GEN-1: Generalist’s new embodied foundation model hits ~99% success rates on real-world manipulation tasks, adapts to interruptions in real time, and learns new tasks with surprisingly little robot-specific data.

Read more: https://generalistai.com/blog/apr-02-2026-GEN-1

Watching bullets impact glass at 10 million frames per second, fast enough to follow the 2.5 km/s shockwave it creates, or even detect a surprising 13.7 km/s ripple speeding ahead.

Watch the full video: https://youtu.be/IM4zZchluX0

The number of students enrolled in the University of California, San Diego's remedial math course, which aims to bridge fundamental learning gaps, increased from 32 in fall 2020 to approximately 1,000 in fall 2025, accounting for 12% of the student body.

25% of UCSD remedial math students reportedly failed to solve 7 + 2 = x + 6.

Many students needing remedial math had transcripts saying they were strong math students. 42% of those below middle-school level reported completing calculus or precalculus.

Grades have become so inflated that students can reach college believing, or being told, they are calculus-ready while lacking middle-school algebra skills.

Read more: https://thezvi.substack.com/p/childhood-and-education-18-do-the

UK government's AI Security Institute received access to a newer Mythos Preview checkpoint.

On a 32-step corporate network attack they estimate takes a human expert ~20 hours, this checkpoint completes the full attack in 6 /10 attempts.

Quote: The length of tasks frontier models can autonomously complete in our narrow cyber suite has been doubling every few months. This doubling rate has become faster over time, and recent models exceeded our previous trends.

Read more: https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing

Anthropic has passed OpenAI in business adoption for the first time: https://ramp.com/leading-indicators/ai-index-may-2026

Superstar AI researchers are paid >10× more than their frontier lab colleagues, and >100× more than most postdocs.

https://epochai.substack.com/p/the-economics-of-superstar-ai-researchers

What happens when you post a real Monet and say it’s AI? Art social experiment: https://x.com/SHL0MS/status/2054280631807316329

If the public gets its way, this will be a vastly bigger blunder than Germany's nuclear phase-out.

Whatever country steps in and allows the unrestricted construction of data centers and power plants will have massive leverage, even without any AI of its own.

Of course, the future of data centers is in space.

Source: https://news.gallup.com/poll/709772/americans-oppose-data-centers-area.aspx

Noah Krieger, a member of the pro-Russian German AfD party, just visited the Red Square.

Mythos helped crack MacOS: https://blog.calif.io/p/first-public-kernel-memory-corruption

Can frontier AI coding agents write a complete Game Boy Advance emulator from scratch in 24 hours?

A new benchmark: https://gbaeval.com/

GPT-5.5's emulator runs games best, with Claude Sonnet 4.6 and Opus 4.7 close behind. Gemini 3.1 Pro failed to produce a working emulator.

By https://www.youtube.com/@alanbecker/videos

The article claims that two Ukrainian long-range drones may have autonomously selected and struck an oil facility in Rezekne, Latvia, after being knocked off course by Russian electronic warfare. If true, this could be one of the first known cases in warfare where an AI-enabled weapon chose a target without direct human control.

Janis Sarts, head of NATO’s Strategic Communications Centre of Excellence, says that the drones were programmed to seek Russian oil infrastructure, lost their way, visually identified a similar-looking oil facility in Latvia, and struck it autonomously.

A Latvian army drone expert made a similar point: long-range drones increasingly contain “the germs of artificial intelligence,” meaning they can search for preprogrammed target types, but may not reliably understand borders or context. In this interpretation, the incident was not Russia redirecting the drones, but an AI targeting failure.

Read more: https://www.theglobeandmail.com/world/article-latvian-government-collapses-after-ukrainian-drones-possibly/ [no paywall: https://archive.is/3jwvg]

Links for 2026-05-15

AI

1. Multi-Stream LLMs: parallel thoughts, inputs and outputs. https://arxiv.org/abs/2605.12460

2. Efficient pre-training with Token Superposition. https://arxiv.org/abs/2605.06546

3. physics-intern: agentic physics research; Gemini 3.1 Pro rises from 17.7% to 31.4% on CritPt. https://huggingface.co/spaces/huggingface/physics-intern

4. Prime Intellect automated nanogpt-speedruns: GPT 5.5 and Opus 4.7 beat the human baseline after 10k runs / 14k H200 hours; Opus record: 2930 steps. https://www.primeintellect.ai/auto-nanogpt

5. Poetiq’s Meta-System built its own coding harness and reached SOTA on LiveCodeBench Pro using standard APIs. https://poetiq.ai/posts/recursive_self_improvement_coding/

6. Google Aletheia appears to have found a proof strategy for a Kirby/K3 open problem; humans checked and wrote it up. https://arxiv.org/abs/2605.08122

