The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
9:58 AM · May 18, 2026 · 447.4K Views
Richard Sutton, University of Alberta professor, shares a 26-word distillation of his Bitter Lesson urging AI researchers to favor computation-scaling methods over embedded human knowledge
AI Judge changed title after evaluation, original title: "Richard Sutton, University of Alberta professor and Turing Award winner, posts 26-word summary of the Bitter Lesson favoring scalable search and learning over human knowledge"
Gary Marcus and others reply debating reliance on human data and priors.
3718.6K1K3.4K710.9K
Sentiment
Positive users praise Richard Sutton's Bitter Lesson insights on scalable search and learning without human knowledge or taste, while negative users dismiss the claims as unrealistic and lacking evidence.
Pos
50.4%
Neg
49.6%
51 comments with sentiment.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS40.3KLIKES361RETWEETS29REPLIES20
Jiaxin Wen@jiaxinwen22
I might be one of the few people who is most bearish on human research taste and bullish on automated research: - "AIs can only do hyperparameter search" is mainly a skill issue with bad automated research setups. - human taste is overrated, e.g. frontier labs / neolabs are doing pretty simlar things. - human taste might win in a low-compute world, but not a high-compute world we're entering.
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 40.3KLikes 361Bookmarks 159
BOOKMARKS184
Herbie Bradley@herbiebradley
Some takes about RSI from discussions with many smart researchers & thinkers:
1. Many RSI (or automated AI R&D) debates converge to similar cruxes: is a 1000x sample efficiency improvement possible, can you just simulate reality and train on it with no sim2real gap, can we easily make models good at "fuzzy" tasks? People like to assume that automated research agents will find such breakthroughs specifically *because* without them, progress could be heavily bottlenecked on data or continued compute scale-ups.
2. The Yudkowsky "genius brain in a box" framing of ASI has latent influence on many researcher views even though people may not be aware of it. A common move is to "flip" predictions, as they go further out, from assuming LLM or deep learning-specific properties of future AI to assuming "von Neumann x1000", human brain-like properties. I'd like to see more thought-out reasoning of why this flip should occur at any particular point (eg pre or post automated AI R&D)—this question is a crux behind many predictions like AI 2027.
3. There are some cracks in this worldview beginning to show: predictions from a few years ago that models would be less jagged now than they are, or that they would be more deceptive, synthetic data would work better, etc. Many of these seem like prediction errors from imagining future models as a "human brain in a box", but LLMs are empirically a different kind of intelligence. Most models of software-only intelligence explosion are also coarse enough to mostly ignore properties of LLMs.
4. Views about fast RSI progress seem to be correlated with (a) belief that synthetic data is all you need (b) belief in very high GDP growth and an industrial explosion because of automated firms (c) having worked only in AI research or in small organizations.
5. Key technical things to track over the next 1-2 years: does RL increase in its generalization, AI lab data spend, can we automate synthetic RL env construction, best practices for FDEs deploying AI into large enterprises, coherency of AI personas, how powerful will multi-agent scaling of test-time compute be, and continual learning.
6. Overall I think the "RSI leading to *fast* takeoff" frame had huge alpha in 2022, moderate in 2024, and potentially is of neutral usefulness in 2026 for predicting the future.
42dViews 20.1KLikes 263Bookmarks 184
Dileep George@dileeplearning
Here's a better lesson, don't fall for bitter lesson.
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 31.6KLikes 231Bookmarks 105
Rishabh Agarwal@agarwl_
Perplexed by this take: Sure, let's not mainly do supervise learning on human knowledge, but it makes sense to build off it instead of the *let's do it from scratch*.
People cite AlphaGo vs AlphaGo Zero as a quintessential example of how using human-generating data is suboptimal but it was *imitating* it that was suboptimal.
What if we learned from that data assuming it was suboptimal in the first place (so not supervised learning but RL like mindset of using that data)
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 24.8KLikes 129Bookmarks 67
kache@yacineMTB
GOD GAVE US THE UNIVERSE, THE ORACLE. WE MUST MINE IT
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 19.8KLikes 221Bookmarks 49
Chirasmita Mallick@chirasmita16
@RichardSSutton wrote "The Bitter Lesson" in 2019.
Silicon Valley read it and scaled LLMs to the moon.
Sutton says they got it wrong. 🧵
41dViews 14.7KLikes 85Bookmarks 61
rohan anil@_arohan_
Two humans armed with same codex / claude code version and can produce completely different outcomes and measured on log scale.
Pranav Shyam@recurseparadox
falseposting. value of understanding at all time highs 📈
42dViews 17.5KLikes 128Bookmarks 18
Gary Marcus@GaryMarcus
i wonder whether this needs an update.
current methods, such as they are, leverage massive amounts of human knowledge as their primary fuel. they would be lost without it.
and they even build some knowledge into their system prompts.
and lately they build knowledge into their harnesses, usually by over 50 tools that have been carefully crafted with human knowledge.
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 7.9KLikes 68Bookmarks 15
Bojan Tunguz@tunguz
One way of looking at the bitter lesson.
