korrents
Chip Huyen

What Chip Huyen thinks about LLMs

@chip-huyen · 37 positions · 1 change of mind · On ppll ↗

Engineer and writer on machine-learning systems; author of Designing Machine Learning Systems and AI Engineering.

Chip Huyen 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.

13 dated positions, 2023 to 2025, in their own words. Our reading of what Chip Huyen has said — not written or endorsed by them.

  1. AI judges must be evaluated and iterated over time, just like all other AI applications.

    ↗Common pitfalls when building generative AI applicationshuyenchip.com 3rd of 3 in this piece

  2. 6 months earlier
  3. Context construction for foundation models is equivalent to feature engineering for classical ML models.

    ↗Building A Generative AI Platformhuyenchip.com 1st of 3 in this piece

  4. In-context learning, learning from the context, is a form of continual learning.

    ↗Building A Generative AI Platformhuyenchip.com 2nd of 3 in this piece

  5. AI models are probabilistic, making their outputs unreliable.

    ↗Building A Generative AI Platformhuyenchip.com 3rd of 3 in this piece

  6. 5 months earlier
  7. Model routing has the potential to increase response quality while reducing costs and latency.

    ↗Predictive Human Preference: From Model Ranking to Model Routinghuyenchip.com 1st of 3 in this piece

  8. One pattern is that for simple prompts, weak models can do (nearly) as well as strong models. For more challenging prompts, however, users are much more likely to prefer stronger models.

    ↗Predictive Human Preference: From Model Ranking to Model Routinghuyenchip.com 2nd of 3 in this piece

  9. With more and more models being developed, each with different capabilities and a cost structure, model routing has clear economic values.

    ↗Predictive Human Preference: From Model Ranking to Model Routinghuyenchip.com 3rd of 3 in this piece

  10. 6 weeks earlier
  11. This probabilistic nature makes AI great for creative tasks.

    ↗Generation configurations: temperature, top-k, top-p, and test time computehuyenchip.com 1st of 3 in this piece

  12. The optimal thing to do with a fickle model, however, is to swap it out for another.

    ↗Generation configurations: temperature, top-k, top-p, and test time computehuyenchip.com 2nd of 3 in this piece

  13. I suspect that in the future, it'll be easier to get models to output exactly what we need with minimal prompting, and these techniques will become less important.

    ↗Generation configurations: temperature, top-k, top-p, and test time computehuyenchip.com 3rd of 3 in this piece

  14. 3 months earlier
  15. A model that can effectively learn from bitstrings or bytestrings will be very powerful, and it can learn from any data mode.

    ↗Multimodality and Large Multimodal Models (LMMs)huyenchip.com 1st of 3 in this piece

  16. Image is perhaps the most versatile format for model inputs, as it can be used to represent text, tabular data, audio, and to some extent, videos. There's also so much more visual data than text data.

    ↗Multimodality and Large Multimodal Models (LMMs)huyenchip.com 2nd of 3 in this piece

  17. Text is a much more powerful mode for model outputs. A model that can generate images can only be used for image generation, whereas a model that can generate text can be used for many tasks: summarization, translation, reasoning, question answering, etc.

    ↗Multimodality and Large Multimodal Models (LMMs)huyenchip.com 3rd of 3 in this piece