> ## Content Index
> Fetch the complete content index at: https://www.antoinebuteau.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Lessons from Percy Liang
- URL: https://www.antoinebuteau.com/lessons-from-percy-liang/
- Published: 2026-05-26T23:18:34.000Z
- Updated: 2026-09-05T03:07:24.000Z
- Description: Percy Liang, a Stanford computer science professor and Together AI co-founder, defined foundation models and built HELM to evaluate them, advancing transparent, open-source infrastructure while probing language understanding, evaluation systems, and the limits of machine reasoning.
- Author: Antoine Buteau
- Tags: Profile, AI & Machine Learning Profiles

![Visual summary of operating lessons from Percy Liang.](https://www.antoinebuteau.com/content/images/2026/05/lessons-from-percy-liang-profile-infographic.webp)

Percy Liang defined the concept of foundation models and built the HELM benchmark to evaluate them. A Stanford computer science professor, director of the Center for Research on Foundation Models, and Together AI co-founder, he advocates for transparent, open-source AI infrastructure. This profile collects his perspectives on language understanding, evaluation systems, and machine reasoning.

## Part 1: The Paradigm of Foundation Models

1. **On the distinction between ML and human intelligence:** "I think it's absolutely right to think of machine learning AI as not chasing human intelligence, but more — they're a different thing... tools that we build to help us." — [*Source: \[Behind the Tech*](https://news.microsoft.com/?ref=antoinebuteau.com)*\]*
2. **On naming the era:** "The term large language models doesn't really underscore the importance... we called it foundation models, because we felt like there's a paradigm shift where we all of a sudden have these huge models trained on broad data, which served as a foundation for developing other products." — [*Source: \[Stanford Engineering*](https://engineering.stanford.edu/?ref=antoinebuteau.com)*\]*
3. **On historical equivalents:** "AI today is kind of like where we were with the internet in 1993\. We were at the beginning of a revolution that would change the world." — [*Source: \[AI2050*](https://ai2050.schmidtfutures.com/?ref=antoinebuteau.com)*\]*
4. **On emergence:** He notes that emergence catches researchers by surprise—the fact that simply training a model to predict the next token can result in answering questions and summarizing text. — [*Source: \[Stanford CRFM*](https://crfm.stanford.edu/?ref=antoinebuteau.com)*\]*
5. **On the risks of homogenization:** "Now almost all NLP models are built on top of BERT, or maybe one of a few of these foundation models... So there's this incredible homogenization that's happening." — [*Source: \[Fast Company*](https://www.fastcompany.com/?ref=antoinebuteau.com)*\]*
6. **On systemic failure:** Foundation models create a single point of failure, meaning any defects, biases, or security vulnerabilities are blindly inherited by downstream tasks. — [*Source: \[Fast Company*](https://www.fastcompany.com/?ref=antoinebuteau.com)*\]*
7. **On changing goals:** "While humans used to be the paragon for general intelligence, I don't think it is productive to use humans as the goal post anymore, as the standards for AI should be higher in some ways." — [*Source: \[Schmidt Sciences*](https://www.schmidtsciences.org/?ref=antoinebuteau.com)*\]*
8. **On the role of academia:** While industry commands the vast compute resources, academia must provide the scientific standards and exploration that companies overlook in the rush to productize. — [*Source: \[No Priors Podcast*](https://www.youtube.com/?ref=antoinebuteau.com)*\]*
9. **On automation in education:** "I want to emphasize that a lot of AI is also going to automate really bad ways of teaching. So \[we need to\] think about it as a way of creating new types of teaching." — [*Source: \[Stanford HAI*](https://hai.stanford.edu/?ref=antoinebuteau.com)*\]*

