Mon, Feb 9 at 6:00 AM
GPT-5 is running an automated lab for Ginkgo Bioworks 🧬
wait, what?

OpenAI and Ginkgo Bioworks gave GPT-5 control of a fully automated cloud lab and told it to optimize protein production. Over six months, the AI designed and analyzed 36,000+ experiments across six iterative cycles, processing 580 384-well plates with robotic execution. Humans handled the physical reagents. GPT-5 handled the thinking.

We connected our autonomous lab to OpenAI's GPT-5 and let it run 36,000 experiments....
@Ginkgo · x.com
...so did it actually work?

Yup. It cut the cost of producing superfolder green fluorescent protein by 40%, from $698/gram down to $422/gram. Ginkgo is already commercializing the optimized reaction mix. Stock jumped 6% on the news. First time a frontier LLM has driven large-scale experimentation like this. The "AI can only summarize text" era is officially over.

Did you expect AI advancements to move this fast?

Login or Subscribe to participate

Chrome just became an autonomous agent 🌐
🤯 what's Google doing now?

Gemini 3 Auto Browse lets Chrome navigate websites and complete multi-step tasks for you. Assign it a job, a new tab opens with a side panel tracking every step as the AI clicks, scrolls, and fills out forms. It can research travel across multiple dates, manage subscriptions, even handle tax documents. Rolling out now to AI Pro and Ultra subscribers in the US.

that sounds... risky?
Gemini in Chrome: Help right where you need it
youtu.be

Yeah, security researchers are already flagging prompt injection risks, where malicious sites could trick the AI into doing things you didn't ask for. Auto Browse can access Google Password Manager with permission, so the stakes are real. Google's adding safeguards: manual confirmation for purchases and posts, labeled browser history for automated sessions, and full activity logs you can review or delete. The browser isn't just a window anymore, it's an operator.

With auto browse in Gemini in Chrome (now available for Google AI Pro and Ultra subscribers in the US), you can delegate routine tasks...
@googlechrome · x.com
weekly scoop 🍦

📈 Meta aggressively scales AI infrastructure for 2026

🛡️ OpenAI forms integrity team for ChatGPT advertisements

🧠 Anthropic unveils Claude Opus 4.6 with adaptive thinking

Introducing Claude Opus 4.6
youtu.be

💻 OpenAI launches GPT 5.3 Codex for agentic software development

🎬 Disney Plus integrates Sora for user generated shorts

💰 ElevenLabs reaches 11 billion dollar valuation in funding

👀 weekly challenge: can you spot real vs AI pictures
what's the challenge?

Challenge: This week’s challenge is simple: Can you spot the real picture vs the AI generated picture?

Which of these pictures from the 1987 stock market crash (Black Monday) is real?
[scroll to the bottom of the newsletter for the answer]

Pic #1:

Pic #2:

😉 Not as easy as you thought, huh?

Will you trust an AI agent to browse the web and book your travel autonomously? And, does GPT 5 running a lab feel like the start of a new scientific revolution? We want to hear your thoughts!

Answer to Challenge: Picture #1 is the real deal. Picture #2 was created with the help of Nano Banana Pro.

Zoe from Overclocked 

Keep Reading