42dViews 2.3KLikes 36Bookmarks 14
Thomas G. Dietterich@tdietterich
@RichardSSutton I think it is worth studying human knowledge, particularly understanding the structure of its abstractions, as they can provide guidance about the kinds of things humans learn and machines do not (yet). I wouldn't call that "distraction".
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 7.1KLikes 65Bookmarks 8
Shubhendu Trivedi@_onionesque
The bitter lesson is that we are doomed to endless bitter lesson exegesis.
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 3.9KLikes 41Bookmarks 7
Jiaxin Wen@jiaxinwen22
However, I hope humans keep doing our own research, with *strong* tastes and priors. Not all research is about outcomes. Sometimes you just want to solve/understand a problem in a way that feels like yours.
Jiaxin Wen@jiaxinwen22
I might be one of the few people who is most bearish on human research taste and bullish on automated research: - "AIs can only do hyperparameter search" is mainly a skill issue with bad automated research setups. - human taste is overrated, e.g. frontier labs / neolabs are doing pretty simlar things. - human taste might win in a low-compute world, but not a high-compute world we're entering.
42dViews 2.6KLikes 33Bookmarks 8
Julian Togelius@togelius
Meanwhile, Claude's system prompt is the size of a novel, and the harness is the size of a small operating system. Modern LLMs are trained on most of human knowledge. "AI" operates in a human world, and intelligence cannot be cleanly separated from knowledge about the world.
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
41dViews 5.2KLikes 28Bookmarks 6
ib@Indian_Bronson
@RichardSSutton yeah cc: @tszzl @khoomeik
https://plato.stanford.edu/entries/epistemology-india/
More AI researchers should read the Nyāya-Sūtras and Tattvacintāmaṇi.
42dViews 674Likes 11Bookmarks 8
Xiao Ma@infoxiao
bitter and sad
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 6.7KLikes 17Bookmarks 7
Pranav Shyam@recurseparadox
falseposting. value of understanding at all time highs 📈
Ryan Brewer@ryanbrewer
My entire job is now codex and managing codex threads, I’m genuinely curious what the software engineering job even is anymore. The value of my understanding of any system goes down every single day. Very weird times
42dViews 20.7KLikes 31Bookmarks 4
Lucas Beyer (bl16)@giffmana
@agarwl_ discarding humand data (be it in LLM or in non-trivial "classic" RL env) is just complete nonsense. Even if Rich says it. But I'm not sure he's actually saying this here.
Guess he should use more than 26 words, but then it wouldn't sound Ilya-style mysterious anymore :)
Rishabh Agarwal@agarwl_
Perplexed by this take: Sure, let's not mainly do supervise learning on human knowledge, but it makes sense to build off it instead of the *let's do it from scratch*.
People cite AlphaGo vs AlphaGo Zero as a quintessential example of how using human-generating data is suboptimal but it was *imitating* it that was suboptimal.
What if we learned from that data assuming it was suboptimal in the first place (so not supervised learning but RL like mindset of using that data)
42dViews 2.2KLikes 15Bookmarks 4
Wenting Zhao@wzhao_nlp
❤️
Richard Sutton@RichardSSutton
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
42dViews 8.3KLikes 17Bookmarks 4
Minh Nhat Nguyen@menhguin
@jiaxinwen22 median human taste is not great. the tails are pretty good, but for the purposes of "will AI replace a profession" the media is not hard to beat.
Jiaxin Wen@jiaxinwen22
I might be one of the few people who is most bearish on human research taste and bullish on automated research: - "AIs can only do hyperparameter search" is mainly a skill issue with bad automated research setups. - human taste is overrated, e.g. frontier labs / neolabs are doing pretty simlar things. - human taste might win in a low-compute world, but not a high-compute world we're entering.
42dViews 1.6KLikes 21Bookmarks 2
Herbie Bradley@herbiebradley
Related interesting point
christian@curious_vii
It's a good question. My own perspective on this is downstream of three pretty strong metaphysical claims:
1. The future is not only uncertain, not only unknowable, not only incalculable, but actually unimaginable. And so input generated from unbounded exploration on some frontier—be it scarce context, digital exhaust, or embodied experience—are essential inputs to the goal-setting process.
2. The second is that goals that are also good cannot be cast arbitrarily and must instead be discovered by meditating upon and then contemplating the aforementioned data in one. This is where I think I fundamentally disagree with some of the synthetic data and automated research maximalists. AI models themselves cannot contemplate. An AI system can be useful as macroscopic lens or sensor/actuator network tissue (and a pretty good substitute for bureaucracy per se), but at the end of the day, only a person—an embodied human being—can actually discern what is good and desirable, latent in that data.
3. The third is that those good goals are emergent only in the context of some relationship between human persons. You might be able to discover something that is technically true through some kind of radical commitment to scientific inquiry, but unless all that is ultimately tied to that which is good that another person can also contemplate, then it is not good or ultimately desirable in the sense that it invites actions at the expense of other actions.
Now, with all that in mind, there are myriad approaches to generating the raw material from which you can derive good goals and the means by which you can iteratively pursue them. But I don't think you can ever take a person out of the loop, even if those loops grow by many orders of magnitude from here on out.
42dViews 1.7KLikes 11Bookmarks 4
Load more posts