korrents
Jay Alammar

Jay Alammar

@jay-alammar · 7 positions · On ppll ↗

Writes Language Models & Co., a newsletter on large language models; known for The Illustrated Transformer, referenced in AI/ML courses at MIT, Stanford and Cornell.

Jay Alammar 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.

Filter All topics · Newest first
Topic
All 7LLMs 4AI agents 1open source 1
Sort
NewestOldest

  1. Our reading

    Running large batches of prompts through an LLM API is a capability unavailable to someone who only uses models through a chat playground interface.

    LLMs

    Their wordsRunning thousands of prompts is one of the superpowers you do not get to use if all of your LLM usage is through a playground.

    ↗Inside NeurIPS 2025: The Year’s AI Research, Mappednewsletter.languagemodels.co 1st of 2 in this piece

  3. Our reading

    Delegating work to many AI agents at once is not yet a wise choice because of the costs involved.

    AI agents

    Their wordsIt's probably not even wise yet to delegate to many agents given the possible costs involved.

    ↗Inside NeurIPS 2025: The Year’s AI Research, Mappednewsletter.languagemodels.co 2nd of 2 in this piece

  4. 2 months earlier

  5. Our reading

    Open source AI models today largely share similar architectural approaches, with little genuinely novel architecture left to introduce.

    open source

    Their wordsNote that little of these architectural details is particularly novel. They're generally similar in the latest SoTA open source MoE models.

    ↗The Illustrated GPT-OSSnewsletter.languagemodels.co 1st of 2 in this piece

  7. Our reading

    More reasoning gives a language model more time and compute to solve harder problems, but increases latency and compute cost.

    LLMs

    Their wordsOn the one hand, more reasoning allows the model more time and compute to reason about a problem which helps it tackle more difficult problems. On the other hand, that comes at a cost of latency and compute.

    ↗The Illustrated GPT-OSSnewsletter.languagemodels.co 2nd of 2 in this piece

  8. 7 months earlier

  9. Our reading

    Base large language models have now crossed a threshold of quality and capability significant enough to change what training methods are viable.

    LLMs

    Their wordsModern base models have crossed a certain threshold of quality and capability (this base model was trained on 14.8 trillion high-quality tokens).

    ↗The Illustrated DeepSeek-R1newsletter.languagemodels.co 1st of 2 in this piece

  11. Our reading

    Reasoning problems, unlike general chat or writing tasks, can be automatically verified or labeled without human annotation.

    Their wordsReasoning problems, in contrast to general chat or writing requests, can be automatically verified or labeled.

    ↗The Illustrated DeepSeek-R1newsletter.languagemodels.co 2nd of 2 in this piece

  12. 3 months earlier

  13. Our reading

    Language models are not merely text generators; they can also form other systems, such as embeddings and classifiers, that are useful for problem solving.

    LLMs

    Their wordsA central concept the general public should know is that language models are not merely text generators, but that they can form other systems (embedding, classification) that are useful for problem solving.

    ↗Our book, Hands-On Large Language Models, Is Now Out!newsletter.languagemodels.co