Chelsea Finn
Professor at Stanford working on robot learning and meta-learning; co-founded Physical Intelligence.
Chelsea Finn did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake
Latest x.com
What they believe
-
Every machine-learning deployment that has paid off so far left a person making the decision, which is the only reason imperfect models were useful.
And in all of these applications, the customer is making a decision based off of the recommendation of the AI model more or less. Uh, and this means that if the customer is ultimately like kind of making the decision, this means that if the system makes a mistake, um, that's okay because usually the person can kind of recognize that or or decide what to do even despite that mistake.
youtube.comSaid 12 Aug 2026This on korrents
The bold line is korrents’ wording; the quote is theirs. These one: what they wrote before what they said aloud, one per source, newest first.
Feed
As its own page →Hiding
24 October
21 October
13 October
9 October
4 October
23 September
15 September
14 September
23 August
12 August
From one piece Chelsea Finn: This is the State of the Art in Robotics 15 beliefs · youtube.com
-
korrents.com
Robot models will reach ChatGPT-level capability within the next few years, even without a ChatGPT-style moment.Their words
at the same time in terms of the capabilities of these models I think that we are really starting to get to the point where these models are actually useful in the real world and I think that getting to the kind of the capabilities of chat GBT I think is um yeah very much on the horizon in the next few years.
-
Their words
I think that the distribution channel for physical models is going to be slower uh unfortunately because you actually need a physical robot there
-
Their words
without metadata prompting when you add lower quality data from 80% data to 100% data the performance actually decreases which is perhaps not too surprising because you're adding lowquality data to your data mixture whereas with the metadata prompting the performance actually increases when you add that lowquality data
+ 12 more
-
Their words
we find that the performance on held out tasks decreases dramatically. Whereas if we um just take out a random 20% of the data that's less diverse than the most diverse subset, the performance um only decreases a little bit. And so this suggests that actually having really diverse data plays an important role in enabling it to generalize to new tasks.
-
scaling lawsroboticsreinforcement learning
Their words
we see that the across the board the single PIO like pre-trained PIO7 model matches or outperforms the fine-tuned specialists that were developed with reinforcement learning post-training for those downstream tasks.
-
Their words
when you can basically bridge the concept of an avocado and a chair and show that you kind of can combine those two, it means that the model has at least some kind of conceptual understanding of what an avocado is and what a chair is to the point that it can combine them into something that exhibits both concepts at the same time.
-
korrents.com
A model you have to fine-tune for each thing you want it to do is not a general-purpose model.Their words
But if you have to fine-tune a model, you actually aren't getting a general purpose model um for the things that you want it to do because you have to fine-tune it for each individual thing.
-
Their words
And then for longer memory, for memory that spans multiple minutes or multiple hours, we don't necessarily need video of exactly what happened in that past uh in that past history.
-
Their words
So, you might be surprised to hear that most state-of-the-art foundation models for robotics have no memory or no context. They're just operating on the current sensor observations, the current camera readings, uh, and predicting actions based off of that.
-
Their words
kind of going back to this reliability question, we took this policy and we ran it not just once, but we ran it for 13 hours straight. Uh and we basically wanted to evaluate is this policy not only good at making a latte once, but can it do so reliably to the extent that it would be needed to be useful in the real world?
-
Their words
and trying to fold two boxes together isn't useful data that will teach the model how to get better at the task. And so that would be kind of wasting a lot of time on the robot attempting to go down the wrong path for solving the problem. And so instead of spending a lot of time trying to do that task, what we'll do is we'll actually have a human intervene and show the robot what to do and how to recover from that situation.
-
LLMsroboticsreinforcement learning
Their words
Now maybe this isn't completely out of the question but this would be quite challenging uh to do and that's because the calculus is a little bit different. We're not just running compute to optimize for a use case. We're actually running the robot in the real world and using the hardware and attempting the task in the real world.
-
Their words
Um, and while this generally improves the reliability of the model, uh, people eventually get tired and it's hard to get really, really high reliability with a person that's manually tuning this. And so what would be even better is if the AI system itself can iterate on the scenario in which you want it to have higher reliability where it on its own automatically seeks out places where it needs more data, where it needs more supervision.
-
Their words
Uh and this means that they're going to be far more useful when they're operating fully autonomously. And as a result, this requires us to develop physical AI systems that make far fewer mistakes than the machine learning systems that have been deployed thus far.
