🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

the first fully automated AI firm with the ability to spawn millions of agent employees is being built as we speak

very impressive indeed. i had a script that i was pitching to TV studios a few years ago. we are very close to building content around our own ideas and not have gatekeepers. i would also like to to experiment with how to break down technical concepts into visual tools.

we are entering the age of curing diseases. it has started ladies and gentlemen. ai won't kill humanity. it will save it. go e/acc 🚀

haters will hate. whatever you take away from this and everyone will react differently to it - the fact that we have such a strong positive message coming would have been unfathomable even a few years ago. the culture has shifted to true e/acc mode what is 'truly' possible

jony ive will surely be using OpenAI's cutting edge AI to build AI design agents to help his team design the next big product. that itself is a multi-billion dollar idea.

the acceleration never stops. hurrah for all AI coders. we have a new model and new benchmark. we have come a long way from 50% on the SWE back in Dec

It would great to more of this flow into companies like Extropic

The AI datacenter regulatory, supply chain, funding, permits pipeline is highly complex and evolving. We will be launching a product very soon to uncover many of these fast-moving shifts in the AI infrastructure landscape across the full stack.

With Claude 4 Sonnet and Opus we are that much closer to building an agentic org. All my code that was stuck using inefficient methods has suddenly been unleashed to just work. I am still at the helm checking and adapting to what the code does for me, but building and running

Claude 4 Opus is really good. Can make major multi-level changes through your complex codebase. But its expensive. For the first time, one needs to really think twice whether to run a code prompt on Opus vs Sonnet. Where I was running sub $50 monthly API bills I am now

E/acc is how we get there

Claude Code is the real deal. Clean crisp and works wonders. No intermittent breaks, picks up where it falters, just works as you want it to. The command line interface just allows you to focus on the task at hand and not get distracted with too many windows.

This looks interesting, but the question is how much compute does this add to pre-training? 10x? 100x? The current paradigm of pre-training and then using after hours inference compute to scale RL seems more efficient.

This 👇

My notes on Anthropic's substantial essay about how they built their multi-agent research system, which has finally talked me around to taking multi-agent LLM prompt engineering seriously simonwillison.net/2025/Jun/14/mu…

If you are an AI builder who is pushing the boundaries of the LLMs and LRMs this talk about whether AGI will come in 2027 or beyond is moot. We have tools that can generate code back to front for fairly complicated software stacks. Just go and build. Ignore the chatter

Using AI as a 10x multiplier is only something people with agency and infinite creative possibilities think about. The rest will keep arguing for the opposite. What are you anon?