## Part 2: Redefining Evaluation and Benchmarking

1. **On benchmarks as guides:** "Benchmarks serve as the North Star for the AI community, providing clear targets to aim for." — [*Source: \[HubSpot Blog*](https://blog.hubspot.com/?ref=antoinebuteau.com)*\]*
2. **On encoding values:** "Benchmarking is more than just measurement—it encodes values. Benchmarks determine what the community prioritizes." — [*Source: \[HubSpot Blog*](https://blog.hubspot.com/?ref=antoinebuteau.com)*\]*
3. **On the crisis of evaluation:** "The pace of innovation is happening so quickly that we're not able to measure... we're living in a crisis in our ability to measure model quality." — [*Source: \[HubSpot Blog*](https://blog.hubspot.com/?ref=antoinebuteau.com)*\]*
4. **On holistic measurement:** To evaluate language models effectively, researchers must acknowledge incompleteness, use multi-metric measurement, and enforce standardization across all models. — [*Source: \[HELM Paper*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
5. **On broader metrics:** Evaluation must extend beyond raw accuracy to encompass robustness, fairness, bias, toxicity, and efficiency. — [*Source: \[Princeton AI*](https://ai.princeton.edu/?ref=antoinebuteau.com)*\]*
6. **On transparency:** "Transparency is the vital first step... Many language models exist, but they are not compared on a unified standard." — [*Source: \[Stanford University*](https://news.stanford.edu/?ref=antoinebuteau.com)*\]*
7. **On democratic alignment:** The evaluation process needs to be more democratic so that the values a model reflects represent what people actually want, not just the decisions of a few people behind closed doors. — [*Source: \[Stanford Engineering*](https://engineering.stanford.edu/?ref=antoinebuteau.com)*\]*
8. **On evaluation as a distinct discipline:** "Evaluation has now... become more of a first-class citizen in a way that it was just not before. It was always: you train a model, you evaluate." — [*Source: \[Lex Fridman Podcast*](https://lexfridman.com/?ref=antoinebuteau.com)*\]*
9. **On the purpose of testing:** "We should start by thinking about the goals of evaluation: is it to provide feedback and incentives for improving a system or is it to assess the absolute quality of a system?" — [*Source: \[NIST Panel*](https://www.nist.gov/?ref=antoinebuteau.com)*\]*
10. **On factual boundaries:** "I would push back on saying that the hallucinations in itself is fundamentally bad... what are the notion of facts is also a little bit hazy. Now in a given application, there turns out to be things that you should say and things that the language model should not output." — [*Source: \[Stanford Engineering*](https://engineering.stanford.edu/?ref=antoinebuteau.com)*\]*

## Part 3: The Mechanics of Semantic Parsing and NLP

1. **On the definition of semantic parsing:** "Semantic parsers map natural language into logical forms, the classic representation for many important linguistic phenomena." — [*Source: \[Communications of the ACM*](https://cacm.acm.org/?ref=antoinebuteau.com)*\]*
2. **On logical forms:** "We can think of the logical form as a program that is executed to yield the desired behavior." — [*Source: \[Communications of the ACM*](https://cacm.acm.org/?ref=antoinebuteau.com)*\]*
3. **On the core problem of semantics:** "How do you map a natural language utterances into some sort of executable program or meaning representation?" — [*Source: \[Behind the Tech*](https://news.microsoft.com/?ref=antoinebuteau.com)*\]*
4. **On the history of the field:** Early rule-based natural language understanding systems were sophisticated but brittle; the modern resurgence uses logical representations paired with statistical learning. — [*Source: \[JHU Seminar*](https://www.jhu.edu/?ref=antoinebuteau.com)*\]*
5. **On learning from denotations:** It is possible to learn a semantic parser purely from question-answer pairs, treating the intermediate logical form as a latent variable rather than requiring expensive manual annotation. — [*Source: \[MIT AI*](https://www.csail.mit.edu/?ref=antoinebuteau.com)*\]*
6. **On bootstrapping parsers:** By generating a canonical formal language and having users paraphrase it into natural language, developers can rapidly build semantic parsers for entirely new domains overnight. — [*Source: \[ACL Anthology*](https://aclanthology.org/?ref=antoinebuteau.com)*\]*
7. **On lambda dependency-based compositional semantics:** λ-DCS was designed as a compact formal language to represent the semantics of natural language more efficiently than traditional lambda calculus. — [*Source: \[OpenReview*](https://openreview.net/?ref=antoinebuteau.com)*\]*
8. **On structural understanding:** Modern large language models still struggle with compositional semantics, exposing gaps in their ability to understand the deep logical structure of language. — [*Source: \[Lex Fridman Podcast*](https://lexfridman.com/?ref=antoinebuteau.com)*\]*
9. **On evaluating language understanding:** The modern twist on classic linguistic representation is dealing with the statistical and computational issues introduced by learning semantic parsers directly from data. — [*Source: \[Communications of the ACM*](https://cacm.acm.org/?ref=antoinebuteau.com)*\]*

