psa: your competitor's next product demo will look like a hollywood production. it'll cost them $200 and one afternoon to make. internally, we've been pushing video models to their absolute limits to see what breaks. here's our newest veo 3 experiment:

Get early access:

Introducing — the fastest way to build AI marketing agents that get users on autopilot. Watch me build a Reddit agent in 60 seconds:

Huge news: we've raised $1.2M to build the first AI marketer that acts like a human. Astral can browse the internet to find potential customers, then convert them while you sleep:

Join the waitlist:

Announcing Astral - an AI marketer that works 24/7 to grow your startup. Astral navigates websites, creates content, and runs marketing across socials. Watch Astral automate Reddit in real-time:

RT @fdotinc: F&#! doing groceries on vacation. Introducing - a supermarket inside your Airbnb.

agent hack night at github 🤝. demos + talks on: .- llm observability & evals by @arizeai .- multi-agent framework from @crewAIInc .- efficient vector search from daxe.- fast and cheap ai inference by @friendliai . ft. @zaw358 🫡

// Working on an API that lets AI agents get real-time feedback from a pool of 200k people. With a few API calls, an agent could: .- Test products, prototypes or mockups on real people.- Gather structured data on market pain points .- Generate feedback forms, and get answers to

Though this may be because we’re in the early stages of understanding. We might see major improvements with better scaffolding, tooling, and post-training, similar to what we saw with GPT models. (Image Source: Situational Awareness by Leopold Aschenbrenner)

Seems I was wrong on this…. o1 doesn’t perform much better on agentic tasks than other frontier models. Which is pretty unexpected given its supposed IQ of 120.

RT @MatthewBerman: Holy sh*t.

Am I missing something, or is the only limitation to achieving AGI better tooling? . If o1 "rivals the performance of top human experts in reasoning" - then it's already as intelligent as the smartest humans. Just give it tools to interact with the world, and you have AGI.

o1 might cause a paradigm shift for agents. today’s agents suck at many-step workflows - since one poor decision snowballs into a chain of errors. but o1 will lead to way fewer reasoning errors, meaning agents will crack processes they just couldn't before.