Talk by James Zou (Together AI / Stanford) on Einstein Arena, an environment where AI agents — and only agents — solve curated open science problems, each with a deterministic verifier, a shared forum, and a live leaderboard exposing all solutions. Within weeks it held the best known answers to eleven problems, including improving the 11-dimensional kissing number bound from 593 to 604 spheres. The same arena setup applied to GPU kernel compilation produced kernels more than twice as fast as the prior state of the art. He also presents DSGym, a data science benchmark whose execution-verified trajectories fine-tune small open source models, designed so that many tasks in popular benchmarks cannot be solved without touching the data. Topics: designing agent environments rather than workflows, collective multi-agent improvement, and executable verification. # Video Identity Observation: Einstein Arena Presentation (James Zou, Together AI) ## Explicit Self-Description - **Speaker and affiliation stated directly**: "My name is James So [Zou]... I am going to explain some of the work we're doing with Together AI and it's also in collaboration with Stanford." - **Stated project scope**: "designing and optimizing environments for AI agents to enable these agents to make new kinds of scientific discoveries." - **Central thesis claimed by presenter**: shifting "away from designing workflows and harnesses towards designing environments" — environments specify "where the agent should work" with "incentives and infrastructure... guardrails and resources" rather than step-by-step workflows. ## Observable Presentation Patterns ### Demonstrated content scope - **Two flagship projects presented in depth**: - **Einstein Arena**: an "agent native" environment where AI agents "collaborate in the wild and compete to really solve open-ended scientific problems," with discussion forums, leaderboards, and deterministic verifiers. - **DS Gym** ("data science gym"): "a unified environment... for evaluating and training data science agents to solve complex data science problems." - **Applied ML engineering results**: agents generating optimized kernels "already used in production at Together AI," with "sometimes over two x speed ups." ### Evidence attached to claims - **Kissing number problem (11 dimensions)**: agents reportedly found a construction of 604 non-overlapping spheres, improving on the 2022 human result (592) and a DeepMind result (593). Presenter states 11 problems received "best solutions that have ever been found" within weeks of the March launch. - **Benchmark critique**: existing data science benchmarks are "very vulnerable to shortcuts," with "20 to 50% of the tasks" solvable "without actually looking at any of the underlying data." - **DS Gym performance claim**: "even the frontier models, often are only still achieves like less than 50% accuracy performance." ## Intended Audience (Observable) - A **technically fluent audience** — assumes familiarity with concepts like Kaggle, Docker containers, fine-tuning, PagedAttention, error-correction codes, and model names ("GPT 5.5 or a cloud [Claude] models"). - Audience likely **research/AI-industry oriented**: talk format with slides, references to published papers, and closing applause suggest a **conference or seminar setting**. ## Claimed Channel/Positioning Signals - The talk positions the creator's organization at the **frontier of agent-environment research**, framing a progression: models → workflows/harnesses → environments, with the presenter claiming "our research shows... we're already moving towards the next stage." - Emphasis on **openness and accessibility** for agents: "any agent in the world can openly and freely participate," while it is "intentionally very hard for humans to enter." - Emphasis on **rigor**: benchmarks "carefully curated" with "human scientists and experts to review each of those tasks." ## Notes for Synthesis - No explicit channel-level self-description (e.g., "this channel covers X") is present; identity signals come from the talk's framing and collaboration attribution (Together AI, Stanford). - Results are presenter-reported claims from a talk, not independently verified within the material; the kissing-number and kernel-speedup figures are cited as the presenter's assertions with slide references. # Delivery Observation: James Zou — Einstein Arena Talk ## Tone & Overall Style - **Conversational academic register**: casual connectives and self-revisions throughout, e.g., *"All right. So so that I guess the current paradigm…"* and *"So, yeah, so so in in 1980…"* - **Emphasis through rhetorical confirmation**: frequently ends clauses with *"right?"* — this is a highly repeated pattern (dozens of instances), e.g., *"that tells the agents you know what to do, right?"*, *"This is where the agent can actually see each other's solutions. Right?"* - Frequent filler phrases: **"you know"** (*"we will compile and benchmark and test and verify… you know"*) and **"sort of"** as a hedge (*"It's sort of like the one of the first environments"*, *"sort of a unified environment"*, *"sort of like a social network"*). ## Teaching Habits - **Concrete example anchoring**: after stating an abstract thesis ("environments, not workflows"), he immediately grounds it in a specific case — the Einstein Arena — then the kissing number problem, then kernel optimization. - **Simple-to-complex exposition**: explains the kissing number problem starting from 1D (*"you can place one sphere to the left and one sphere to the right… is two"*) before moving to higher dimensions. - **Historical timeline framing**: builds the 11-dimension kissing number narrative through dated milestones — *"in the 1980s… 440 spheres… in 2022, a mathematician… 592 spheres… DeepMind the following year… 593… with Alpha Zero… 604."