## Part 4: Navigating the Open Source vs. Closed Model Debate

1. **On the fallacy of closed security:** "Security by obscurity has always been frowned upon as something that can't be relied on. Malicious, incentivized actors often use that obscurity to get early access to zero-days." — [*Source: \[Together Talks*](https://www.together.ai/blog?ref=antoinebuteau.com)*\]*
2. **On internet-scale infrastructure:** "If you look at the history of computer security... we should build the entire internet ecosystem on Open Standards like SSL... where everyone can stare at each other." — [*Source: \[Together Talks*](https://www.together.ai/blog?ref=antoinebuteau.com)*\]*
3. **On true open source:** Current open weight models are not truly open source because the critical recipes—the training data, code, and evaluation logs—remain secret. — [*Source: \[Together AI Blog*](https://www.together.ai/blog?ref=antoinebuteau.com)*\]*
4. **On building a Linux for AI:** The mission is to create a decentralized, transparent ecosystem for foundation models that prevents the capture of AI by a few closed-source corporations. — [*Source: \[Redpoint Podcast*](https://www.redpoint.com/?ref=antoinebuteau.com)*\]*
5. **On decentralizing compute:** Distributed training across disparate networks and data centers is essential so that frontier research is not confined to a single massive supercomputer. — [*Source: \[Together AI Tech Blog*](https://www.together.ai/blog?ref=antoinebuteau.com)*\]*
6. **On architectural efficiency:** Memory-aware attention mechanisms like FlashAttention drastically reduce the bottlenecks of Transformers, making it viable for the broader community to train and run large models. — [*Source: \[ArXiv*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
7. **On open development:** It is critical to practice open development, where every experiment, success, and failure is preregistered and visible to the community in real-time. — [*Source: \[Marin Lab*](https://crfm.stanford.edu/?ref=antoinebuteau.com)*\]*
8. **On the transparency index:** The AI industry is trending toward dangerous secrecy regarding training data and labor practices, necessitating a formal index to track and audit corporate transparency. — [*Source: \[Stanford HAI*](https://hai.stanford.edu/?ref=antoinebuteau.com)*\]*
9. **On global accessibility:** Open-source models are essential for reducing English-language bias and making advanced AI accessible to developers worldwide. — [*Source: \[No Priors Podcast*](https://www.youtube.com/?ref=antoinebuteau.com)*\]*
10. **On data efficiency:** Rigorous, open scientific tuning of data recipes can lead to massive increases in data efficiency, allowing state-of-the-art performance with a fraction of the traditional data volume. — [*Source: \[Together AI Research*](https://www.together.ai/?ref=antoinebuteau.com)*\]*

## Part 5: Reliability and Distribution Shifts

1. **On average-case failure:** "Standard machine learning produces models that are accurate on average but degrade dramatically when the test distribution of interest deviates from the training distribution." — [*Source: \[WILDS Benchmark*](https://wilds.stanford.edu/?ref=antoinebuteau.com)*\]*
2. **On the limits of scale:** "We have found many surprises in our quest for robustness: for example, that the 'more data' and 'bigger models' strategy that works so well for average accuracy sometimes fails out-of-domain." — [*Source: \[MLR Press*](https://proceedings.mlr.press/?ref=antoinebuteau.com)*\]*
3. **On redefining reliability:** "What is one to do with an 80 percent reliable system? ... The conventional metric for success—average accuracy—is not a good interface for AI safety." — [*Source: \[ADT Mag*](https://adtmag.com/?ref=antoinebuteau.com)*\]*
4. **On uncertainty as safety:** "Safe deployment requires AI systems that expose uncertainty to humans... ensuring that an AI understands, and is able to communicate, its limits." — [*Source: \[Stanford Engineering*](https://engineering.stanford.edu/?ref=antoinebuteau.com)*\]*
5. **On real-world fragility:** Distribution shifts that happen naturally—such as deploying a medical model in a new hospital—consistently degrade system accuracy in ways standard testing fails to capture. — [*Source: \[ICML*](https://icml.cc/?ref=antoinebuteau.com)*\]*
6. **On ignored data shifts:** Despite being ubiquitous in real-world deployments, practical distribution shifts have historically been under-represented in the benchmark datasets favored by the ML community. — [*Source: \[WILDS Benchmark*](https://wilds.stanford.edu/?ref=antoinebuteau.com)*\]*
7. **On the failure of robust algorithms:** Standard training methods often outperform specialized robustness algorithms on real-world in-the-wild data, highlighting the need for fundamentally new approaches. — [*Source: \[ResearchGate*](https://www.researchgate.net/?ref=antoinebuteau.com)*\]*
8. **On mission-critical risk:** "I think it's really important that we focus on the issue of robustness, especially when these systems are going out into more mission-critical situations like self-driving cars or medical diagnostics." — [*Source: \[Behind the Tech*](https://news.microsoft.com/?ref=antoinebuteau.com)*\]*
9. **On debugging predictions:** Influence functions provide a mathematical framework for debugging model reliability by identifying the exact training points that caused a potentially unsafe prediction. — [*Source: \[ICML*](https://icml.cc/?ref=antoinebuteau.com)*\]*

