Frankly going to implore progressives to follow better what is happening in China tech. It's not a VC bubble. It's a dozen top labs, it's most large co getting heavily into AI (at a pretraining level) and robotics, it's a deep bet on the next industrial revolution.

Pink-sorrel flowering at the back of @theul.bsky.social by Grange Road. #WildflowerHour

You can slice a cube in half and get a hexagon ??

If you are an AI person recently here, follow Naomi's guide to getting a Bluesky feed that works for you. I'll also plug my "AI for Grownups" feed: bsky.app/profile/did:...

The Intelligence is in the User https://tomstafford.substack.com/p/the-intelligence-is-in-the-user Today's newsletter, on ways we underestimate how much user skill goes into getting good results from AI models #genai #cogsci #llms #chatgpt

Many people find ways to compliment and extend their cognitive abilities with AI (and many unfortunately don't extend but replace). In that context, what we have now and what would continue to be valuable in the future is ACI: Artificial Complimentary Intelligence...

might be time

Thinking about the genius of withholding funding from American universities, making it harder to pay for college with financial aid and hostility to foreign students & foreign workers then saying you’re in an AI race with China.

I deeply appreciate how machine learning advances are encouraging us to look deeply into the cognitive sciences. The better we understand biological consciousness and agency, the better chance we have of greater empathy for other people and the ecosystems we shepherd.

thinking about the $100k H-1b price, it disproportionately affects more junior people, in an environment where i think AI already is making it difficult to see value in junior roles it’s going to be a rough few years

🌱 Good morning🌞 #gardening #gardensky #bloomscrolling

Really exciting to see the ideas from DreamCoder working here!

This is not OK. I don't submit often to NeurIPS, but I reviewed papers for this conference almost every year. As a reviewer, why would I spend time trying to give a fair opinion on papers if it's what happens in the end???

We got accepted at #NeurIPS2025. I am very happy that I could merge my knowledge of Mathematics with AI to create sth new and useful for the community. ☺️ The paper: arxiv.org/abs/2505.17190 The code: github.com/Baran-phys/T...

🧪 This is an interesting paper! Researchers built Delphi-2M, an AI that adapts LLM principles to health records. By analyzing the sequence and timing of past diagnoses, it can forecast a patient's future risk for over 1,000 different conditions. #MedSky #MLSky

I finally got around to making a tool to compare completions from SFT vs. RLHF trained models. This is a mini site for the RLHF book that I've wanted for a while. buff.ly/lqDL5wa It's always been hard to say what RLHF does to a model within a more complex post-training pipeline.

If you have opportunities to get out and do things with others, jump at the opportunity: meetings, classes, social events, etc... Get out and be with other people. It's essential. If virtual is more convenient, then take advantage of those opportunities to be with other people.

I asked "what is a door, anyway?" after trying to generate "an open door" in a lot of midjourney styles. In my latest newsletter, along with more "door" pics. And the usual AI/games/tools news (and narrative research). arnicas.substack.com/p/titaa-705-...

🤖 As AI-generated content is shared in movies/TV/across the web, there's one simple low-hanging fruit 🍇 to help know what's real: Visible watermarks. With others @hf.co, I've made sure it's trivially easy to add this disclosure to images, video, chatbot text. See how: huggingface.co/blog/waterma...

my political beliefs: - everyone should have real freedom, not just the in-group - everyone should have a decent standard of living - abundance + tech go brrr - YIMBY urbanism with lots of trains - staunch bodily autonomy (drugs, transition, etc.) - immigration is good, actually

As AI becomes increasingly capable of conducting analyses and following instructions, my prediction is that the role of scientists will increasingly focus on identifying and selecting important problems to work on ("selector"), and effectively evaluating analyses performed by AI ("evaluator").

What will AI look like by 2030 if current trends hold? Our new report zooms in on two things: (1) whether scaling continues (compute, data, power, capital), and (2) the capabilities this enables—especially for scientific R&D.

📢 New #COLM2025 paper 📢 Standard benchmarks give every LLM the same questions. This is like testing 5th graders and college seniors with *one* exam! 🥴 Meet Fluid Benchmarking, a capability-adaptive eval method delivering lower variance, higher validity, and reduced cost. 🧵

cannot tell you how awesome i find it when people are super interested in or hyper competent at something i have never ever thought about before.

I've been testing the accuracy of LLMs in transcribing printed and handwritten historical texts stored as images. I put together a corpus of 400 pages from the Library of Congress. Current leader is GPT4.1-mini, which at 11 cents per 1K pages, is also the cheapest option. github.com/nealcaren/In...

