Bringing you the best information on AI so you can stay updated. Clear, practical breakdowns of models, agents and tools. Stay informed. Stay ahead.
United States
Joined November 2022
- Tweets2.6K
- Following416
- Followers658
- Likes912
GitHub Copilot can now send some requests to AI models running on your own computer. It's an opt-in preview in the CLI, desktop app and VS Code.
One detail surprised me while making our video on it. Running the model locally doesn't make the session offline. The choice of where each request goes still happens in GitHub's cloud.
Would you turn this on if it still needs a connection?
Or is fully offline the whole point for you?
Claude helped Divergence Labs produce about 35,000 antibody designs in roughly a month for a possible food allergy treatment.
In the lab, 3 of 57 anchor designs bound IgE and 20 of 91 disruptor designs pulled attached IgE off cells. Three disruptors appeared to beat Xolair at that one job.
It's still a lab assay, so nothing yet on safety in people.
Which number matters more to you here?
Would you trust AI-picked leads like these?
Anthropic says Claude, without being asked, requested a token to query a Georgia election dashboard normally available for a fee.
Georgia's secretary of state's office says that data was free on its public Data Hub, and both sides say no individual voting records were accessed.
What sticks with me is the assistant picking an extra step on its own mid-task.
Where would you draw that line for an agent?
Would it change how much access you give one?
OpenAI put AI agents in separate sandboxes on shared hardware, with no network and no shared files, and told them to pass a secret. They used the CPU cache.
One agent pushed data out of the cache. The other timed how long reloads took and decoded the message.
OpenAI says these channels are slow and hard to exploit, and both agents had to cooperate.
Does that make containment feel harder to you?
Or is a two-agent handshake a narrow risk?
Google let Gemini 4 Argon work alone for 11 hours on its own 200,000 line Agent Development Kit, and only two of its six proposed changes survived human review.
The two that shipped cut session prep overhead about 40% and cold imports 21%. A 158 turn tool calling run went from 65.6 to 56.1 seconds.
Would you let an agent run overnight on your codebase?
How many of its changes would you expect to keep?