* - **Relevance justification**: closes each example by connecting it to application, e.g., kissing numbers → *"better coding systems including ways of like doing error correction codes."* ## Recurring Vocabulary & Phrases - **"curated" / "carefully curated"** appears repeatedly: *"a list of curated problems"*, *"we curated these problems so that…"*, *"we actually carefully curated at our own our own benchmarks."* - **"collaborate" and "compete"** as a paired framing: *"there's both a collaboration dynamics and also a competition dynamics."* - **"unified"** for DS Gym: *"a unified environment"*, *"a unified list of different data sets"*, *"a unified interface"*, *"a unified infrastructure"*, *"a unified execution layer."* - **"actually"** as a near-constant intensifier: *"they'll see actually see a list"*, *"the agents were actually able to already discover new solutions"*, *"are actually already used in production."* - **"environment" vs. "workflow"** as an explicit contrastive pair, returned to in the summary: *"we're trying to shift away from designing workflows and harnesses towards designing environments."* ## Structural Habits - **Signposting transitions**: explicit pivots like *"So the first example I want to share is…"*, *"So, I'll just give you example of one such solution…"*, *"So in the last few minutes, I want to show like a second example…"*, *"So just to summarize…"* - **Summarize-and-recap close** for each section: *"So just to summarize the this part was the data science gym. Right?"* - **Data points as evidence**: cites figures repeatedly — *"11 problems"*, *"20 to 50% of the tasks"*, *"less than 50% accuracy"*, *"over two x speed up"*, *"over a dozen different tasks."* ## Uncertainty Markers - The speaker frequently hedges exact figures: *"I think in March"*, *"in I think 19—"*, *"sometimes up to 20 to 50%"*, *"sometimes over two x."* This verbal hedging appears to be a stable pattern rather than isolated. ## Isolated Moments (Not Stable Patterns) - Quoting an agent forum exchange (*"Have you tried you know, some of these approaches?"*) is a one-time device. - Audience-directed framing like *"we're very impressed and very surprised"* occurs only once as an explicit emotional report. ## Summary The delivery is a **hedged, example-driven academic talk** with a strong signature of *"right?"*-tagged emphasis, the intensifier "actually," and a consistent pedagogical move from thesis → concrete example → quantified result → practical relevance → explicit summary. # Observation Document: Einstein Arena & DS Gym — James Zou (Together AI / Stanford collaboration) ## Speaker - **James Zou**, presenting work by **Together AI** in collaboration with **Stanford**. ## Core Thesis: Environments over Workflows - Current paradigm: designing **workflows** that tell agents *what to do* (steps, prompts, tools, instructions). - Zou's alternative: environments should specify **where** agents work — providing **incentives, infrastructure, guardrails, and resources**. - **Claim:** designing workflows "can limit the capabilities and creativity of the agents"; well-designed environments let creativity and intelligence "naturally emerge." - Framed progression: individual models/tools → agents/harnesses/workflows → **environments**. ## Example 1: Einstein Arena - Described as one of the first environments enabling AI agents to **collaborate and compete** in the wild on open-ended scientific problems. - **Design features:** - Agent-native access; intentionally **hard for humans to enter** (puzzle to prove agent-ness). - Curated problems meeting two criteria: (1) existing human research community interest, (2) a **deterministic verifier** can assess solution quality. - Includes discussion forum (social-network-like), real-time **leaderboard**, ability to view/download other agents' solutions. - Both collaboration and competition dynamics — analogized to human researchers. - **Claimed result:** launched ~March; within weeks agents found best-known solutions to **11 problems**, beating previous human solutions and specialized AI tools. ### Kissing Number Problem (11 dimensions) - Problem explained: maximum number of non-overlapping spheres around a central sphere; trivial in 1D (2) and 2D (6), open in higher dimensions; noted Newton worked on a version. - Stated progression: - ~1980: construction of **582** spheres (Zou also said "440" for the earlier best-known, with some hesitation in delivery). - 2022: mathematician improves to **592**. - Following year: DeepMind breakthrough to **593**. - Agents on Einstein Arena: **604** within a few days. - Relevance: constructions relate to better **error-correction codes** for information transfer. - **Collaboration essential:** no single agent (he names "GPT 5.5" or "cloud models" — as spoken) could solve it alone; lineage traces show agents refining each other's solutions; forum information sharing (e.g., agents discussing SDP approaches) was important. ## Example 1b: Kernel Optimization - Same environment with a different backend verifier: compile, benchmark, test kernels; real-time leaderboard feedback. - **Persona agents:** distinct roles/priors — profiling, memory consumption, precision/tensor computation. - **Claimed results:** sometimes over **2x speedups** vs. previous state-of-the-art kernels (e.g., paged attention for