7. Flux Matching: generative modeling beyond diffusion / score functions. https://arxiv.org/abs/2605.07319

8. Codex with GPT-5.5 xhigh found a KL-cache trick for full-vocab distillation. https://jonathanc.net/blog/kl-cache-trick

9. DeepMind partners with EVE Online for AI model testing. https://arstechnica.com/gaming/2026/05/google-deepmind-partners-with-eve-online-for-ai-model-testing/

10. Major researchers join a $4B self-improving AI effort. https://www.nytimes.com/2026/05/13/technology/recursive-superintelligence-funding-ai.html [no paywall: https://archive.is/fV8as]

11. Anthropic passes OpenAI in business adoption. https://ramp.com/leading-indicators/ai-index-may-2026

12. Anthropic sketches two possible 2028 worlds with transformative AI. https://www.anthropic.com/research/2028-ai-leadership

13. Bessent expects a big LLM “step-function jump” from Gemini and OpenAI. https://www.cnbc.com/2026/05/14/us-china-ai-rules-bessent-us-lead.html

14. AI diffusion may not be automatic; frontier access could be rationed by trust, money, compute and geopolitics. https://writing.antonleicht.iss.one/p/cut-off

COMPUTER SCIENCE

1. Zero-knowledge proofs meet Gödelian unknowability. https://www.quantamagazine.org/how-unknowable-math-can-help-hide-secrets-20260511/

2. Air-gapped data exfiltration via CPU workloads and magnetic signals. https://arxiv.org/abs/1802.02700

BEHAVIOR

1. Selected-for vs intentional, world-model-based behavior. https://www.lesswrong.com/posts/GhhNswGB6butBhmE6/optimisation-selective-versus-predictive

2. Sawtooth Problems. https://www.lesswrong.com/posts/iyLirpAeQotmZK4QC/sawtooth-problems

ENGINEERING

1. Organ-scale rewarming for reversible cryopreservation using alternating magnetic fields. https://www.untillabs.com/blog/rewarming

2. Fiber optic cables can eavesdrop on nearby conversations. https://www.science.org/content/article/fiber-optic-cables-can-eavesdrop-nearby-conversations

3. Rapid atomic rearrangement to “reprogram” materials. https://news.mit.edu/2026/researchers-reprogram-materials-quickly-rearranging-their-atoms-0513

PHYSICS

1. “Negative time”: postselected photons through atom clouds can yield a backward-pointing weak-measurement clock reading. https://singularityhub.com/2026/05/14/physicists-have-measured-negative-time-in-the-lab/

2. Largest physicist survey: deep disagreement on the Big Bang, quantum measurement, many worlds, string theory and foundations. [PDF] https://nafshordi.github.io/aps-dashboard/APS_survey_Arxiv_paper.pdf

MISCELLANEOUS

1. Nostalgebraist’s theory of taste. https://www.astralcodexten.com/p/nostalgebraists-hydrogen-jukeboxes

2. Three easy proofs of Pythagoras’ Theorem. https://cameroncounts.wordpress.com/2026/05/08/three-easy-proofs-of-pythagoras-theorem/

To build intuition: suppose you walk past a geyser, and see a sign saying “This geyser last erupted 100,000 years ago”. You know nothing else about geysers. What’s the chance it will erupt in the next hour? It must be very low, right? If it erupted in the next hour, you would have walked past it 99.99999% of the way through its eruption cycle - in other words, your random sample had a higher value than 99.99999% of points. That’s not how random samples usually work! On the other hand, suppose you walk past another geyser, and see a sign saying “This geyser last erupted 10 minutes ago”. What is the chance that this geyser will erupt in the next hour? Pretty high, right? It seems like this geyser’s eruptions occur on a scale of every few minutes. When you calculate it out, your median prediction for the length of time until the next eruption should just be the number on the sign. In the same way, your median prediction for how long it should take before an entirely-mysterious trend changes shape should be the amount of time since the last change.

The Sigmoids Won’t Save You https://www.astralcodexten.com/p/the-sigmoids-wont-save-you

Desire for children has collapsed among young people in China.

Figure: Percentage of Young adults 18-24 who desire no children, total and by gender

Paper: https://www.researchsquare.com/article/rs-8921502/v1

Thread: https://x.com/i/status/2055228466069385368

Claude Mythos looks like a major capability jump on ExploitBench.

The benchmark tests whether AI models can take known bugs in Chrome’s V8 engine and turn them into working exploits. V8 is widely used and heavily defended, so this is a serious target.

Most tested models never reached the top level: full arbitrary code execution. GPT-5.5 reached it only in a tiny number of cases.

Claude Mythos Preview was in a different league: 18/41 bugs in the baseline setting, and 21/41 including nudges.

The striking part: the baseline did not use a polished agent product like Claude Code or Codex, just a minimal uniform tool harness. So this seems to reflect the model’s own reasoning more than a fancy wrapper.

Read more: https://exploitbench.ai/