James MacGlashan
@jmac-ai.bsky.social
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.
Are policy gradient methods hopeless without other machinery? Deep learning works well when the Hessian is well conditioned. But the policy objective is under no obligation to give us that. And no amount of architecture engineering can rescue us from problems w/ the objective.
In the Assassin's Creed universe, Amodei would be a Templar, hands down. He seeks a perfect control he cannot have and in the quest for it will only increase the chaos he claims to fight. www.anthropic.com/news/positio...
Our position on open-weights models
Anthropic CEO Dario Amodei on open-weights models
anthropic.com
At Brown, I'm privileged to organize an AI policy summer school where we bring grad students from across the nation together to learn about policymaking. We travel down to DC as a group and learn about policy implementation and careers, and the students conduct meetings with Congressional offices.
Hugging Face used open AI models to combat the OpenAI model security breach because OpenAI models are closed AI models. Makes sense, right?
The AI community failed to exit twitter. What implications can we draw about what people believe from this?
What it means that the AI community can't quit twitter
Some quickly jotted thoughts working through the implications of the AI community remaining on twitter
rl-blogging.leaflet.pub
If we wanted more people to like and embrace AI, maybe we shouldn't have left a bunch of the worst possible people in charge? And maybe we should all collectively show a tiny modicum of empathy to people worried about their livelihoods and meaning in life... I'm really just spit-balling...
the secret sauce to the Mamdani Moment, which I think will produce a genuine difference in the civic and governing outcomes we come to expect from local leaders: this sense of a shared fate changes the game entirely he acts like he’s In It with his city, and that changes what he can ask OF them
An open model with this performance is great, but perhaps more striking is it looks like it's a hybrid model with 3/4th linear attention and 1/4 standard attention. This kind of performance with a hybrid model is a big deal! Enough with the memory hungry KV-caches!
mr. TIM@timkellogg.me · 3w agothis wasn’t supposed to happen yet
I resigned from Google DeepMind bc it broke its founding promise by selling AI to the military without restrictions against killer robots or mass spying. For months, I worked to stop this but watched powerful ethicists and institutions choose silence. Here's what happened. 🧵
Decouple reviews and curation from publication. Make reviews opt-in from anyone, usually deanonymized, and a contribution in its own right that others use to find good work. Build better social network tooling for discovery. The current game isn't winnable. Let the slop be published, but ignored.
Tony S.F.@tonysf.bsky.social · 4w agoBig lack of qualified reviewers? The cost of generation seems headed to 0 relative to the cost of verification; seems obvious that we must change norms so verification is seen as more of a contribution? What if we require authors to review for some number of conferences before being able to submit?
This is the feature I've been waiting for for some time.
Zed@zed.dev · last mo.A new (very Telescope) way to search has landed as well. Open it with the `text finder: toggle` action in the command palette. Thanks ozacod!
In various settings I see people who will say things like "I hate AI, but I used it to do <thing> because <grand defense of why their use is okay>" Then you don't hate AI. You might hate specific things about how specific AI tech was developed and deployed. It's okay to have a nuanced position.
Five years ago we asked: can an AI agent outrace the world's best #GranTurismo drivers? The answer became a @nature.com cover, a game feature and a research frontier that's still open. #GTSophy, five years on ↓ bit.ly/4vHyxpF
Gears are the "muscles" of robots and yet if you are like me you think robots are cool but have no mechanical engineering training So this weekend I made an explainer of different types of gears/reducers and where they are used now in robotics cpaxton.github.io/gears/index....
A little bird told her: scientist wins $100,000 prize for decoding birdsong #ai #news #artificialintelligence #technology
A little bird told her: scientist wins $100,000 prize for decoding birdsong
Julie Elie worked out how zebra finches announce who they are, what they are doing and use individual signatures
theguardian.com
Just in case there weren't enough reasons for people to consider GLM 5.2 as an alternative.
Zack Whittaker@zackwhittaker.com · last mo.In a new privacy policy, Anthropic says Claude will soon require you to upload your driver's license or passport for a variety of reasons. The ID checker is Persona, a firm funded by Trump ally Peter Thiel. As a U.S. company, Persona is also subject to gov't demands for people's verification data. 👀
I don't see the cost of memory coming down until we replace Multi-head attention with linear complexity methods such as those in the linear recurrence class (gated delta nets, etc.). We really need more work in that space.
TMLR has the best policies and approach to conventional publishing for AI. But I've long been of the mind that their approach still isn't enough to combat the systemic issues in conventional publishing. Quotas might help short-term, but we ultimately need to radically change our systems.
Transactions on Machine Learning Research@tmlrorg.bsky.social · last mo.TMLR has been facing an significant uptick in the number of submissions since the start of 2026. This is placing an extreme burden on our amazing team of reviewers & action editors. To ease this burden, TMLR will be implementing submission quotas, effective July 1. 1/n medium.com/@TmlrOrg/ann...
I talk about the research on when AI undermines, versus supporting, thinking and learning here: www.oneusefulthing.org/p/choosing-t...
Choosing to Stay Human
If you go to your favorite social media site, you will find it full of posts that start to look suspiciously similar to each other:
oneusefulthing.org
"Like a cat leaving a dead bird at your doorstep, Anthropic catalogs the grim future that its products might produce, shrugs its shoulders and then returns to its furious efforts to make these warnings a reality." 😂 www.nytimes.com/2026/06/17/o...
Opinion | Dear A.I. Companies: The Doom Trolling Needs to Stop
nytimes.com
Tbc, you cannot find an AI researcher who will approve of this nonsense. This is coming from business folks
Tim Fernholz@timfernholz.com · 2mo agopeople who work and invest in AI frequently ask me why their work polls so badly
I wrote a piece for the Yale Review on AI and "jagged intelligence". (Note: headline was not written by me.) yalereview.org/article/mela...
Melanie Mitchell: What We Get Wrong About AI
Melanie Mitchell probes the jagged landscape of AI and its uncertain future.
yalereview.org
@maphramusic.bsky.social continues to kill it. I really can't think of another vocalist like her. www.youtube.com/watch?v=uGSx...
Faouzia - Unethical (MAPHRA Vocal Cover)
YouTube video by MAPHRA
youtube.com
"The brain is a computer" is not an analogy. It's not likening it to a Von Neumann architecture, nor re-progammability. It's making a claim that is either true or false, but the claim that its false is far more outrageous. This claim also gives us useful tools, so its insight is the right question.
Sure feels like we're experiencing the most extreme version of Moravec's Paradox that anyone could imagine.
This is why I think we need a further shake up of how peer review and publishing works. Publishing should not be gate kept. Review should be public, ad hoc, discoverable, and part of a scientist's body of work that leads to tenure etc. Curation then sits on top of these distributed adhoc systems.