specific shapes, generalized to other shapes/hardware); improved kernels **already used in production at Together AI**. ## Example 2: DS Gym (Data Science Gym) - Unified environment for **evaluating and training** data-science agents; unified dataset/task interface via code execution; Docker containers run in parallel. - **Benchmark critique:** existing popular data-science benchmarks are vulnerable to **shortcuts** — agents solve 20–50% of tasks without using the underlying datasets. - **Their curation:** - Scientific analysis: tasks curated from recent papers, reviewed by human scientists/experts. - Predictive modeling: curated from recent **open Kaggle competitions** with high-quality data and evaluations. - **Scale:** over a dozen tasks, dozens of scientific domains (biology, physics, economics), many data modalities. - **Claimed findings:** - Frontier models achieve **<50% accuracy** — benchmarks not saturated. - Execution-verified trajectories used to fine-tune **small open-source models** — claimed best-in-class open-source for such tasks, runnable on laptops. ## Summary Position (as stated by Zou) > "We're already moving towards the next stage... what we really want to do is to design environments, which is a set of infrastructure and incentives that... motivates the agents to solve more and more challenging problems." - Appropriate environment design can "unlock much more creativity and collective intelligence" beyond what workflows permit. - References to published papers cited but not detailed in the talk. ## Notes on Qualifications - Some numerical uncertainty in Zou's delivery (e.g., "I think in March," the 440/582 historical figures, "GPT 5.5"), recorded as spoken. - Claims about agent performance and production use are the speaker's own; no independent verification presented within the talk. # Video Attribution Record ## Video Metadata - **Title:** Einstein Arena: Harnessing Collective Agent Intelligence for Open Science — James Zou, Together AI - **Video ID:** `01a15ca7-c472-7f5f-9a77-79a3196809b3` ## Speaker Inventory | Speaker | Role | Evidence | |---|---|---| | **James Zou** (transcribed as "James So") | **Presenter / Creator** | Self-introduces: "my name is James So. I am... going to explain some of the work we're doing with Together AI and it's also in collaboration with Stanford" | | Unidentified moderator voice | Brief opener ("All right. I think we'll go ahead and get started") — role not explicitly identified in source | | | Audience | **Audience (non-speaking)** | "[applause]" at end | - No guests, narrators, or other named individuals speak in this video. - **Note:** Transcript spelling "James So" likely corresponds to title's "James Zou"; title metadata is the authority for the name. ## Affiliations Attributed by Speaker - **Together AI** — Zou's stated affiliation; improved agent-designed kernels stated as "already used in production at Together AI." - **Stanford** — stated collaboration partner. ## Provenance of Claims and Referenced Sources ### First-party claims (speaker's own work) - **Einstein Arena** — environment for agent collaboration/competition; launched "earlier this year, I think in March." - Kissing number problem result: agents reportedly constructed a 604-sphere solution in 11 dimensions "within a few days." - **Kernel optimization** — agent-designed kernels with "sometimes over two x speed ups," in production at Together AI. - **DS Gym** — data science evaluation/training environment with "over a dozen" curated tasks. ### Third-party claims / cited external sources - **Isaac Newton** — attributed as having worked on a version of the kissing number problem. - **1980 result** — 582 spheres in 11 dimensions (attribution to specific researcher not stated in transcript). - **2022 mathematician** — breakthrough to 592 spheres (unnamed in transcript). - **DeepMind (following year)** — Alpha Zero–related advance to 593 spheres. - **Kaggle competitions** — source of curated predictive modeling tasks. - **Recently published scientific papers** — source of scientific analysis tasks, with human expert review. - **Existing benchmarks** — three unnamed "common popular data science benchmarks" critiqued for shortcut vulnerability (20–50% solvable without data). - **References slide** — speaker mentions "references for the papers that we published," but specific citations are not named in the transcript. - Models referenced generically: "GPT 5.5," "cloud models" (likely Claude, per transcription ambiguity — uncertain). ## Promotional Segments - **Kernel speed-up section** (production use at Together AI, PagedAttention examples) functions partly as a showcase of Together AI's own deployment — flagged as self-promotional in character, though presented as research results. - No third-party sponsorships, ads, or external promotional segments identified in the transcript. ## Unavailable Evidence Channels - **Curated metadata:** none supplied beyond title. - **Timestamps:** none in transcript. - **Comment availability:** not indicated. - **Slide content / reference list:** referenced by speaker but not present in transcript. All right. I think we'll go ahead and get started with the with the presentation. So my name is James So. I am uh going to explain some of the work we're doing with Together AI and it's also in collaboration with Stanford around designing and optimizing environments for AI agents to enable these agents to make new kinds of scientific discoveries. All right. So so that I guess the current paradigm of how people often are using or deploying AI agents is often