## Part 6: Simulating Society and Generative Agents

1. **On pure exploration:** "I think that was a really fun project because it was pure kind of exploration. It was like, 'We don’t know what’s going to happen, let’s build it and see and simulate.'" — [*Source: \[The AI Podcast*](https://www.redpoint.com/?ref=antoinebuteau.com)*\]*
2. **On believability versus validity:** "While generative agents was about creating believable simulations... if you could get simulations that were actually valid in the sense that they reflected something in reality, then you unlock a lot of different new vistas." — [*Source: \[Redpoint Ventures*](https://www.redpoint.com/?ref=antoinebuteau.com)*\]*
3. **On digital twins:** "For the first time, because we have these models, you can actually simulate something in much greater detail than was ever possible... allowing you to run policy experiments \[on a\] digital twin of society." — [*Source: \[TWiML AI Podcast*](https://twimlai.com/?ref=antoinebuteau.com)*\]*
4. **On emergent social dynamics:** When agents are equipped with a memory stream, reflection, and planning capabilities, they naturally exhibit complex social behaviors like spreading information and coordinating events. — [*Source: \[Generative Agents Paper*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
5. **On the limits of sandbox games:** While creating agents that simply look human is sufficient for video games, the true scientific potential lies in modeling specific real-world populations for sociological research. — [*Source: \[Stanford Research*](https://news.stanford.edu/?ref=antoinebuteau.com)*\]*
6. **On architectural importance:** The success of generative agents relies less on the raw scaling of the underlying foundation model and more on the cognitive architecture—specifically how memory and interaction are structured. — [*Source: \[Lex Fridman Podcast*](https://lexfridman.com/?ref=antoinebuteau.com)*\]*
7. **On simulating interaction:** Language models act as a cognitive engine, but the environment and the structured memory retrieval are what actually ground the agents in a continuous social reality. — [*Source: \[ArXiv*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
8. **On policy testing:** Validated social simulations could eventually allow researchers to test the downstream effects of urban planning or economic policies before implementing them in the real world. — [*Source: \[The AI Podcast*](https://www.redpoint.com/?ref=antoinebuteau.com)*\]*
9. **On autonomous agents:** The transition from static prompt-response chatbots to autonomous agents capable of long-term problem-solving requires entirely new evaluation metrics focused on trajectories rather than static answers. — [*Source: \[Together AI Blog*](https://www.together.ai/blog?ref=antoinebuteau.com)*\]*

