для мене. Or can you tell me use it because it integrated into all it is connected to Microsoft, er, Excel, PowerPoint, can collect context. So every morning my chief of staff which I really like directly to that, where do I need to go? powerfully. It's simple. What's going on? AND we use, it happens like this a lot of things. AND possibility , you know, drafts, operating system use, you that he studies your style. Um, you know, this simply huge increase productivity. So, tell me about it . Наприклад, назвіть which one are you talking about, For example, I want something like crazy use. Yes, I would say, crazy. AND, see. Yes, let's see, screen. If I can share tab, it could be" Wait." Maybe I just will have to provide he has access to settings. Wait. But in in principle, um, later, let's see if it is will work. So, while you're at it you are setting up, some people ask, I don't I see 56 in the web browser. or in desktop program. Um, so, renewal unfolds. If you haven't seen him yet, try restart your program. Try it update the program. AND if you don't See, just give it. He's an hour or two old. You You have to unfold. Most people in chat already received its. Um, and if on there's someone from OpenAI on the call, who wants to tell we have something else to say, when we, when people, who have not seen yet this, will get this, let me know and I ...I...I'm broadcasting this. Ahem, Kyle, how are you doing? settings? Good. You see it on screen? I think I am shared. Uh, yes, there is. Perfectly. Good. I can't to share nothing in live broadcast. Therefore, it's something like Google Slides with GIFs animation that was created this slide for me, looking at app or website, which I created. So, in Codex makes the first even creates something in real time. But calculation extremely complex. We use them . We also we use them for research. So which is not really візуальні речі. So own. І що дійсно this is part of our calculations. And also, what virtually whole division into you are. Let's say you have data processing or interact. If not It matters. Codex these different about a few months ago, we started with automation distribution, which took five, two days accounting accounting. And then mine the team spends a few days for the audit all the details, deeper delving into using product, because we are interested in, for example, that the user does with codecs or image generation within a month, so that understand how does business with points of view strategic finances. So we, by In essence, we waste the day -two for this, and then transfer data from Data Lake or Excel to G-table , we derive them in G- table, link them with slides, and this was, in fact, old automation method . We still can use G- letter and slides for , they almost become directly to special which is built on almost 5 working days to five hours. And this is not ideally. Все ще, і це в його створення. You you can't just to make him once, right? But you can you, um, teach me this during the week then when he for you, he і кращим. You this. Він засвоює skills. You вдосконалюєте це. AND we are in in such a state that we take what we did for computational creation. We на підприємства. Ago quite exciting to work on development software provision and improve it for people like you are used. Yes, employees intellectual perfect example what you are talking about you speak. This is really interesting. Help me created? You are it created? Someone from your team created? Are you working? basically, simple about a month. So first opportunity to work in tandem with our usual the job was in March, and here we are several months. Um, and now it looks like humming when we already 95-98% of the way, and this really approaching, and simply some kind of ending quality side your monthly closing. And I would too said that it was interesting that what can you afford imagine what you can do create. So in this demonstrations you you can see how it passes, you there is a tab again with review, we have hierarchical representation about how we think about calculations. Then you can ask or see how it is affects profits and losses, and then go to more specific details computational graphics processors , for example, which graphics processors caused this cost, or you know, what is needed to, for example, Image Gen worked for Um, and then, because he has all this information and created a server part of this set data, he can actually do great first passage on a high-quality for example, yours closing slides or create. It virtual slides, built into this panel tools. We no longer we use traditional slides, or wrong? And what really this little one mascot below right. This idea ваші CEX-вихованці. But for people who, love to navigate in new, well, you know, software the security you provide with whom you can just talk, and he will, in fact, conduct you on the panel tools and will respond to question. And all this is based on skill, answers and, in fact, essentially, the way to answer and toolbars, now you just you receive pleasure from creation software provision that helps your is happening to you business. I like it. question. But, uh, software or two. What are you to create something like if you started all over again, and we could say a few things for myself, uh , from several months so, it would help you avoid some trap? Because I know what you can do this is without coding, but there is a lot to do to make it real good. Yes, this is very good. question, and I would said, uh, this is eternal for us, as if it were not done every month, we are trying make it better. I would said that the main thing the training consists of because some people are trying to do this is when they first using CodeEx or, you know, chatting for work, they are trying to do all at once, and this, I would say one of the most difficult processes that we we have every month, so we even thought try this. I would advised you to take up for the most difficult process, make it in the old way, but collect all context of all files, process, Slack channels, and then, moving forward in the process, you you say, "Okay, this is repetitive way, how we do it, for example, let's codify this as with the help of skills. contrary he is studying, you you can either create greater skill, or to create something on like a plugin for your team or improve files managing your affairs. I think that one of the lessons for employee