involves designing workflows that sort of tells the agents you know what to do, right? Or how the agent should work. And it's typically done through a series of steps or prompts, tools, and instructions. In contrast, the way we imagine the environment is that the environment should really specify not how the agent should work, but really where the agent should work, right? And the environment then should provide a set of incentives and infrastructure for the agents and guardrails and resources so that agent can then flexibly work within that environment. Right. And our thesis here is that as agents become more and more powerful, right? If we try to design workflows that often can limit the capabilities and creativity of the agents. Whereas if we properly design the environment, this can enables a lot more creativity and capabilities and intelligence for the agents to naturally emerge. This why I think we're trying to shift away from designing workflows and harnesses towards designing environments. So what I want to do today is to give a few examples of the how we design environments for agents. And in particular also show how they're able to then with with the right environment able to actually solve some really interesting and innovative problems. So, the first example I want to share is the system that we environment that we created called the Einstein Arena. It's sort of like the one of the first environments that enables AI agents to be able to collaborate in the wild and to compete to really solve open-ended scientific problems. So, we designed this Einstein Arena to be really agent native. So, I So, that means that it's very easy for agents to just read the skills talk on our on our arena and be able to access the arena. And it's actually also designed so that it's intentionally very hard for humans to enter the arena, right? So, you actually have to solve a little puzzle to prove that you're an AI agent in order to participate in this arena. But, any agent in the world can openly and freely participate on the arena. And once the agent actually enters into the Einstein Arena, this is what they'll see, right? They'll see actually see a list of curated problems. Each of these problems is actually a problem that we curated, so it's a scientifically interesting problem. And we curated these problems so that first, there's actually an existing community of human researchers that are interested in these problems. So, these are important problems for human scientists. And second is that for each of these problems, we can actually create a well-defined and deterministic deterministic verifier to assess the quality of the solutions to each of these problems. And I'll give some examples in a couple of slides. So, So, the agents can actually decide which of these problems they're interested in once they log onto the arena, right? So, if they enter into a particular problem space, this is what they'll see, right? They'll see some description that precisely explains what is the problem. We have a discussion forum where the agents can communicate. It's almost like a social network where the agents can actually communicate and talk to each other and ask for help or give recommendations. Um and we also have a leaderboard. This is where the agent can actually see each other's solutions. Right? So in any in at any time they want, the agent can actually submit a solution to one of these problems. And because we have this verifier, we can actually then determine what is the quality of that solution and provide a score in real time. So this leaderboard is being constantly updated in real time. And the agents can also see how other agents are doing on this problem. And they can also see other agents' solutions and download those solutions. So there's both a collaboration dynamics and also a competition dynamics in this arena, right? They can collaborate and ask each other questions and help in the discussion forum. But agents are also competing with each other. And that's why I think this also sort of simulates how human researchers can compete and also collaborate to solve interesting problems. So we launched this AI instant arena environment earlier this year, I think in March. And within a few weeks, it's already actually we're very impressed and very surprised that the agents were actually able to already discover new solutions to 11 problems that are of the best solutions that have ever been found. Right? So that means that the solutions that they discovered by the agents on AI instant arena were better than any previous human solutions or any solutions that we acquired using more specialized AI tools. So I'll just give you example of one such solution or one such problem which is called the kissing number problem. So this is actually a very famous problem. It's been around for hundreds of years. So for example, Isaac Newton was already working on some version of this kissing number problem. And it's actually relatively easy to state. Right? So the kissing number problem basically asks that what is the maximum number of spheres that you can place around the central sphere so that these additional spheres do not overlap each other? So for example, in one dimensions, right? So around the central sphere I can place one sphere to the left and one sphere to the right without overlap. So the kissing number in one dimension is easy to compute. This is two. In two dimensions, it's also easy to show that you can at most place six spheres. So, that's the kissing number in two dimensions is six. But, it turns out that in higher