OpenAI introduces their new Agent with full computer use capabilities, Anthropic is now evaluating models with other models, what the data leak at Tea can and can't tell us about vibe coding security, and more reasoning time isn't always better

Further reading:
https://openai.com/index/introducing-chatgpt-agent/
https://alignment.anthropic.com/2025/automated-auditing/
https://www.nytimes.com/2025/07/26/us/tea-safety-dating-app-hack.html
https://safety-research.github.io/inverse-scaling-ttc/

Allocating enough time to find your passions, Bell Labs vs Frontier Labs, project based classes, and getting involved in communities

Perplexity releases their AI browser Comet, Grok 4 and xAI's safety alignment strategy, the David vs Goliath of language models, and the Muon optimizer challenging Adam for its title

Anthropic & OpenAI's new approach for announcing models, double vs single blind peer review, and distribution channels for researchers

Inception Lab's new diffusion architecture Mercury, Google's emissions up 51% since 2019, DeepFleet warehouse scheduling robots, and Microsoft's delayed AI chip