juust

This week's Brainstorm: Our breakdown of the Anthropic and SpaceX deal and the broader AI infrastructure buildout 00:00 — Anthropic ran out of compute. Frank breaks down how they solved it with SpaceX 02:41 — Why leasing Colossus One makes sense for SpaceX heading into its IPO 07:24 — The economics of a gigawatt: $60B to build, $15B/year in infrastructure revenue, $30B/year if you're the model company 11:03 — Revenue per watt is going up. How NVIDIA's chip gains flow through the entire AI stack 14:44 — Space vs terrestrial data centers: can you launch a gigawatt for under $20B? 20:45 — Earth is getting scarce. Why that changes the math on space-based compute entirely 24:01 — When does space compute go from concept to reality? Brett's take Follow @wintonARK, @downingARK, @skorusARK.

Who amongst the X #AI 200 are talking about an #LLM over the past 48hrs ? A new tool of my AI Signals app, semantic search. Hook into the discussion on your favorite AI subjects...

We said in college, 'if as financial you still have to work past 45, you didn't get it'

@brockpierson Now I am having an anxiety attack

Are you old enough to have used a dial up modem yourself to get onto the internet?

Okay, I am game. I can ask any LLM about the Pure Language (I am fluent in it, sometimes, when I am relaxed) and LLM's know little or nothing about it. Only very few people know anything about the Pure Language. But it is in the books, the Gospel is part a demonstration of it, it contains numerous examples. It's one of these '2%' things. If your Claude Code can tell me in three hours, before midnight overhere, what the Pure Language is in 200 words, and give me three examples with an explanation, I'll stand corrected. I'll ditch OpenAI and become a big fan of Claude. I would *love* for any LLM to produce a coherent article on the Pure Language. Serious. Make my day ! Make my life !

@juust @haider1 Use Claude code and you know what you said is outdate by at lease a year they get that 2% within 2-3 hour human need 2-3week to do the same

Yann LeCun says LLMs are strongest in domains where language itself is the substrate of reasoning, like math and code They can solve problems, prove theorems, and write programs — but they are not creative mathematicians, software architects, or computer scientists "their role is to help humans build"

No good, no food

I am having fun... work should be fun like this, always, when you can just make whatever you want #AI imagine it, prompt it, run it

#Ai #Dev Forum Today ▸ Demi Moore sparks ‘fascist propaganda’ backlash at Cannes a… ▸ 2028: Two scenarios for global AI leadership ▸ Detroit automakers have cut more than 20,000 U.S. salaried… 👉 Read: aidevforum.com/blog/ai-news-o… #AI #AIDev

@shafu0x @merit_systems We also do that as deckhands. I ask around if a skipper is okay, who he or she is, and go talk to them : 'hi... bla bla bla...'. If we like one another, we sail a round. That's how it works. The 'mobile first' generation have to learn the hard way

This is actually how you get a job! I want to work at @merit_systems. Find the CEO, their verified email, what they seem interested in lately, and draft a short personalized email that I could actually send.

@elonmusk ..and I am the AI Mind Reader ! Investor intelligence from the top 200 AI accounts on Twitter, from research to development, with VC's, let's see the mind at work...

𝕏 is the mind of humanity

@vibin_live Nice idea, a quick newsflash page for devs !

Where do you stay up to date on AI news?

@rand_longevity You seldom die of old age, usually you get cancer or a heart attack

I genuinely believe I will never die of old age and I know how insane that sounds but the data is the data

@van00sa I am happy with just a little, to buy birthday presents If Claus pays my presents, I'll settle for nothing

You will own nothing and you will be happy

@astnkennedy Hey Austin, I am an old coding fossil ! We're in a hype fase, everything is 'double speed, double shifts'. It'll settle down. Learn to work with the new tools. We went from gorilla.bas on DOS6 to Windows with Visual Studio in W95. Wow... but we all learned (except Linus).

I'm 22 and it feels like I have no time. Yesterday I was talking with my good friend that's 19, and it seems like we only have a "7 year window" to do big things. This could be young immaturity, but I think there's a degree of truth in this. Three reasons why: (1) The models have democratized information. Some people are using ChatGPT, Claude, etc. to learn at a rate that wasn't possible before. I see those models as a PhD in your pocket (given many of them are trained off of PhDs). This along with YouTube kinda cover it all. (2) Social media has democratized distribution. If you have a phone you can record content and click post in 30 seconds. Allowing human relationships to scale in a way not previously done before. Instagram, TikTok, and LinkedIn have been around forever. But posting content is only just now becoming normalized. Most of my self-employed friends got there through content. (3) Coding agents have democratized building. Via a $20 Claude Code or Lovable subscription, you can ship from your laptop in hours. Solo founders are doing what used to require a team. Right now, I feel like everybody really can't keep up with everything going on, and there's a lot of fog in the air. But eventually it feels like the air will clear and the window will close. Again, idk if this is just a 22 year old not knowing the way the world works, but would def be curious people's opinions that lived through the dot-com era or the early App Store days.

@briansolis @DarioAmodei @sama @ylecun I agree that causality is one main issue. I solved it with a cloud guardrail of ISO9001/double entry accounting (causality is the basis of accounting, and ISO9001 is a good framework for budgets and mandates, the lock on the wallet).

At Davos, in the same room where @DarioAmodei was claiming 50% of white collar jobs will disappear in five years and @sama was talking about slipping past AGI into superintelligence, @ylecun told the entire agentic AI industry that they're building castles on sand. His point was that popular large language models predict words well but can't foresee real-world action outcomes. He argues that without internal world models simulating physics and cause-and-effect, these AI agents face cascading failures from mistakes or hallucinations. His view reignites divides among builders racing to deploy LLM agents for tasks like coding and research, while skeptics like Gary Marcus and Demis Hassabis call for brain-inspired alternatives amid billions in investments. LeCun left Meta at the end of 2025 to lead Paris-based @amilabs , a new company focused on world models. @drfeifei and @theworldlabs are also focused on World models. They are initially ideal for applications requiring physical understanding, spatial reasoning, and safe, reliable action planning beyond language-based AI. Key focus areas include robotics, industrial automation, healthcare, and 3D simulation, where AI must understand and act upon physical reality rather than just processing text

Second edition of the Daily X #AI Signals ! Works so far. aidevforum.com/blog/x-signals… It's already an old format. 280 chars offers little context. This morning, I added an auto-building semantic 'world' system. Now I'm making it self-learning with a cron job, some probability and a scraper. That should improve the quality a tat, that, and a better model for the report itself. #buildinpublic

@elonmusk Is that Chinese red wine ?

@levie I built my own mini-SAP for Claws. Without LLM's, just code. I didn't want it to be subject to the LLM armsrace. It aims to provide a causality framework to eradicate hallucinations and make them usable for business.

He just spent a year building scaffolding for his agent harness. Now release a new model update that makes all of it obsolete.

@BoringBiz_ You got a point there :)

We told an entire generation to learn how to code and then spent trillion of dollars on data centers to never have to hire them again Let that sink in