## Part 7: Test-Time Compute and Reasoning

1. **On the next paradigm:** "From a research perspective, I think this signals sort of a change that I think we'll see going in the future. And I think the idea of test-time compute—it's been around, but as with many things, it needs a certain kind of scale for it to sink in for people." — [*Source: \[Redpoint Podcast*](https://www.redpoint.com/?ref=antoinebuteau.com)*\]*
2. **On the limits of early reasoning products:** While test-time compute is a research breakthrough, early implementations were often slow and difficult to integrate into practical user experiences. — [*Source: \[The AI Podcast*](https://www.redpoint.com/?ref=antoinebuteau.com)*\]*
3. **On scaling inference:** "Test-time scaling is a promising new approach to language modeling that uses extra test-time compute to improve performance... setting the stage for a new scaling paradigm." — [*Source: \[s1 Paper*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
4. **On budget forcing:** "We seek the simplest approach to achieve test-time scaling... 1,000 examples + budget forcing." — [*Source: \[s1 Paper*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
5. **On suppressing early termination:** By suppressing a model's end-of-thought tokens and forcing it to generate a "Wait" token, researchers can force the model to think longer and arrive at better answers. — [*Source: \[ArXiv*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
6. **On System 2 thinking:** Test-time compute represents the transition from fast, intuitive token prediction (System 1) to slow, deliberate, multi-step reasoning (System 2). — [*Source: \[Together AI Blog*](https://www.together.ai/blog?ref=antoinebuteau.com)*\]*
7. **On latent reasoning:** Complex reasoning capabilities are often already latent in pre-trained models; they do not necessarily require millions of reinforcement learning steps to unlock. — [*Source: \[s1 Paper*](https://arxiv.org/?ref=antoinebuteau.com)*\]*
8. **On data quality for reasoning:** A mere 1,000 highly curated reasoning traces can be sufficient to train a model that matches the reasoning capabilities of much larger, RL-heavy systems. — [*Source: \[ACL Anthology*](https://aclanthology.org/?ref=antoinebuteau.com)*\]*
9. **On sequential scaling:** Letting a single model think deeper sequentially is generally a more effective strategy for solving complex problems than running parallel generations and taking a majority vote. — [*Source: \[Stanford CRFM*](https://crfm.stanford.edu/?ref=antoinebuteau.com)*\]*
10. **On the bitter lesson:** While raw compute and data scaling have driven recent progress, achieving true logical reasoning requires a return to structural and algorithmic innovations at inference time. — [*Source: \[Lex Fridman Podcast*](https://lexfridman.com/?ref=antoinebuteau.com)*\]*

## Part 8: The Pursuit of Reproducible Research

1. **On capturing pipelines:** "Worksheets allow you to capture complex research pipelines in a reproducible way and create 'executable papers' where researchers attempting to reproduce the results should be able to execute the experiment within the same environment." — [*Source: \[CodaLab Documentation*](https://codalab.org/?ref=antoinebuteau.com)*\]*
2. **On the mission of reproducibility:** "In general, I am a strong proponent of efficient and reproducible research. I created CodaLab Worksheets... to maintain the full provenance of an experiment from raw data to final results." — [*Source: \[Stanford Faculty Profile*](https://profiles.stanford.edu/?ref=antoinebuteau.com)*\]*
3. **On collaborative ecosystems:** The goal of computational research platforms is to establish a collaborative ecosystem where experiments are conducted efficiently and can be universally verified. — [*Source: \[Microsoft Faculty Summit*](https://www.microsoft.com/?ref=antoinebuteau.com)*\]*
4. **On the lack of provenance:** The current machine learning research process suffers from a severe lack of provenance, making it incredibly difficult for peers to reuse or validate published findings. — [*Source: \[The Gradient Podcast*](https://thegradient.pub/?ref=antoinebuteau.com)*\]*
5. **On executable papers:** Shipping academic papers as executable environments ensures that the scientific community can instantly verify, modify, and build upon the precise code and data used in the study. — [*Source: \[Generally Intelligent Podcast*](https://imbue.com/?ref=antoinebuteau.com)*\]*
6. **On true reproducibility:** Reproducibility is not simply about posting source code on GitHub; it requires sharing the complete, immutable computational environment and the historical execution graph. — [*Source: \[Simons Foundation*](https://www.simonsfoundation.org/?ref=antoinebuteau.com)*\]*
7. **On immutability in research:** For an experiment to be truly reproducible, the underlying data and code must be fixed and immutable, preventing silent updates from altering historical results. — [*Source: \[CodaLab Documentation*](https://codalab.org/?ref=antoinebuteau.com)*\]*
8. **On transparency of the process:** Every single step, from raw data ingestion to the generation of the final evaluation plot, must be transparently visible to independent reviewers. — [*Source: \[Stanford CRFM*](https://crfm.stanford.edu/?ref=antoinebuteau.com)*\]*
9. **On the common task method:** The AI community advances fastest when researchers compete and collaborate on shared, rigorously standardized benchmarks with fully reproducible execution pipelines. — [*Source: \[Simons Foundation*](https://www.simonsfoundation.org/?ref=antoinebuteau.com)*\]*