-
Their words
And in all of these applications, the customer is making a decision based off of the recommendation of the AI model more or less. Uh, and this means that if the customer is ultimately like kind of making the decision, this means that if the system makes a mistake, um, that's okay because usually the person can kind of recognize that or or decide what to do even despite that mistake.
9 August
9 July
3 July
26 June
19 January 2023
3 February 2022
21 January 2022
9 February 2021
19 January 2020
Nothing matches.
Beliefs
Korrents What they believe 22 beliefs — each backed by an exact quote.
Each is a — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.
Recent
Every machine-learning deployment that has paid off so far left a person making the decision, which is the only reason imperfect models were useful.
And in all of these applications, the customer is making a decision based off of the recommendation of the AI model more or less. Uh, and this means that if the customer is ultimately like kind of making the decision, this means that if the system makes a mistake, um, that's okay because usually the person can kind of recognize that or or decide what to do even despite that mistake.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Physical AI has to make far fewer mistakes than any AI deployed so far, because it is only useful when it acts autonomously on the world.
Uh and this means that they're going to be far more useful when they're operating fully autonomously. And as a result, this requires us to develop physical AI systems that make far fewer mistakes than the machine learning systems that have been deployed thus far.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Hand-tuning a robot's data set cannot reach very high reliability, because people get tired; the system has to seek out its own missing data.
Um, and while this generally improves the reliability of the model, uh, people eventually get tired and it's hard to get really, really high reliability with a person that's manually tuning this. And so what would be even better is if the AI system itself can iterate on the scenario in which you want it to have higher reliability where it on its own automatically seeks out places where it needs more data, where it needs more supervision.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Show 19 more
Reinforcement learning cannot be scaled for robots the way it was for language models, because every attempt spends real robot hours instead of data-centre compute.
Now maybe this isn't completely out of the question but this would be quite challenging uh to do and that's because the calculus is a little bit different. We're not just running compute to optimize for a use case. We're actually running the robot in the real world and using the hardware and attempting the task in the real world.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
A robot going down a wrong path is not producing training data; the efficient move is for a human to intervene and show the recovery.
and trying to fold two boxes together isn't useful data that will teach the model how to get better at the task. And so that would be kind of wasting a lot of time on the robot attempting to go down the wrong path for solving the problem. And so instead of spending a lot of time trying to do that task, what we'll do is we'll actually have a human intervene and show the robot what to do and how to recover from that situation.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Whether a robot policy is any good is settled by running it for thirteen hours straight, not by whether it completes the task once.
kind of going back to this reliability question, we took this policy and we ran it not just once, but we ran it for 13 hours straight. Uh and we basically wanted to evaluate is this policy not only good at making a latte once, but can it do so reliably to the extent that it would be needed to be useful in the real world?
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Most state-of-the-art robot foundation models have no memory at all, and that is what stops them doing long multi-step tasks.
So, you might be surprised to hear that most state-of-the-art foundation models for robotics have no memory or no context. They're just operating on the current sensor observations, the current camera readings, uh, and predicting actions based off of that.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
A robot does not need video to remember the last hour: past a few seconds, memory can be a text summary of what happened.
And then for longer memory, for memory that spans multiple minutes or multiple hours, we don't necessarily need video of exactly what happened in that past uh in that past history.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
A model you have to fine-tune for each thing you want it to do is not a general-purpose model.
But if you have to fine-tune a model, you actually aren't getting a general purpose model um for the things that you want it to do because you have to fine-tune it for each individual thing.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Combining an avocado and a chair into one image is evidence a model conceptually understands both, not that it memorised pictures of them.
when you can basically bridge the concept of an avocado and a chair and show that you kind of can combine those two, it means that the model has at least some kind of conceptual understanding of what an avocado is and what a chair is to the point that it can combine them into something that exhibits both concepts at the same time.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
One pre-trained robot model now matches or beats the specialists that were fine-tuned with reinforcement learning for the very tasks they were built for.
we see that the across the board the single PIO like pre-trained PIO7 model matches or outperforms the fine-tuned specialists that were developed with reinforcement learning post-training for those downstream tasks.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
It is the diversity of a robot data set, not its size, that produces generalization: dropping the most diverse slice hurts far more than dropping a random fifth.
