---
schema: 1
type: video
title: 'Special Topics in Kernels, RL, Reward Hacking in Agents — Daniel Han, Unsloth'
description: >-
  An advanced seminar (good prerequisites: Daniel's 2024 and 2025 hit AIE
  workshops, but all are welcome!) PLS WATCH:
  https://www.youtube.com/@aiDotEngineer/search?query=daniel%20han Timestamps:
  0:00 Introduction to Unsloth and model distribution 2:32 The State of AI:
  Meter plots and performance trends 20:26 Open Source vs. Closed Source models
  38:51 Throughput maxing and accuracy minimizing 1:03:00 Benchmarking and
  cheating in AI 1:37:49 Kernels and algorithmic improvements 2:04:16
  Reinforcement learning primer 2:05:19 Reward hacking and AI agents Viral
  Quotes & Pull Quotes: "If you make the model 86% smaller, it does not get 86%
  dumber... it only gets 14% less dumb." (29:52) "Reinforcement learning is
  terrible, but everything else is even worse." (20:43) "The model becomes not
  important anymore; it's the harness or the tool that is actually the most
  important thing." (38:23) "If a model can finish a task that takes a human 16
  hours, can a model finish that task?" (2:49) --- Watch along with the
  corrected transcript, summary, timestamps and resources on this talk's
  official AIE page:
  https://ai.engineer/talks/uIiA6DquRiE-special-topics-in-kernels-rl-reward-hacking-in-agents
  Join our upcoming AI Engineer conferences (see https://ai.engineer landing
  page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs
  and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference
  returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS
  2026 in Sydney ---
canonical: >-
  https://liquidgold.cc/channel/aidotengineer/explore/video/special-topics-in-kernels-rl-reward-hacking-in-agen-59e8ded2b3bc
markdown: >-
  https://liquidgold.cc/channel/aidotengineer/video/special-topics-in-kernels-rl-reward-hacking-in-agen-59e8ded2b3bc.md
llms: null
parent: 'https://liquidgold.cc/channel/aidotengineer.md'
channel: aidotengineer
updated: '2026-10-24T07:32:44.601Z'
language: null
children: []
videoId: uIiA6DquRiE
durationSeconds: 8421
publishedAt: '2026-07-17T17:00:06.000Z'
searchChunks:
  - ordinal: 0
    text: null
    startSeconds: 166
    endSeconds: 278
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=166'
    heldReason: rights
  - ordinal: 1
    text: null
    startSeconds: 288
    endSeconds: 368
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=288'
    heldReason: rights
  - ordinal: 2
    text: null
    startSeconds: 379
    endSeconds: 483
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=379'
    heldReason: rights
  - ordinal: 3
    text: null
    startSeconds: 494
    endSeconds: 594
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=494'
    heldReason: rights
  - ordinal: 4
    text: null
    startSeconds: 695
    endSeconds: 707
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=695'
    heldReason: rights
  - ordinal: 5
    text: null
    startSeconds: 796
    endSeconds: 929
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=796'
    heldReason: rights
  - ordinal: 6
    text: null
    startSeconds: 939
    endSeconds: 1051
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=939'
    heldReason: rights
  - ordinal: 7
    text: null
    startSeconds: 1051
    endSeconds: 1063
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=1051'
    heldReason: rights
  - ordinal: 8
    text: null
    startSeconds: 1478
    endSeconds: 1599
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=1478'
    heldReason: rights
  - ordinal: 9
    text: null
    startSeconds: 1610
    endSeconds: 1691
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=1610'
    heldReason: rights
  - ordinal: 10
    text: null
    startSeconds: 1702
    endSeconds: 1820
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=1702'
    heldReason: rights
  - ordinal: 11
    text: null
    startSeconds: 1831
    endSeconds: 1930
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=1831'
    heldReason: rights
  - ordinal: 12
    text: null
    startSeconds: 2060
    endSeconds: 2071
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=2060'
    heldReason: rights
  - ordinal: 13
    text: null
    startSeconds: 2486
    endSeconds: 2577
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=2486'
    heldReason: rights
  - ordinal: 14
    text: null
    startSeconds: 2915
    endSeconds: 3014
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=2915'
    heldReason: rights
  - ordinal: 15
    text: null
    startSeconds: 3025
    endSeconds: 3125
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=3025'
    heldReason: rights
  - ordinal: 16
    text: null
    startSeconds: 3136
    endSeconds: 3215
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=3136'
    heldReason: rights
  - ordinal: 17
    text: null
    startSeconds: 3227
    endSeconds: 3307
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=3227'
    heldReason: rights
  - ordinal: 18
    text: null
    startSeconds: 3802
    endSeconds: 3939
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=3802'
    heldReason: rights
  - ordinal: 19
    text: null
    startSeconds: 4083
    endSeconds: 4194
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=4083'
    heldReason: rights
  - ordinal: 20
    text: null
    startSeconds: 4205
    endSeconds: 4292
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=4205'
    heldReason: rights
  - ordinal: 21
    text: null
    startSeconds: 4304
    endSeconds: 4390
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=4304'
    heldReason: rights
  - ordinal: 22
    text: null
    startSeconds: 4493
    endSeconds: 4607
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=4493'
    heldReason: rights
  - ordinal: 23
    text: null
    startSeconds: 4617
    endSeconds: 4730
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=4617'
    heldReason: rights
  - ordinal: 24
    text: null
    startSeconds: 4849
    endSeconds: 4859
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=4849'
    heldReason: rights
  - ordinal: 25
    text: null
    startSeconds: 5195
    endSeconds: 5297
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=5195'
    heldReason: rights
  - ordinal: 26
    text: null
    startSeconds: 5429
    endSeconds: 5522
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=5429'
    heldReason: rights
  - ordinal: 27
    text: null
    startSeconds: 5534
    endSeconds: 5613
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=5534'
    heldReason: rights
  - ordinal: 28
    text: null
    startSeconds: 6176
    endSeconds: 6297
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=6176'
    heldReason: rights
  - ordinal: 29
    text: null
    startSeconds: 6310
    endSeconds: 6389
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=6310'
    heldReason: rights
  - ordinal: 30
    text: null
    startSeconds: 6399
    endSeconds: 6531
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=6399'
    heldReason: rights
  - ordinal: 31
    text: null
    startSeconds: 6542
    endSeconds: 6622
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=6542'
    heldReason: rights
  - ordinal: 32
    text: null
    startSeconds: 6724
    endSeconds: 6818
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=6724'
    heldReason: rights
  - ordinal: 33
    text: null
    startSeconds: 6830
    endSeconds: 6918
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=6830'
    heldReason: rights
  - ordinal: 34
    text: null
    startSeconds: 7077
    endSeconds: 7189
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=7077'
    heldReason: rights
  - ordinal: 35
    text: null
    startSeconds: 7298
    endSeconds: 7418
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=7298'
    heldReason: rights
  - ordinal: 36
    text: null
    startSeconds: 7529
    endSeconds: 7629
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=7529'
    heldReason: rights
  - ordinal: 37
    text: null
    startSeconds: 7641
    endSeconds: 7745
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=7641'
    heldReason: rights
  - ordinal: 38
    text: null
    startSeconds: 7756
    endSeconds: 7859
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=7756'
    heldReason: rights
  - ordinal: 39
    text: null
    startSeconds: 8322
    endSeconds: 8421
    watch: 'https://liquidgold.cc/watch/uIiA6DquRiE?t=8322'
    heldReason: rights
---


