Someone needs to stop this. A dev is running full macOS inside a terminal window. Settings, wallpapers, Finder, all clickable. Why does this even work?
35
Someone put an entire macOS inside your terminal.
Your AI agent gets its own Mac to click around in, and you keep yours. You watch it work live, right next to your terminal.
Free and open source. Would you let an agent drive a whole Mac?
1
1
51
gpt-6.2-astra appeared in Codex usage breakdowns over the weekend.
Not in the API and not announced, but raw responses already show it in the model field. Looks like a quiet staged rollout.
1
129
A model OpenAI hasn't announced is showing up in Codex usage logs: gpt-6.2-astra.
One user's week: 1.6%
Another's: 19.8%
One day: 37.7%
The picker still says GPT-6 Astra. Some of your Codex requests may already be hitting the new model.
Careful though: half the "GPT-6.2 outputs" going viral today are from the old model.
93
Right now you can't confirm which model actually answered you. FlashProve fixes that with a zk proof.
The catch: proving costs 70–95× the GPU time of normal serving, and it only covers greedy decoding for now.
64
Anthropic open-sourced FlashProve.
It attaches a proof to a model's answer, so you can check which model actually wrote it, without seeing the weights.
1 H100 → 110 tok/s of proofs
verify on a CPU in ~17s
tested on a 400B MoE
Swap in a cheaper model and the proof fails.
20
Agents are now a scaling problem for Git itself.
OpenAI's Pando: ~80% lower push latency, ~1 PB of Git data a day. Engineers and agents use Git the same way they did before.
54
OpenAI built Pando, a new Git backend for its monorepo.
Push latency is down ~80%, and it serves nearly a petabyte of Git data a day.
The reason: agents push so much code that Git itself had to scale.
17
Grok Bot usage got reset back to normal today, one day after users hit weekly caps and Musk called serving costs unsustainable. Did your limit come back?
1
134
Fable 5.5 leak status: Azure listing pulled, codename spotted and removed, partners reportedly testing. Today people say a newer Fable is already answering in Claude. Anyone else seeing it?
94
Anthropic hasn't announced Fable 5.5. But it's showing up everywhere:
> Azure listing marked "generally available", then pulled
> codename "onigiri" spotted, then removed
> partners reportedly testing it
> now a "new Fable" in Claude answering 2026 questions
Fable 5.5 its over for video and image models 🤯
THIS IS PURE CODE
one shotted this video holy shit , how this is more beautiful than ai video models
Two ways to check this yourself:
Ask: “Where did @karpathy go in 2026?” with web search and memory turned off. Only the new Fable knows the answer. Then give it the prompt.
Run the same prompt I used for the test (it’s in the screenshot) and see for yourself.
Don’t rely on Grok for basic reasoning like this. Grok doesn’t have up-to-date knowledge of niche stuff. Either join the community or run the test yourself.
Also, a knowledge cutoff doesn’t mean a model knows everything before that date. Sometimes a model doesn’t know about an event even if it happened before its cutoff. This test shows that too.
Run the same prompt on regular Opus 5.5 and the quality drops. On Fable 5.1 it’s worse.
89