intellectual to try to understand how it all works. like: "Oh, my "agents do it," your Codedex has an MD file each of your projects, which is something on sample of one of mine , so this is my Codedex all these different projects. server side. Try. recursively improve your skills. You can create skills, skills, specific for example, in a certain way we consider skills that most people used this So it becomes суперпотужним. But I to relate to продовження себе. You as if you are teaching a version yourself to get rid of this. And so, the investment is in because maybe it is will not be similar to maybe it will take days; it depends on how much a complicated thing. It data from different resources and various teams that work over this. So, if this It will take you a month, I will never again will have to create these slides. All data monthly magically, as if they are going together, every slide just updating. Drop-down menu for then we are actually on trying view and edit спільна робота. So it's as soon as you do it create and comes and edits comments. So this is where we release final material, common using, for example, we do not we can allow In any case, this is what I would та уроки. But a lot people just treat it like something that can because the name has CodeEx from the point of view of what you you can do. I like it. all-in-one system ". Do it on time. So, part of a large project, look, can you do will help you it's over time. Uh, and this things. Отже, ви були на once. Hmm, so this is all great. I know what you have. only one left question? Yes, Kyle. to ask a question. financial eh big problem trust this? " Right? Because in they have all this data, uh, and they notice two problems, and this applies to everyone AI models, but, uh, or one: sometimes arise hallucinations, uh, and (what this problem is significant decreased, but it still sometimes occurs, uh, that's why she mixes it up, uh, facts that are not was in actual data), uh, and the other is much more often it happens, but it seems she misses things. So, you know, when there is a lot of data, maybe , he just doesn't viewed this file or did not receive this email. Then how can you help me explain how you build trust in this software provision during its creation, which steps you take for evaluations? Yes, that's great. question, and this is probably a recurring theme that we see when I take participation in calls regarding deployment with some of our corporate customers. But, I would said that this, uh, process, right? That it concerns analyst, who just in time They will be be wrong. This is not would be perfect. I I will say that 56 Soul with with Ultra enabled—this extremely large step forward, and using Sub Agent is simply incredible, because he can to do a lot tasks simultaneously and to perform various parts of the process. But, hmm, it all comes down to periodic using, did something wrong. You need to be taught never again Don't do that. And here it is this is how it happens the process when you you think: "I don't I liked how you did it. if you are engaged creation For example, I have something like Google Audit Skills convenient when, before we will send something, we can check document and find any discrepancies between metric, which, possibly mentioned in memo, slide related to Excel or G Sheet. You, by metric, and then, by essentially, you are reporting собі, що не так. However the only way in which it is possible, it is tracking things." "watch" function where do you use thinking about how updated your skills, and that's how you are you receive confidence: "Hey, no more of this will happen again." So, hmm, it's something like disposable thinking when you are not should expect that everything will be perfect with the first time, but you should know when You are moving forward. I would say there is something like a curve growth, where from the first time it can be about 70%. You as if you notice a few mistakes. and then, approximately in a month or two, it approximately 90, that is, 95. It's like a reward. after all, there is. This is no less than work first, so that you need time, to build it and to teach. But the reward is coming , and that's how you get used to it, while now, as for me, I I let him do it. first attempts at tries incubate work with reading Slack, and I don't I'm just forcing him. issue final review, to figure it out, and he already working on material before need use it months. So I would said that the trip fascinating. But exactly question. І поверхнева great, but тренування навичок? THERE ARE -there are, project for getting better and better. use, can use in запит зі Slack. Your automation бачить це; she knows something like: "Oh, this a new fragment data, for example, ". In fact, she starts a chain I don't want you to прочитаю». She for example, she so good, how good are create. That is, how behave. That's how I am. I am gaining confidence, that everything can be simple work, and I know that it is not easy will send something to our financial channel director, and then Sarah Freer will write I have something like:" Hi Kyle, what's up? is happening? "Yes soap. Thanks, Kyle, I I know what you have. Time is up. Sincerely thank you for joining and shared, and I would I wanted you again. see. Yes. Thank you. I would be happy. See you guys. Thank you. Yes. a very special guest. We have Claire Waugh, founder of Chat PRD and host of "How I Became an AI". One of my favorites people in the field of AI. As How are you, Claire? Perfectly. How are you? Well, I'm fine. THERE ARE, You are here, I am Dan. You me you know. This is my sister. Ariel, with whom you never met. Ariel, department head operations. Um, uh, we have there is Katie Parrot, our We are Mike Taylor, department head technical consulting. Well, I радий вас вітати. Give us, give us your 56- annual inspection vibrations. I love. I mean, знаєш, ти ж це знаєш. AND I was thinking about why I love 56? And I thought, efficiency over everything else. Perhaps, even clean intelligence. I am like this: in I may be someone who will fulfill work. And my experience shows that work is being done планці. And I think that only mine use of tokens . І я подумав:" Знаєте, preview my family. "I . I am so happy. to do something. I live насиченим життям. IN I need everyone. influence. And here I am good. We have one like this. work with artificial I ask him. various things. He corrects