# Special Topics in Kernels, RL, Reward Hacking in Agents — Daniel Han, Unsloth

An advanced seminar (good prerequisites: Daniel's 2024 and 2025 hit AIE workshops, but all are welcome!) PLS WATCH: https://www.youtube.com/@aiDotEngineer/search?query=daniel%20han Timestamps: 0:00 Introduction to Unsloth and model distribution 2:32 The State of AI: Meter plots and performance trends 20:26 Open Source vs. Closed Source models 38:51 Throughput maxing and accuracy minimizing 1:03:00 Benchmarking and cheating in AI 1:37:49 Kernels and algorithmic improvements 2:04:16 Reinforcement learning primer 2:05:19 Reward hacking and AI agents Viral Quotes & Pull Quotes: "If you make the model 86% smaller, it does not get 86% dumber... it only gets 14% less dumb." (29:52) "Reinforcement learning is terrible, but everything else is even worse." (20:43) "The model becomes not important anymore; it's the harness or the tool that is actually the most important thing." (38:23) "If a model can finish a task that takes a human 16 hours, can a model finish that task?" (2:49) --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/uIiA6DquRiE-special-topics-in-kernels-rl-reward-hacking-in-agents Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney ---

Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/special-topics-in-kernels-rl-reward-hacking-in-agen-59e8ded2b3bc · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md)

> Search chunk text is held until public-source provenance is verified.

## Document chunks

> Missing: no stored document chunks are available.