dimensions, it actually becomes really hard to compute what's the maximum number of over non-overlapping spheres. And the kissing number problem in higher dimensions is actually open, right? It's not been It's not clear what is the optimal number. And so, scientists have been trying to work on this problem for the last several centuries. And in particular, right, so the kissing number problem in 11 dimensions has attracted a lot of interest for various reasons. So, this is actually sort of a progression of the solutions in 11 dimensions. So, in the 1980s, right, so it's best known that there you can place 440 spheres, right, in 11 dimensions without overlap. And in I think 19 uh So, yeah, so so in in 1980, there was a big advance that the first for the first time showed that you can actually just construct with 582 spheres in 11 dimensions without overlap. Uh and then that sort of stuck there for about 40 years, right, until 2022, where a mathematician is able to publish a new advance, right, a breakthrough that's able to improve that to 592 spheres. And then there's another breakthrough from DeepMind the following year that advances that to 593 spheres. But, with Alpha Zero, we know by having these agents able to collaborate actively, right, in the wild, within a few days they were actually able to construct a new solution that shows that for the first time you can create 604 spheres in 11 dimensions that do not overlap. And this is not just a problem that's of mathematical interest, because it turns out that the more of these sort of spheres you can place in higher dimensions without overlap that actually creates the better coding systems including ways of like doing error correction codes for information transfer. Right, so this actually is by creating this better constructions that also leads to this better engineering algorithms. And in this case actually the collaborations among these agents is really critical for making these advances, right? So this is a problem where not a single agent is able to solve by itself, right? Not you know, GPT 5.5 or a cloud models that can't really solve the problem by itself. So the collaboration among multiple agents is really critical. And here we're actually able to show that there's like this sort of a lineage trace of how the agents are able to collaborate and then basically take each other's solutions and refine that and further optimize it to arrive at this breakthrough. And you can also see some of these interactions and discussions on Einstein Arena, right? Where here's an example where the one agent actually was asking other agents, "Have you tried you know, some of these approaches?" Um, with uh, these STP approaches and then the other agents showed that yes, we have tried these approaches and here are some of the things that we found. Right, so the information sharing on the forums on the arena is actually really important to help the agents to arrive at this solution together. So in addition to solving these interesting scientific problems, but we've also been using platforms like the right Einstein Arena uh, to help to improve uh, you know, machine learning and AI itself. Right, so here's one example where we actually use these agents to basically help us to create better kernels for and to speed up those kernels. Right, and here we use the same environment, right? Where the agents can compete and they also can collaborate and they see these leaderboards. And we basically change the back end instead of trying to verify the solutions to this mathematics problem, here we're basically trying to you know, we will compile and benchmark and test and verify the quality and the speed of the individual kernels, right? And then we'll provide a feedback to the agents in real time in the form of these leaderboards. In these kernel settings, we also found it to be quite useful to have different agents with different personas, right? And these different personas actually corresponds to a different uh roles and priors that agents can actually have. So, for example, we have one agent that looks at tends to look at more of the profiling, another agent that tends to look at more of the memory consumptions, a third agent that looks at, you know, the precisions, the tensor computations. And these agents can and then across different personas, they can able to collaborate and a compete on the arena to speed up the kernels. And in this case, right here, the agents were also able to collaborate and lead to really quite substantial speed ups, uh including sometimes over two two x two-fold speed ups in some of these production kernels. So, here I'm just showing you a few examples where for things like page attention, uh and these are sort of for specific shapes, but we also have generalized this to many different shapes and different uh hardware types, right? Where we're actually seeing that we're getting up to sometimes over two x speed up in these kernels, and they uh compared to the previous state-of-the-art kernels for these problems. And these improved kernels created designed by the agents are actually already used in in production at Together AI. So, in the last few minutes, I want to show like a second example of a kind of environment that we created as a way to uh train and to create better data scientist agents, right? So, we call this DS Gym, which stands for data science gym, which is sort of like a unified environment that we created for both for evaluating and for training data science agents to solve complex data science problems. So, here in this DS Gym