James MacGlashan
@jmac-ai.bsky.social
2.2K followers 1.1K following 610 posts
Ask me about Reinforcement Learning Research @ Sony AI AI should learn from its experiences, not copy your data. My website for answering RL questions: https://www.decisionsanddragons.com/ Views and posts are my own.
Posts Media Videos Starter Packs
Pinned
jmac-ai.bsky.social
Reinforcement learning in #AI is hard, so I’ve made a website to collect answers I’ve given to common RL questions.

It's named Decisions & Dragons. It’s launching with 8 questions and answers, but I will add to it in the future.

A 🧵 to give a preview with the link below.
Reposted by James MacGlashan
atrupar.com
C-SPAN caller to Mike Johnson: "Hearing you say that 'everyone is smiling' in cities where troops have been rolled into feels dystopian and insane watching the response to it."
jmac-ai.bsky.social
What a catastrophic misunderstanding of science. I wish he was just a kook with no impact on the world. Instead he's affecting policy at the highest level.
atrupar.com
RFK Jr on Tylenol and autism: "It is not proof. We're doing the studies to make the proof."
jmac-ai.bsky.social
If you do a good empirical evaluation and also can provide some understanding about the choice, that is definitely valuable.

Many conferences will probably penalize it, but I'm hopeful it's the kind of thing that would still be accepted at TMLR or similar places.
jmac-ai.bsky.social
I'm not sure what point you're trying to make about algorithms because I didn't say anything about them.

You shouldn't buy the hype about LLMs. You certainly won't hear me telling you to.

You asked if there was non-LLM AI actively used and developed, and the answer is yes, there is a lot!
jmac-ai.bsky.social
Beyond that there is a long history of AI applications covering things from logistics planning to recommender systems.

These varied technologies, both old and new, came from the AI research community. LLMs are just trendy and the natural language interface catches the general public's attention.
jmac-ai.bsky.social
You may be pleased to learn that there's an enormity of AI research and application that is not LLMs :)

For example, I worked on GT Sophy which learned from experience to out-race the best players in GT:Sport. You can race against variants of it in GT7 on PS5.

www.gran-turismo.com/us/gran-turi...
Gran Turismo Sophy
Gran Turismo Sophy's website. GT Sophy is a revolutionary superhuman racing AI agent that has been featured on the cover of the world's leading science journal, Nature. Project overview, innovative AI...
www.gran-turismo.com
Reposted by James MacGlashan
markkho.bsky.social
I'm recruiting grad students!! 🎓

The CoDec Lab @ NYU (codec-lab.github.io) is looking for PhD students (Fall 2026) interested in computational approaches to social cognition & problem solving 🧠

Applications through Psych (tinyurl.com/nyucp) are due Dec 1. Reach out with Qs & please repost! 🙏
codec lab
codec-lab.github.io
jmac-ai.bsky.social
It's imperative that we begin having trusted digital signing of photo/video content and editors that allow you track original camera source and editors/encoders that are used to modify the content.

On top of that, we need tooling to verify signatures.

No, we don't need blockchain for that.
jmac-ai.bsky.social
There's a very important difference. Data was designed to be an artificial person complete with personal goals and, eventually, human-like emotions.

Modern AI is not and it's an anti goal to pursue AP. It's important that we all understand this difference & that AI does not imply AP.
Reposted by James MacGlashan
cristianfarias.com
This video of Chicagoans intervening to save a man from being abducted off the streets by ICE is making the rounds on Instagram.

Community action works.

Source: www.instagram.com/reel/DPZL2AL...
Reposted by James MacGlashan
beenwrekt.bsky.social
In non-LLM AI news, Waymo has quietly been involved in two fatal car crashes in 2025. Phil Koopman details that this fatality rate is now comparable to humans.

People: 7/billion miles in all conditions
Waymo: 5/billion miles in geofenced conditions
Blame-Free Robotaxi Crashes Are Still Crashes
The tricky logic behind robotaxi crash metrics
philkoopman.substack.com
Reposted by James MacGlashan
kojamf.bsky.social
Dr. Jane Goodall filmed an interview with Netflix in March 2025 that she understood would only be released after her death.
Reposted by James MacGlashan
katmabu.bsky.social
Let us in. Hold ICE accountable. Tear down the fence.

This is what we are demanding from the Broadview ICE facility, and nothing less.

Our leaders can decide whether they stand with the people or with ICE. There is no middle ground.
Reposted by James MacGlashan
jmac-ai.bsky.social
I don't care if you like AOC's politics. She's one of the very few federal politicians making an effort to lead against the unprecedented authoritarianism of the Trump regime.
the-goddess-speaks.bsky.social
🗣️THIS is the message!!

Brava, Congresswoman @aoc.bsky.social!!
Reposted by James MacGlashan
Reposted by James MacGlashan
propublica.org
The rate of sepsis in Houston surged 63% after Texas banned abortion.

In the Dallas-Fort Worth area, where hospital leaders empowered doctors to intervene before patients’ conditions worsened, it rose 29%.

(Published May)
Why Hospital Policies Matter in States That Ban Abortion
Our first-of-its-kind data analysis found that a seismic split emerged in how medical institutions in Texas' two largest metro areas, Houston and Dallas, treated miscarrying patients — and in their ou...
www.propublica.org
jmac-ai.bsky.social
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...
Reinforcement Learning for Gran Turismo: Success and Sharp Edges - James MacGlashen (Sony AI)
YouTube video by Reinforcement Learning and Video Games Workshop
www.youtube.com