If you want to hear about why I'm simultaneously grumpy and optimistic about RL for video games, you can now view my talk at the RLC 2025 RL and Video Game Workshop below! www.youtube.com/watch?v=8DSl...

Missed our workshop at @rl-conference.bsky.social? Watch all keynote presentations and the panel discussion on YouTube: youtube.com/playlist?lis... Big thanks to our awesome speakers: @togelius.bsky.social, @jmac-ai.bsky.social, Roberta Raileanu, Michael Bowling and @pcastr.bsky.social!

A very difficult lesson to learn as tech-nerd since my very earliest memories is that, yes, we will build the technology of your dreams. And then people will use it in the worst and dumbest ways that you cannot even imagine.

Crockett: "I'm gonna honor someone who decided they were going to proclaim I was involved in the 'great white replacement'? Yeah, I'm not honoring that ... it's unfortunate more of my colleagues even on my side couldn't see the amount of harm this man was attempting to inflict upon our communities"

NEW: ChatGPT is causing chaos in marriages, as one spouse becomes deeply fixated on AI therapy/advice/spiritual wisdom — alienating the other spouse and, often, resulting in divorce. In some cases, ChatGPT-enmeshed spouses are using the tech to bully their partners. futurism.com/chatgpt-marr...

I wrote a response to Thomas Friedman's "magical thinking" on AI here: aiguide.substack.com/p/magical-th...

Sony AI is hiring game integration engineers! We do awesome RL applications to modern video games. If that excites you, check out the posting! We have positions for senior and staff level developers. Senior dev link (staff level link next in 🧵): sonyglobal.wd1.myworkdayjobs.com/en-US/SonyGl...

Looooool

I do think you’ve gotta hand it to Ezra Klein for this one, which we need more of, much more, from his peers www.nytimes.com/2025/09/07/o...

9 CDC Directors going back to 1977 speak out. What RFK Jr has done to our nation’s public health system "should alarm every American." It "is unlike anything we have ever seen at the agency, and unlike anything our country has ever experienced." www.nytimes.com/2025/09/01/o...

New: In 7 months, Kennedy has broken the CDC, perhaps irreparably. For all he’s done with mass firings, gutting expertise, and attacking vaccines, Wednesday’s loss of its most senior and competent leaders—truly the CDC’s backbone—is the most consequential attack yet. www.msnbc.com/msnbc/news/c...

The Sony AI Game AI team has reinforcement learning internships opens for 2026! It is remote for people in the US & Canada, mixed remote/onsite in Europe (onsite in Zurich), and onsite in Tokyo. If you want to work on RL with cool applications, sign up! ai.sony/joinus/job-r...

Sony AI has new post on why Reinforcement Learning matters, featuring yours truly and some of my awesome colleagues: Varun Kompella, Harm van Seijen, and Peter Stone. You can read it here: ai.sony/blog/Sony-AI...

Monday's talk "Two Tales of Reward Design: GT Sophy and Factored Value Functions" is online: youtu.be/PrsKX5ZWt_4 #RL #GranTurismo #GTSophy

The "TIME100 AI 2025" list makes me want to vomit.

This imagery is the kind of thing you see in video games to let you know you’re in a totalitarian city or country.

A few years in, a CDC drowning prevention program was ready to share its findings on how to mitigate the leading cause of death among young children. Then the administration terminated that staff.

America now

Adam Raine, 16, died from suicide in April after months on ChatGPT discussing plans to end his life. His parents have filed the first known case against OpenAI for wrongful death. Overwhelming at times to work on this story, but here it is. My latest on AI chatbots: www.nytimes.com/2025/08/26/t...

"A scientist believes..." isn't noteworthy unless it can be followed by "because science shows..."

Can folks do me a favor? I was on Bluesky early & it was great. Mostly. I hit the eject button in January. I came back, many colleagues are here, but still don’t know I am. I co-created Halt and Catch Fire. I write TV & comics. Here’s two favorite set photos. Repost? I am in the wilderness.