we find that the performance on held out tasks decreases dramatically. Whereas if we um just take out a random 20% of the data that's less diverse than the most diverse subset, the performance um only decreases a little bit. And so this suggests that actually having really diverse data plays an important role in enabling it to generalize to new tasks.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Low-quality data makes a robot model worse unless you tell the model the data is low quality; labelled as such, the same data makes it better.
without metadata prompting when you add lower quality data from 80% data to 100% data the performance actually decreases which is perhaps not too surprising because you're adding lowquality data to your data mixture whereas with the metadata prompting the performance actually increases when you add that lowquality data
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Robotics will not get a ChatGPT moment, because the distribution channel for a physical model is a physical robot and that cannot spread in five days.
I think that the distribution channel for physical models is going to be slower uh unfortunately because you actually need a physical robot there
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Robot models will reach ChatGPT-level capability within the next few years, even without a ChatGPT-style moment.
at the same time in terms of the capabilities of these models I think that we are really starting to get to the point where these models are actually useful in the real world and I think that getting to the kind of the capabilities of chat GBT I think is um yeah very much on the horizon in the next few years.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
A small team should start from an open-source generalist policy and fine-tune it on day one, rather than scale up its own specialist models first.
I mean at the very least I actually think that just starting with a generalist policy and then fine-tuning it even like right off the bat uh can be really effective.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Watching humans is not enough for a robot: there is no substitute for experience gathered on its own body.
Uh and likewise um robots can't like watch a person doing something and then figure out how to do it themselves directly. They really need their experience on their own platform um to learn effectively.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Most future robotics training data will be robots' own autonomous attempts, the way most language-model data is now generated by the model itself.
Uh and just like how we see in language models how now a lot of time is spent actually generating data, generating synthetic data by actually running the model and having it think through things. I think a lot of the data in the future in robotics is going to be the robot attempting to do lots of tasks in lots of real world circumstances.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Letting a robot imagine the next image helps, but it is not essential — the model was surprisingly good without it.
uh the we found that that leads to improvement and we saw in the shirt folding example we saw like a quantitative bump from using that sort of imagination compared to not using it. At the same time I think that the model actually performed surprisingly well without that as well.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Robots are slow because the people teleoperating their training data were slow, so the only two ways out are faster data or a policy faster than its data.
and yeah, I think it's either you need to figure out how to make the data faster or you need to figure out how to be faster than the data. We've seen the evidence of being able to be a little bit faster than the data.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
The lasting value of a PhD is learning to handle uncertainty and pick your own problems, which is exactly what working at the frontier of AI demands.
But I think the PhD is an incredible opportunity to first learn a lot about how to handle uncertainty, how to pick good problems to work on.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
Robot policies already show emergent capability: one transferred a skill from its right hand to its left with no such example anywhere in its training data.
Um the robot essentially had learned this sort of equivariance between his left hand and his right hand so that it could actually transfer uh behaviors from one hand to another. uh despite the fact that that was never in the data.
Chelsea Finn: This is the State of the Art in Robotics Said 12 Aug 2026
What is a korrent?
A korrent is a belief a person has stated in their own words: one sentence stating the claim, backed by a quote and a source, kept at korrents.com.
Under a name here, the quoted block is what they actually said. The korrent beneath it is the claim those words support, in korrents' wording — tap it to see the record, its source, and who else holds it.
Nobody here wrote their own korrents. They are compiled from public statements, and a person can change their mind, which is recorded too.
Where they publish
Site Chelsea Finn Her faculty page: papers, teaching and her group.
Link verified 20 Sept 2026.
x.com @chelseabfinn Their X account.
