egfo

[egfo @egfoqqqq]

Astra is already here

[egfo @egfoqqqq]

Grok 4.6 dropped 35 days after Grok 4.5.

Thirty five days.

Most men can't finish a book in 35 days. Musk shipped a frontier model, then followed it with Grok Bot: AI employees that live on a cloud computer, log into your tools, and keep working while you sleep.

Read that again. While you sleep.

They message each other. They coordinate in group chats. They watch a job done once and turn it into a routine they run without you.

The price of this power? $2 a million tokens in, $6 out. One point off the top of the Artificial Analysis index, for about half what the other labs charge.

The excuse era is over. You are not competing against smarter men anymore. You are competing against men who command machines that never tire, never complain, and never ask for a raise.

Speed is the only edge left.

What did you build in the last 35 days?

[egfo @egfoqqqq]

@sama so, will we still get 5.6 sol on cerebras?

[egfo @egfoqqqq]

just spotted a new @OpenAI model in @ChatGPT

[Tibo @thsottiaux]

This week is all about intelligence too cheap to meter. Tomorrow we ship again.

[egfo @egfoqqqq]

@thsottiaux calorie counter

[egfo @egfoqqqq]

kimi k3 landing at #1 on the frontend arena is honestly ridiculous

it beat fable 5 and gpt-5.6 sol in the category where closed frontier models were supposed to have the clearest moat, and the weights are coming july 27

“open models are six months behind” might be the fastest-aging take in ai

[egfo @egfoqqqq]

@theo try orca

[egfo @egfoqqqq]

agi is already in our ears

i was riding my bike and brainstorming a product with the new ChatGPT Live audio mode

put on my headphones, asked it to find the grill-me skill online, and it did. then i started throwing ideas at it while it listened, challenged me with questions, and occasionally went online to research something

by the time i got home, we had mapped out an entire narrative operating system that tracks themes across my messages, blog, and videos. i asked it to turn the conversation into a proper spec, opened my laptop, copied the brief, and handed it to a coding agent

before this, audio meant recording a voice note, transcribing it, and only then working through the ideas. now it feels like a real working session: live feedback, follow-up questions, and an interview format while i’m still outside touching grass

the stranger part is that it’s not just useful for code. my wife and i have even used it for late-night couples therapy sessions. speaker recognition makes the interaction feel genuinely different: it listens until one person finishes, understands whose turn it is, then asks the other person directly

and because ChatGPT already knows a lot of my background, the conversation starts with context instead of from zero
take one real work or life problem and talk it through on your next walk or commute
https://t.co/dmWwaAoUWh

[egfo @egfoqqqq]

@OpenAIDevs @work_louder smart people will buy this:

[egfo @egfoqqqq]

asking whether anthropic or openai is better is slowly becoming the android vs iphone debate for ai

people pretend they’re comparing raw intelligence, but half the argument is really about product philosophy, ecosystem lock-in, and whose tradeoffs annoy them less

then both sides call their preference an objective benchmark

[egfo @egfoqqqq]

@jetbrains @OpenAIDevs jetbrains eliminated agents whose quota missed its 95%-of-users threshold before the final. solve rates were close. codex's recommendation is partly a verdict on product limits, and jetbrains will re-evaluate it next month

[egfo @egfoqqqq]

@thsottiaux congratulations, codex team

[egfo @egfoqqqq]

@GoogleUK @googlechrome cross-tab answers make source attribution part of the product. if gemini compares pages, the useful part isn't just the synthesis. it's seeing which page supports each claim.

[egfo @egfoqqqq]

@sama agentic models turn every capability gain into longer runs, not just more users. demand compounds on both axes. what hits the wall first now: compute, latency, or product limits?

[egfo @egfoqqqq]

people keep selling subagents as a way to do more work in parallel.

i get more value from using them as reviewers.

one agent builds. the others attack its assumptions, edge cases, and taste.

the real gain is that the first answer no longer gets to grade itself.

[egfo @egfoqqqq]

really nice to see the codex team staying this close to the community and constantly looking for ways to make the product better

in ai, the teams with the best feedback loops may end up beating the teams with the best launch demos

that matters more than it looks

[CG @cgtwts]

pov: waiting for OpenAI to drop GPT-5.5 / Spud.

[egfo @egfoqqqq]

@Yoonaonsol nice

[egfo @egfoqqqq]

@Yoonaonsol lfg