An expectant mom got a $750 bill for a blood test to check for genetic abnormalities in her baby. Then she tried to figure out why it was so high. n.pr/3V0E221

Anthropic caring about "model well being" for their LLMs is pseudoscientific nonsense, and it's an alarming position to take. Do you notice how the people who claim to worry about existential threat of AI to people are suddenly moving to care about models instead? 🧵 1/4

We can’t work from the old playbook. It’s time for a new generation of unapologetically progressive leadership.

I scrolled to this while I had Hans Zimmer's "No Time for Caution" from Interstellar playing in the background. Perfection.

This one's been a long time coming. In this post on Decisions & Dragons I answer "Should we abandon RL?" The answer is obviously no, but people ask because they have a fundamental misunderstanding of what RL is. RL is a problem, not an approach. www.decisionsanddragons.com/posts/should...

It sucks that I have to think about whether I should delete an em dash from my text in fear that people will think I had an LLM write it.

Back on June 23rd I posted a .pdf of a piece on AI hype that I had written 37 years ago. Here it is transcribed in my blog, with some commentary on two of the things that I said in 1988 had been overhyped in 1949 and 1962. rodneybrooks.com/ai-great-exp...

This was a fantastic talk!

Today is my birthday (turning 29), which is the only time all year that I ask for a repost 🌿 I hope you get lost in fantasy worlds as much as I do 🔮

Even if @rust-lang.org never fully takes off and becomes one of the more dominant languages, I think learning it was worthwhile because it made me a better programmer in other languages. It forces you to write in a way that is inherently clearer and easier to maintain. That translates.

Yes, but have we considered writing another book about how Biden was old?

I agree with this quote. However, many people assume that researchers are not interested in it. That couldn't be further from the truth. The amount of work motivated by robots/AI doing chores is boundless. We are here only because writing/image gen is easier and a step toward those goals. 1/3

Pretty sure documenting code well takes me about twice as long as writing it.

The boy who cried wolf would out right lie about events to prank people. The story wasn't about sounding the alarm for legitimate warning signs. Are we really failing to understand childhood fables?

Need for Speed or: How I Learned to Stop Worrying About Sample Efficiency Part II of my blog series "Getting SAC to Work on a Massive Parallel Simulator" is out! I've included everything I tried that didn't work (and why Jax PPO was different from PyTorch PPO) araffin.github.io/post/tune-sa...

I am grieving the barbarism that is going to unfurl from all this. People are going to die. Livelihoods gone. All to feed a corrupt kleptocracy. I see every day up close how different it is from the first time around. There are no guardrails. A disaster. I’m sorry we have to live through this.

NEW: Elon Musk’s lobbyists spent tens of thousands of dollars on things like gifts and meals for Texas elected officials this year. But, in most cases, state transparency laws do not require them to disclose which elected officials they wine and dine. With @kutnews.bsky.social & @texastribune.org

We have @littlelobbyists.com families on the way to Washington DC right now, they'll be on the House-side of the Capitol tomorrow. We're going to make them see us before they vote to kill disabled children. If you can help cover the costs of this urgent action: secure.actblue.com/donate/littl...

Yes, I flipped off a bigot harassing toddlers at a drag queen story hour. No, I won't apologize for it: www.youtube.com/watch?v=SXd7...

At my current stage in life, my opinion is that p-values are fine, actually. It’s people’s interpretation that’s not. As long as you use them as just another statistic, much like a mean, they’re handy. Granted, I’m not sure what to do about it being difficult for people to interpret correctly.

And how will it do that?

The space of tasks where LLMs are useful is similar in structure to NP problems. That is, problems where you can verify correctness quickly, but generating the answer (via LLM) is time consuming.

live from the huge spontaneous anti-ICE protest after the agency used military-style tactics during a series of raids in Los Angeles today.

No door dash, I really don't need to snap a photo of my taco bell before I dig in.

What's more remarkable to me (well, not really. I expect it from him.) is that he has no self awareness that buying an election isn't something you should be proud of. And short of that, not something you should advertise.

Cuts have consequences, illustrated. As seen on TV 📺

Video shows two unknown men, without uniforms or identifying insignias, attacking and kidnapping a man from a Nashua, NH courthouse back in February. During the kidnapping, the assailants, presumed to be ICE agents, knocked an elderly bystander who walks with a cane to the ground.

It is more important to lower the variance RL algorithms than it is to increase the mean performance.