New paper from Microsoft on Self-Evolving Agent Skills
INNOVATION
-
Dan Shipper: Future of work inside Codex or Claude Code
By
–
My biggest takeaways from @danshipper
: 1. The future of work will happen inside Codex or Claude Code. Instead of putting AI into your SaaS tool, you’ll use your SaaS tools inside your favorite AI agents' in-app browser. Dan spends all his time in Codex now—writing documents, -
Tau Scaling Law offers new path beyond Moore’s Law limits
By
–
But Moore’s Law is approaching physical and economic limits. The industry now needs new ways to keep improving compute performance. That is why the Tau Scaling Law she presented matters. It reframes the question from: “How small can we make transistors?” To: “How much time
-
Chinese AI models are now leading and frontier
By
–
That's outdated Chinese models are already leading and frontier: Alibaba, Kimi, Xiaomi, Z AI
-
GIPO: Gaussian Importance Sampling for AI Learning Without Forgetting
By
–
What if your AI agent could learn faster from older data without forgetting or crashing? Researchers from Sany Group present GIPO (Gaussian Importance Sampling Policy Optimization). Instead of hard clipping, it uses a soft Gaussian trust weight to gently limit extreme updates
-
Figure’s 24/7 Humanoid Robot Production Stream Achievement
By
–
Figure just made millions of people watch robots move boxes for 9 straight days.😱
— Pascal Bornet (@pascal_bornet) 25 mai 2026
And honestly, that is a bigger moment than it sounds.
Nailed it.
The challenge was simple:
Put humanoid robots on a live 24/7 stream and let the world watch them work in a real production loop.… pic.twitter.com/MZuPD9ZZSmFigure just made millions of people watch robots move boxes for 9 straight days. And honestly, that is a bigger moment than it sounds. Nailed it. The challenge was simple: Put humanoid robots on a live 24/7 stream and let the world watch them work in a real production loop.
-
Grok 5 Model Ready: 1.5T Parameters, Public Release in Weeks
By
–
SPEACEXAI : The next Grok model is expected to be ready for public release in 2-3 weeks. > 1.5T V9-Medium base model in comparison to 0.5T v8-Small, used for Grok 4.3
> Cursor data being used for supplementary training Grok 5? -
0.5T Model Open-Source Release Announced
By
–
We will open source the 0.5T model towards the end of this year. It should still be quite useful.
-
CommaAI openpilot: affordable autonomous driving kit
By
–
CommaAI built an #autopilot you can add to a regular #car 🚗
— Amitav Bhattacharjee (@bamitav) 25 mai 2026
Their openpilot system costs around 1,000 dollars and enables semi #autonomous #driving. The kit includes 3 HDR #cameras, two for the road plus a separate night camera for the cabin to monitor the driver and help… pic.twitter.com/hs69EbWIdyCommaAI built an #autopilot you can add to a regular #car Their openpilot system costs around 1,000 dollars and enables semi #autonomous #driving. The kit includes 3 HDR #cameras, two for the road plus a separate night camera for the cabin to monitor the driver and help
-
Self-Play RL vs Static Bug Generation in SWE Training
By
–
The key distinction: Synthetic bug generation is static. Self-play SWE-RL is online. The bug generator and bug solver improve together, so the curriculum changes as the model changes. That is what makes this much more interesting than simply manufacturing more training data.