I missed @teorth.bsky.social ‘s public lecture on “Mathematics in the Age of AI” last night due to an obligation for the IMU. But I’m looking forward to watching the video: Terry’s posted his thoughtprovoking slides already at teorth.github.io/tao-web/slid... #ICM2026

On this day in 2009, Wii Sports Resort was released in North America.

All go to the same place; all come from 4-bit, and to 4-bit all return.

GPT-5.6 comes in three model tiers. → Sol handles long-horizon coding and agentic work that demands planning, tool use, and follow-through. → Terra balances performance and cost for everyday work. → Luna brings speed to well-defined, high-volume work. (1/2)

Model guidance | OpenAI API

Compare model features, migration guidance, and prompting best practices across OpenAI models.

developers.openai.com

Now that Fable 5 is ready to build (again), we've reset everyone's 5-hour and weekly rate limits.

GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We’re expanding preview access globally now.

Collective intelligence is the practical hedge against this concentration of power. Fugu simply routes around vendor restrictions by relying on an entirely swappable agent pool.

As the developers who programmed the original SNES Star Fox and created the Super FX chip, we'd like to take this opportunity to congratulate you on bringing our boys back. It's time to pass the baton and do a barrel roll! ⭐🦊

Star Fox 64 Gameplay

ALT: Star Fox 64 Gameplay

static.klipy.com

We’ve been passionately working on this for a while, and we’re so proud to finally share it with the world: Velan Studios is the developer of Star Fox™! Collaborating with Nintendo and bringing Fox McCloud and crew to Nintendo Switch 2 is truly an honor. #StarFox #NintendoSwitch2

#HPC must read, "FP8 is All You Need (Part 1): Debunking Hardware FP64 as the HPC Holy Grail" from Satoshi Matsuoka arxiv.org/pdf/2606.06510

Warm congratulations to Ilias, Gautam, Daniel, Jerry, Ankur, and Alistair!

Clément Canonne@ccanonne.github.io · 2mo ago

Huge congratulations to Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Ankur Moitra, and Alistair Stewart on being awarded the Gödel prize for their breakthrough work on algorithmic robustness! www.sigact.org/prizes/g%C3%...

The 2026 Gödel Prize is awarded to Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Ankur Moitra, and Alistair Stewart for Robust Estimators in High Dimensions without the Computational Intractability Paper: doi.org/10.1137/17M1... Citation: www.sigact.org/prize...

Inlining: the hard parts bernsteinbear.com/blog/inlinin...

"People submit too many papers to ML conferences" Meanwhile, at IEEE Transactions on Wireless Communications: "You may not submit more than 36 papers per year." Apparently this is a new policy. I would be super curious to see the stats on how many people are submitting more.

Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor. It’s happening faster than we thought, and the implications deserve greater attention.

When AI builds itself

Our progress toward recursive self-improvement, and its implications.

anthropic.com

Microsoft AI's report on MAI-1 training is full of interesting infrastructure anecdotes, esp around storage: - Async checkpoint from DRAM to object (no node-local SSD) - Restart is read once, then broadcast - Anyscale Ray for fast restart - Bitwise repro > MFU microsoft.ai/wp-content/u... #AI

I'm at the Claude w/ Code event in San Francisco, and I'll be live blogging the keynote here: simonwillison.net/2026/May/6/c...

The Most Underrated Layer Inside Every AI Model Virtually every AI model has normalization layers. BUT, what makes them so essential? 🤔 New video on learning the role of normalization in stabilizing training and alternatives like DyT and Derf. youtu.be/JHl_gwVoh-k

Added a 1500 word mini history to my book on the path to on-policy distillation being a core post-training optimization technique. Is a fun rapidly growing area now! rlhfbook.com/c/12-synthet...

Synthetic Data & CAI | RLHF Book by Nathan Lambert

The Reinforcement Learning from Human Feedback Book

rlhfbook.com

Live from Code with Claude: we're launching dreaming in Claude Managed Agents as a research preview. Outcomes, multiagent orchestration, and webhooks are now in public beta.

In this paper is presented the first comprehensive characterization of SPEC CPU2026 across nine platforms spanning recent Intel, AMD, Ampere, and Nvidia processors. arxiv.org/pdf/2605.03713

@leesharkey.bsky.social has been working on a new way to do interpretability: decompose a language model’s weights, not its activations. Their decomposition natively handles attention (!) and behaves less like a lookup table and more like a generalizing algorithm.

New Anthropic Fellows research: Model Spec Midtraining (MSM). Standard alignment methods train AIs on examples of desired behavior. But this can fail to generalize to new situations. MSM addresses this by first teaching AIs how we would like them to generalize and why.

This GCC patch reveals some #Hygon CPUIDs, ISA levels, and codenames: github.com/gcc-mirror/g... - 900F40 Dharma #AVX2 - 900F60 Shanghai #AVX2 - 900F70 Chengdu #AVX512

New for financial services: ready-to-run Claude agent templates for building pitches, conducting valuation reviews, closing the books at month-end, and more. Install them as plugins in Cowork and Claude Code, or use our cookbooks to run them in production as Managed Agents.

Fully-funded EPCC PhD: investigate the potential of novel accelerators. Run in collaboration with AWE, who will provide real world mini-applications and algorithms of genuine scientific interest. Applicants must be UK nationals. Apply by 28 May 2026. Read more at: edin.ac/4979QJO

railroads were expensive from finn on twitter

Hey Ford, GM, Stellantis, and maybe Tesla: Jensen says you’re losers.

Countdown Standard xkcd.com/3232/

A NASA technical paper on Gas-Core Nuclear Rockets shows their Isp scaling to 6500s if they can retain the uranium plasma at 2000 bar: ntrs.nasa.gov/citations/19... At the 12 GW scale, they'd reach 58 kW/kg. 7.5% of reactor output becomes waste heat at 1400K to be removed by radiators.