Jon Conley (e/acc)

E / A C C - The United States of Acceleration

@soraofficialapp @ns123abc

@askalphaxiv 👀 @SakanaAILabs @hardmaru

With how promising Evolution Strategies is as an RL alternative, we just completed our own research to compare and evaluate it against GRPO! Our major finding lines up with the ES paper: Evolution Strategies can beat GRPO even when you have only a little fine-tuning data. However, GRPO becomes the better choice once data grows or you start from base models. Hopefully this research we've conducted can serve as an additional rule of thumb if you want to work with ES. Check out our blog below👇

@ns123abc Deploying local models that are on par with Opus 4.6 and GPT 5 at scale will soon be a valuable skill.

I thought it was just me…

@mil0theminer No Idia

Is anyone using Dia? Did anyone ever use Dia?

@productaizery @BlakeHer_on @karpathy @ryancarson @RhysSullivan Very good question.

@BlakeHer_on @karpathy @ryancarson @RhysSullivan How do you know it works, and by what measure?

Thank you Sarah, my pleasure to come on the pod! And happy to do some more Q&A in the replies.

@w643423152354 @mil0theminer It’s not even recurring anymore, it is now “Annualized Revenue”.

@mil0theminer got a feeling that their ARR is actually most definitely not 400m

So they’re just trying to pivot to what everyone else is doing so they can fill their next round

@2kySzn @RPPharmacist @TukiFromKL The dotcom crash happened first, before the internet and the web took off.

@RPPharmacist @TukiFromKL People said the same thing about the internet… then it quietly took over everything.

🚨 Let me tell you why this Goldman Sachs headline is the most dangerous one you'll read today.. Companies spent $450 billion on AI last year.. fired tens of thousands of people to "restructure around AI".. replaced entire departments with chatbots.. And Goldman Sachs just said it contributed basically zero to economic growth.. so where did the money go? > It went to Nvidia.. $130 billion in GPU sales.. Jensen is the only man on earth who got rich from AI that hasn't produced anything yet.. > It went to stock buybacks.. companies fired people, cut costs, reported "record profits" and bought back their own shares.. the money went UP not OUT.. Jesus! > It went to a bubble.. the same way crypto money went to Lamborghinis and not infrastructure.. AI money is going to valuations and not productivity.. here's the part that should terrify you.. They already fired the people.. Atlassian 1,600.. Meta 21,000.. Block 40%.. Amazon warehouses.. the jobs are already gone.. But the growth didn't come.. the productivity didn't come.. the revenue didn't come.. they burned the village to build a city that doesn't exist yet.. and Goldman Sachs just looked at the empty lot and said "there's nothing here"

@TheGeorgePu Cursor is closer to being WeWork at 50B. Can’t see how they can justify that valuation.

Replit just hit $9B. It doesn't make the AI. It doesn't own the models. It's a front door to someone else's house. A year ago I'd have said that's a dead business. Maybe it still is. The last time someone paid this much for a layer they didn't own. It was called WeWork.

@Teknium @championswimmer @steipete @OpenAI @modal Yes

@championswimmer @steipete @OpenAI Isn’t that just @modal persistent sandboxes

This might be what our boy @steipete will work on at @OpenAI "Stateful Runtime Environments" - or as I have called them "Remote Execution Environments" or as Cloudflare calls them Sandbox VMs. Tiny compute boxes for LLMs.

@broadfield_dev @hyhieu226 Hope not

If you are only using codex to write code, you are missing out a lot. Codex takes away a lot of my mental burden: - git, uv, ssh keys - excel functions - Python plots (yes looking at you pyplot, manim and streamlit you are next!)

@karpathy @ChristosTzamos Are Neural Turing Machines going to make a come back because of this?

@ChristosTzamos Wait this is so awesome!! Both 1) the C compiler to LLM weights and 2) the logarithmic complexity hard-max attention and its potential generalizations. Inspiring!

1/4 LLMs solve research grade math problems but struggle with basic calculations. We bridge this gap by turning them to computers. We built a computer INSIDE a transformer that can run programs for millions of steps in seconds solving even the hardest Sudokus with 100% accuracy

@ns123abc If Deepseek v4 releases between now and May, and it performs better on benchmarks than Opus 4.6 and Gemini, then would be bad news for Llama before it even releases. So far it is not looking good for Meta, but we’ll see what happens.

🚨BREAKING: META delays AI model “Avocado” after it underperforms Google, OpenAI, Anthropic, xAI >be zucc >spend $600B on data centers >$135B this year alone >hire Scale AI CEO for $14.3B >poach top researchers from OAI >create elite lab called meta superintelligence lab >name your model Avocado >still loses to Gemini 3.0 Avocado delayed from March to May Meta leadership discussing licensing Google Gemini to power Meta AI products while they figure it out Zucc…

@TheAhmadOsman Perplexity has a great product and design team. The only issue is their undisclosed influencer campaigns ruining this app.

the OG grifter AI company they’re so irrelevant that they’re trying to jump on this Mac mini trend like a cheap slop influencer this HAS NOTHING to do with local AI

@beffjezos Not sure about xAI. They would need to spend 30B of cash + stock to pay for that 30B valuation. Amazon might be the ones that are more likely to buy Perplexity.

xAI should acquire Perplexity They need the product design taste of the team Would give Grok a nice agentic frontend

@S954814431683 @im_roy_lee ?

@JonyConley @im_roy_lee Ride him harder why don't ya

ngl i was sweating for 2 mins

@flesheatingemu @XFreeze My previous tweet agrees with you. Anthropic knows who are the biggest users of their models and will clone the ones that are the biggest revenue drivers which is why it is a trap. This happened to Cursor with Claude Code.

@JonyConley @XFreeze meh, like i said, claude code is a bullshit trap for fools who would agree not to compete with a thing that does all their work for them

Jason Calacanis issues a brutal warning to every developer: “If I were any kind of developer, I would never work with Sam Altman and OpenAI” “This is a warning for anybody dumb enough to use Sam Altman’s OpenAI API” He says Altman is actively studying exactly how you use the API because he wants every single dollar of revenue from the entire ecosystem. No prisoners Sam comes from the Zuckerberg school of business: give people tools, study what they build, and like the Borg, steal every innovation they create Exactly like Bill Gates did at Microsoft - let people build Lotus 1-2-3 then release Excel, let them build WordPerfect and WordStar then release Microsoft Word

@flesheatingemu @XFreeze Cursor crawled so that Claude Code could sprint.

@XFreeze anthropic doing the same exact thing, every claude code user agrees to a noncompete, and they automatically opt you in to train claude on your codebase that means - when they take your business, you insta-lose in court - you are paying them to learn to take your business

@stuffybag @klos @meaning_matters 🤣

@klos @meaning_matters so made up bullshit

3 years ago, I hired a programmer to write a feature. He tried for almost 3 months. It cost me ~$10k, and the result was so buggy that I decided to turn it off. Today, I'm deploying the same feature written by Opus in about 1h. It's not binary, but many programer s are worse than AI in most tasks.

@Shilllin Undisclosed paid partnerships are so widespread on X. Not just crypto but also in AI. It’s even out in the open.

Anyone willing to post “paid partnership” I will support, very few integral people left in cryptotwitter Thank you to those who choose integrity