environment, we also curated and created a unified list of different data sets and tasks, right? So, these data sets can combine uh spans across many different settings. And the agents are then able to interact with these different data sets that we have through a unified uh interface and through code execution. In the DSGM environment, we also provide a unified infrastructure for the agents. So, for example, the agents can actually spin up many different Docker containers to test their data science algorithms and actually run them in parallel. So, in the process of actually creating the data sets and tasks for the data DSGM environment, so we initially actually wanted to incorporate some of the existing data science benchmarks that have been used to evaluate agents. But we actually quickly realized that many of the existing widely-used benchmarks actually have many problems. And one big problem is that they're actually very vulnerable to shortcuts. By shortcut, I mean here is that uh down here what I'm showing are three different common popular data science benchmarks. Right? And the in green here we see shows like the performance of the agents on these benchmarks. Uh but the red bar also shows how well they're able to the what fraction of the benchmark the agents can actually solve without actually using the data sets themselves. Right? So, just by reasoning or by, you know, uh doing other shortcuts without actually actually do working with the underlying data sets. And across many of these different benchmarks, right, sometimes up to 20 to 50% of the tasks can be solved without actually looking at any of the underlying data. Which I think is uh really a significant problem with many of the existing benchmarks. So, to address that, we actually carefully curated at our own our own benchmarks, right, for both for scientific analysis and also for predictive modeling. So, for scientific analysis and discovery, the way we did this is that we actually went through recently published papers and then carefully curated data and then also tasks from those papers. And then we also had human scientists and experts to review each of those tasks. And for predictive modeling, the way we did this is go through all the different Kaggle competitions to look for some of the recent Kaggle competitions that are still open and and where also you have high quality data sets and also high quality evaluations. Then we curated those into the DS Gym as a kind of task for evaluating how well models agents can actually build predictive models. So all together in the DS Gym, we actually have created over a dozen different tasks. They span across dozens of different scientific domains ranging from biology to physics to economics. It also involves many different data types and data modalities. So this actually makes it very easy for us to evaluate different models, both open and closed source models. And one thing we found is that the existing models, even the frontier models, often are only still achieves like less than 50% accuracy performance on the DS Gym tasks. Right? So these are definitely not saturated benchmarks. We can also use a DS Gym as sort of like a training factory to improve these open source models. Right? So one thing we did here is actually generate in the DS Gym actually the gym itself will actually create all these execution verified trajectories, which means that these are trajectories generated by the agents that have been verified through the through through actually executing the code from the agents. Right? So by generating these execution verified trajectories, then we are able to like fine-tune sort of small open source models that actually now achieve sort of the they're sort of the best in class open source models in terms of solving these kind of data science tasks. Right? And these models are small enough that you can actually run them locally on your laptops and your computers. So just to summarize the this part was the data science gym. Right? So we with DS Gym, we created this unified execution layer so people can actually run and all these different tasks across dozens of different tasks across many different domains. We have carefully verified that there are no shortcuts in these tasks, which has been sort of a common challenge with existing data science benchmarks. And we also enable in the DSG and a way to generate synthetic data, so you can easily use that to improve and to train your own data science agents. So, just to summarize the presentation, um I think the main takeaway here is that I think we're in seeing this interesting progression as in terms of how we build different AI systems. Right. So, then in the past, people have been building these AI systems mostly by designing individual models or individual tools. And currently, there's a lot of focus on creating designing agents or harnesses and workflows around agents. But what our research shows is that I think we're already moving towards the next stage, where you're not then trying to design workflows or specific or specific agents, what we really want to do is to design environments, which is a set of infrastructure and incentives that in that motivates the agents that you solve more and more challenging problems. And with appropriate designs these environments can actually unlock much more creativity and collective intelligence from the agents that's that's limited by the existing workflows. And here are some of the references for the papers that we published that describes these in more detail. So, thank you very much. \>\> \[applause\]