Recent
- Generative value classifiers are a new parametrization of Q-functions that don't require action conditioning. We think these may combine gracefully with large, pre-trained policies and learn/generalize better from a richer output signal. Paper: 24 Oct 2026 alexanderswerdlow.github.io
- How do we scale up the context of LLMs? Agentic search scales to arbitrarily large amounts of info, but can easily miss relevant info. Idea💡: Optimize organization of info for downstream utility. Leads to 30+% gains even on OOD domains Paper + code: 21 Oct 2026
- The Pi API is officially in beta! 🤖 Many people have already been using it to fine-tune π models for their robots -- below are a few We’re hoping to expand testing significantly. Let us know if you are interested! Blog: https://t.co/JBTahDOmg7 Signup: 13 Oct 2026
Show 15 more
- As robots become more reliable, DAgger becomes far less efficient. eg 90% success means the person spends >90% of the time passively watching the robot Key idea: failures come from similar initial states & we can focus learning on these states! Paper: 9 Oct 2026 arxiv.org
- Scientific discovery is one of the most exciting applications of LLMs, but we lack data & evals on the scientific process. ScholarCatalyst captures which prior papers can advance new research on a problem, rather than citations or topical relevance. Strikingly, even frontier models struggle to identify which prior research is helpful -- lots of room for future work! Paper, Code, Data: 4 Oct 2026 ohmyksh.github.io
- New blog post with @perryadong on what we need for robots to be broadly useful in the real world. https://t.co/jxsNeMUhFg Reliability is the one of the biggest open challenges in AI right now. Current models work out okay if a person will be reviewing the outputs (eg drafting code), but it will become more of a bottleneck as we want systems to act with more autonomy and more trust. 23 Sept 2026
- RL for dynamic tasks requires handling latency. Building on EXPO-FT, we let the base VLA operate on older images while a small edit policy operates in real time. Outperforms RTC by a significant margin, across the board. Paper: 23 Sept 2026 pd-perry.github.io
- A video from a Pi robot deployed at Dandelion Chocolate, fully autonomous w/ no interventions. 🤖 Deploying robots has taught us surprising lessons about the gap between proof-of-concept (i.e. building one box) and real-world utility (productively building boxes for hours). I expected that the hard part is building the box, since it’s the most dexterous, but that wasn't what we found. Counterintuitively, the hardest part was reliably stacking the boxes. Our original table-mounted robot had poor visibility of the stack without special separately-mounted cameras. Plus, stacking requires more ge… 15 Sept 2026
- Congrats to @zipengfu, @chenwang_j, and co on the release! 14 Sept 2026
- One of the most important aspects of scientific discovery is deciding where to draw insights from. While LLMs are promising tools for science, we lack datasets & evaluations for this step. Help contribute to a public dataset for exactly this: 23 Aug 2026 tinyurl.com
- Pretraining a Q-function often doesn’t actually help RL finetuning, compared to initializing Q from scratch. We find that pretraining Q-functions on data from diverse policies is critical to see improvements from pretraining. Paper: 9 Aug 2026 arxiv.org
- I'm giving a talk tomorrow at ICML on emergent physical generalization, including π0.7 🤖 3:15 pm @ SCALE workshop in Ballroom 201 9 Jul 2026 scale-icml-2026.github.io
- I'm giving a talk on how we can move beyond the scalar reward bottleneck for both robotics & LLMs. ICML RLxF workshop tomorrow at 1:30 pm. 9 Jul 2026
- We’ve been approaching reward supervision for robots the wrong way. I think freeform preferences are part of the answer. A short 🧵 3 Jul 2026
- LLM RL optimizes for sequential reasoning We also optimize over the reasoning strategy, incl parallel trains of thought, aggregation of parallel traces, & sequential reasoning This allows the model to better explore & allocate compute at test time 26 Jun 2026 arxiv.org
- Can we translate a rough sense for what to do + VLA prior into successful behavior? Flow reversal steering: 1) runs the VLA's flow ODE backwards to back out noise that's closest to coarse traj 2) runs flow forwards to get closest good behavior Paper: 26 Jun 2026 flow-reversal-steering.github.io
- How does test-time scaling impact robots? We find that larger models, more thinking, and more context help significantly for some prompts but not others. Like LLMs, we can also train a router to for a better performance/latency tradeoff! Paper: 12 Jun 2026
- Can robot foundation models collaborate with themself? We finetune a VLA to be able to control any robot in a team. - matches or outperforms training separate models or a single centralized model for all robots - readily scales to large teams Paper: 12 Jun 2026 arxiv.org
Link verified 23 Sept 2026. Recent items update automatically from the channel.
GitHub @cbfinn Their code.
Recent
- maml_rl — Code for RL experiments in "Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks" 19 Jan 2023
- bcz-videos 3 Feb 2022
- bair_camp 21 Jan 2022
Show 2 more
Link verified 23 Sept 2026. Recent items update automatically from the channel.
About the English under a post
Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.
Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.