korrents
Lilian Weng

Lilian Weng

@lilian-weng · 16 positions · 0 changes of mind

Machine-learning researcher; has written the Lil'Log survey posts on how a model technique works since 2017, and worked at OpenAI from 2018 to 2024.

Everything they publish, on ppll ↗

Lilian Weng did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

  1. the layer between the raw model and the real-world context seems to be as important as the model's raw intelligence

    Harness Engineering for Self-Improvementlilianweng.github.io 1st of 3 in this piece

  2. recursive self-improvement

    Recursive structure alone is not enough. The base model must be capable enough to improve the mechanism.

    Harness Engineering for Self-Improvementlilianweng.github.io 2nd of 3 in this piece

  3. harness improvement enables better deployment of the model but intelligence is still the core.

    Harness Engineering for Self-Improvementlilianweng.github.io 3rd of 3 in this piece

  4. 10 days earlier
  5. As the model size grows significantly, we are running out of enough high-quality unique tokens.

    Scaling Laws, Carefullylilianweng.github.io

  6. 14 months earlier
  7. LLMs

    this self-correction capability turns out to not exist intrinsically among LLMs and does not easily work out of the box, due to various failure modes

    Why We Thinklilianweng.github.io 1st of 3 in this piece

  8. thinking for longer should be especially useful when the model is presented with an unusual input, such as an adversarial example or jailbreak attempt

    Why We Thinklilianweng.github.io 2nd of 3 in this piece

  9. LLMs

    model CoTs could be biased due to lack of explicit training objectives aimed at encouraging faithful reasoning

    Why We Thinklilianweng.github.io 3rd of 3 in this piece

  10. 5 months earlier
  11. designreinforcement learning

    Reward hacking exists because RL environments are often imperfect, and it is fundamentally challenging to accurately specify a reward function.

    Reward Hacking in Reinforcement Learninglilianweng.github.io 1st of 2 in this piece

  12. designAI agents

    A more intelligent agent is more capable of finding "holes" in the design of reward function and exploiting the task specification-in other words, achieving higher proxy rewards but lower true rewards.

    Reward Hacking in Reinforcement Learninglilianweng.github.io 2nd of 2 in this piece

  13. 5 months earlier
  14. LLMs

    To avoid hallucination, LLMs need to be (1) factual and (2) acknowledge not knowing the answer when applicable.

    Extrinsic Hallucinations in LLMslilianweng.github.io 1st of 2 in this piece

  15. LLMs

    These empirical results from Gekhman et al. (2024) point out the risk of using supervised fine-tuning for updating LLMs' knowledge.

    Extrinsic Hallucinations in LLMslilianweng.github.io 2nd of 2 in this piece

  16. 3 months earlier
  17. It has extra requirements on temporal consistency across frames in time, which naturally demands more world knowledge to be encoded into the model.

    Diffusion Models for Video Generationlilianweng.github.io 1st of 2 in this piece

  18. In comparison to text or images, it is more difficult to collect large amounts of high-quality, high-dimensional video data, let along text-video pairs.

    Diffusion Models for Video Generationlilianweng.github.io 2nd of 2 in this piece

  19. 2 months earlier
  20. Lots of ML techniques in the post can help with data quality, but fundamentally human data collection involves attention to details and careful execution.

    Thinking about High-Quality Human Datalilianweng.github.io

  21. 3 months earlier
  22. One drawback with UAT (Universal Adversarial Trigger) attacks is that it is easy to detect them because the learned triggers are often nonsensical.

    Adversarial Attacks on LLMslilianweng.github.io 1st of 2 in this piece

  23. High perplexity makes an attack more vulnerable to be detected and mitigated.

    Adversarial Attacks on LLMslilianweng.github.io 2nd of 2 in this piece