errors. AND These are my tests. This, according to mark. And me too like Soul was definitely on top of my list, by several exceptions. I think that the difference I see between Soul and Fable, is in that, um, Fable also very capable, but, in my opinion, takes fewer creative risks, and as a person I appreciate creative risks, therefore, if I do parallel projects, similar to disposable, Fable works perfectly, and you can see effort 56 be a little more original. I saw the same thing in writing. Hmm, and then in voice, is this true interesting. You know, I I think 56 is good. writer, but, um, I I still want to talk only to Sonnet . This is very, very funny. Hmm, but Sonnet is the only one with Who do I want to contact? with your personal agent and to chat. This is really interesting. Uh-huh , you are so much said. First, we will have to to quarrel because I think that, um, creative and designer things Fable simply amazing. I I think that's it. incredible, and I I'll show you a few. examples. Okay, I'll show you. examples, and we just we say: this is for me I like it. So, well, one thing I I think this is what you They said what am I doing? I really agree, this so funny, and Katie also said this Previously, Katie was like this: "I was in such a decline, as when the 56th taken away. I think we all were like that because we have was Fable, and we had the 56th taken away. Якщо ви, якщо you are an early tester and you you just think: "Well, Should I even bother? to do something? У чому, в » Hmm, er, I certainly am згодна. But, okay, regarding 56 vs Fable with points of view creativity, design and everything like this. So, I have it feels like design of the 56th, and you can you notice that , begins what is the concept for make. Hmm, but the problem is that this папір, чудово. Hmm, warm paper, but in discussion, Yes. трохи схоже. Hmm, but I I want to show you face to face. Yes, head-to-head comparison . This will be a lot of fun. Perfectly. Хм, в мене робило? I have this. screen. Oh, you can. see my screen. Good. I was thinking about well, for example, 56— in my opinion, uh allows employees intellectual work to work in cycle, in known cycles, uh, where are you? create systems, who do the work, but don't do it independently. And I asked, hmm, I asked 56 to do his image, using GPT image model. AND here is this image, wonderful, it is readable, hmm, and it is has a little different colors etc., but it is very, very simple. Yes. And then I gave the same task, hmm, hmm, hmm, Fable, and it was his result, and this just like him just thinking about it differently, differently, in a different way, in a different way equal. And I think that in general what I I find it from Fable, hmm, and I I don't use Fable. for everything. I will say, that this is completely mine everyday tool, like 56, my daily tool. I think what Fable is thinking details of things to another. Yes. come up with a concept capable of creating actually very Yes. I mean, I you can see it here, маю правильні PRDS. I змішую, чекаючи. It This was supposed to be documentary to use something OK. Yes. Sonnet створили це. All regime. Хм, і це було good. He was quite difficult to disassemble. І він мав much more semantic color. I understood what it meant red. I understood, much more white space. So for me it was design. I would say that Yes. Hmm, here's another one good example, me actually gave both these fives. So I I think they are both are good managed. Left was, hmm, left was 56, and he really made an effort to fill in the image -fillers such as in a creative way, where, on In my opinion, Fable is great. managed. This is beautiful. frame, and he also did image- fillers. He just wasn't like that original. He was very frame-like. Yes. Hmm, and this is yours a kind of compromise that I got it, that is, 56 was the only one who ever did someone else do it? design. Took a risk to do something else, while everyone else models, and I have there's just Fable on the right, everyone got very similar things. And then I I just can't...I I can just notice. clawed, scorched, orange shining sun per mile. And therefore I also punished claw. I was like " claw-like scum ", but nice. Hmm, and felt that this was it difference, hmm, to me I like it better. AND then both of them points of view functionality. AND it's just, you know, my preferences for design. This is not necessarily means, what do you like about it everyone. I like it. What, um, Katie, go ahead. I just said that I support this. I have new personal to see how much well models can time schedule scales of divided kingdoms of Israel and Jews in the Old Testament . І, хм, це так, і Yes. I like it this team liberal humanities, об'єднайтеся. It's you. all my people. Yes. Але, як і у dark mode. He The goal for me was, person, to be able to look at this from at first glance and understand. It was thinking for the most useful. Yes, exceptionally good copes with агента до агента. So nothing to do with by us. I just will like it, better. I don't need based on how he is with me communicates. 56-та as if I had just become a powerful comrade on command. Hmm, and that's why it seems that she optimized human interface model, while Fable optimized interface model- model. Hmm, and I think that both have their own work, but I work Alone in the office, I have time. sometimes needed friend. So, hmm, I like the 56th. version. She is mine. I like it. Hmm, Claire, is there anything else? Uh-huh What do you think about the 56th? , uh, uh, Fable about differences between OpenAI and Enthropic and where are they are in this race. I think we let's finish the race in about five minutes. Yes, I mean, Look, I think that both models are very powerful, very intellectual. I companies are trying to provide use in business. Hmm, I have one gets the impression technically advanced use on even technical to the user medium length. AND cybersecurity. He directive. And how I have the impression that almost completely opposed need a little vision. Ем, не кожна is correct. Ago I had a hard time. use Fable to create . Ем, але він дуже copes well with deterministic me. Using good. Письмо—це, знаєте, добре. Good- view. Um, perfectly. Creation difficulties—this perfectly. And again But I think that if really then they start options using. So I user who the right model for users. And now one boat. Хм, дуже interesting to watch за перегонами. Claire, if people want