"What Can You Do When You Have Zero Rewards During RL?" by Jatin Prakash* (NYU), Anirudh Buvanesh* (MILA) spiffy-airbus-472.notion.site/What-Can-You...

Post-training 101: A hitchhiker's guide into LLM post-training by Han Fang and Karthik Abinav Sankararaman tokens-for-thoughts.notion.site/post-trainin...

This is an extremely nice thread if you need any kind of pick-me up. (Read the quote posts too!)

Morning! Society for Hopeful Technologists newsletter #2 is arriving in inboxes now. A little bit on Musk and X, much more about future plans + organising buttondown.com/society-of-h... and great to have a group of fellow organisers coming together 💪

Getting ready to invest more time into the RLHF book to prepare for print edition. What do you wish was clearer or had more coverage in it?

Why do LLMs fail at long horizon tasks? Because errors in execution, not planning i.e. an error made early on conditions the LLM into a bad state later. Newer agentic LLMs deal with early errors better arxiv.org/abs/2509.09677

Comparing AI Labs in the United States and China, by Jingyuan Liu

I love this analytical take on video game categorization from Antoine Mayerowitz and Julie Belzanne: hushcrasher.substack.com/p/taxonomy-o... Instead of trying to vibe-intuit the definition of an "indie" game, the authors analyzed the data from the perspective of game size and credits length

Gemini Live is giving me the confidence to speak my second language The anxiety I experience when trying to speak my second language disappears when I use Gemini Live. #ai #gemini #geminiai

its absurd how much progress there’s been

This week's media circus following the Charlie Kirk assassination has making me think, yet again, of how we're in an abusive relationship with MAGA and the so-called "centrist" media are enablers (at best) As I wrote about almost a year ago

TIL that for some people with irreparably damaged corneas, you can pull one of their teeth, drill a hole in it, install an artificial lens, sew it into their cheek (so tissue attaches), then pull it out and stick it in their damaged eye, and ... it works fine!?! www.nbcboston.com/news/health/...

I don't like doing this. It feels fucked up to constantly point out the negative dimensions of someone who just died in a horrific, worrying way. But I would like the media to tell basic truths about this person and his impact on public life rather than resorting to euphemism and outright lies!

Suburban rainbow. #scape 📷 #photography

Pet parrots which typically live alone (whilst those in the wild live in large flocks) were given the technology to call each other. They would use it for up to three hours a day, and developed favourite friends 💔 on.ft.com/3K05vhS

🌱Good Morning Gardeners🌱 Friday #Fuchsia Berries & blooms to wish a peaceful & pleasant weekend in the garden … ☘️😊☘️

Dear friends from other countries: please join the Brazilians as we celebrate this groundbreaking decision.

Spoke with @swyx.io about the evolving role of the AI Engineer for @redmonk.com's MonkCast. We discuss the definition of an AI Engineer, differences between AI Engineers & traditional software devs, vibe coding, frontend observability, & the AI Engineer World's Fair. redmonk.com/blog/2025/07...

Crazy random line in the Cursor RL blog post saying they're collecting RL data from real users, updating the checkpoint every 90 to 120 minutes. This would've been unthinkable if you told me it 12 months ago. What a wild year of reasoning models. buff.ly/8LqreS2

Alibaba's Qwen3-Next-80B-A3B 🔹80B params, but only 3B activated per token → 10x cheaper training, 10x faster inference than Qwen3-32B.(esp. @ 32K+ context!) 🔹Hybrid Architecture: Gated DeltaNet + Gated Attention → best of speed & recall 🔹Ultra-sparse MoE: 512 experts, 10 routed + 1 shared

Latest open artifacts (#14): NVIDIA's rise, "Swiss & UAE DeepSeek," and a resurgence of open data While Qwen takes some rest, others continue to fuel the open model space. buff.ly/9MKUgYR 43 of the best models/datasets from 27 different organizations.

Q. Who aligns the aligners? A. alignmentalignment.ai Today I’m humbled to announce an epoch-defining event: the launch of the 𝗖𝗲𝗻𝘁𝗲𝗿 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 𝗼𝗳 𝗔𝗜 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 𝗖𝗲𝗻𝘁𝗲𝗿𝘀.

A California bill to regulate AI companion chatbots is nearing final approval. The legislation, SB 243, would require safety protocols, user alerts, and transparency from AI firms to protect minors and hold companies legally accountable. #MLSky

Would you like some good news? A new vaccine has just been approved for protecting koalas from chlamydia, which is one of their leading causes of death.

🌱 #Gardening #FlowerFeed