Top Story
A preview version is available now for testing.
DeepMind's Andreas Kirsch warns compute providers capture the returns.
Moonshot will split subscriptions into general and coding tiers
Critics say labs lobby agencies to restrict open-weight models.
He faced public pushback after analyzing Chinese AI model Kimi.
China's Kimi matches leading public models in agentic coding

Operators deploy temporary AI wrappers to capture brief arbitrage.
Sol performed better than Kimi K3 but cost much more.
The debate highlights recurring community speculation about proprietary model architectures
Tan highlights Markdown as a resilient, universal data format.
The solution replaces a 44-page proof from 1991.
They note proprietary labs rely on PyTorch and JAX.
The transition sparked a debate on SaaS market survival.
The service will split into standard and coding subscription tiers
Simo Ryu noted users can manually adjust ChatGPT's personality.
The agent optimized code during a 10-hour hyperparameter sweep.
Isolated single-agent overnight sessions bypassed compounding drift.
Another user joked about building a floating datacenter there.
Peter Yang added "vibe engineering" to the satirical roles.
He warns isolation from open communities threatens future funding.