if your goal is to beat Hyperliquid, go back to the drawing board your TAM needs to be bigger if you think about it, everything is a market crypto market commodities market forex market let's push it further job market prediction market insurance market carbon credit market influencer market advertising market whatever you think your TAM is, it's 100x bigger

Do you remember when you joined X? I do! #MyXAnniversary

ai agents got into anthropic's sales, hr and it systems in under 12 hours it was an authorized bug bounty test by a startup called cathedral, and it went in through third-party business software. anthropic says user chats, source code and model weights weren't touched, but slack messages, meeting transcripts and salary data were reachable if the ai company can get hit through its vendors, so can everyone else

Is a job a way to get things made, or just a social construct? My take on work for the next generation

gbulk!

STONKINDEXCOMMODITIESEQUITIES

> run a campaign >give $50 for people who tradedonce a day for 5 days >mid campaign change the term and condition the cabal is strong in here

scientists think your brain might talk more like chatgpt than we thought the old idea was that we build sentences like ikea furniture, following a hidden blueprint of grammar rules in our heads this study says it's more like autocomplete. we stack familiar chunks like "at the end of the day" and guess what comes next so maybe "it's just predicting the next word" isn't the insult people think it is

when you're building a robot but accidentally put evil qwen instead of normal qwen as it brain

blink once and it's suddenly halfway to 2B volume.

found a neuron on qwen 3.5-4b that remind me of her.

Gbulk, lets get it.

a rule computer scientists have relied on for over 20 years just got disproved. and it wasn't a human who did it, it was Openai's AI let me explain why people are going crazy over this. imagine a giant padlock with thousands of switches. each switch is on or off, and only certain combinations open it. there's no hint, so the only way is trying combinations one by one every switch you add doubles the number of combinations. 10 switches is about a thousand tries. 100 switches is more than every computer on earth could try before the sun burns out this kind of puzzle shows up everywhere. scheduling flights, designing chips, planning deliveries. a lot of hard real world problems are secretly this padlock in the late 90s, three researchers made a bet called the exponential time hypothesis. basically: there's no smart shortcut. you will always be stuck doing something close to trying every combination nobody could prove it, but almost everyone believed it. people built whole careers and tons of research on top of it, using it as the floor for "this is as fast as it can ever get" now Openai's system says the bet was wrong. there is a shortcut and this isn't a chatbot just saying something confident. the proof was checked by lean, a program that verifies math line by line. the same researchers who proposed the hypothesis are now working with ryan williams to double check it, and the whole thing is public for anyone to look at to be clear, your passwords don't break tomorrow and nobody has a faster app to use today. turning this into real world speedups still takes a lot of work but think about what happened. a "this is impossible" that experts trusted for decades just got a crack in it, and a machine found the crack if AI can do this to a 20 year old assumption, how many other things we call impossible are just things nobody has found the shortcut for yet

anthropic just filed to go public what it is: an S-1 is the paperwork you send the SEC before an IPO. it's not live yet, the shares can't be sold until the SEC signs off what we know: NASDAQ, ticker $ANT goldman sachs + morgan stanley leading share count and price range not set yet why it matters: until now the only way to get exposure to frontier ai labs was big tech stocks or private rounds. this would make one of the top labs something anyone can buy it also means anthropic has to open its books. revenue, compute spend, losses, all public. we're about to see what it actually costs to build these models and once it trades, every claude launch or safety headline becomes a stock event would you buy $ANT on day one or wait?

translation: lenders stopped funding companies whose whole business is renting out gpus spacex can borrow $40B because starlink and launches pay the bills. a neocloud with no revenue outside ai data centers can't ai money isn't drying up, it's getting picky. real revenue wins do you think the 200+ neoclouds really shrink to a handful?

gBulk. Bagworking at a medieval theme festival.

The discourse against using AI for programming is dumb and backward. You could argue AI was bad at coding two years ago. But with models like Opus 5.5 and increasingly capable open source LLMs, calling everything AI generated “slop” feels disingenuous. Whether it’s slop depends on who’s using it. A senior programmer with AI can build far better than someone who doesn’t even understand HTML. It’s like painting. A skilled painter can do far more with a brush than someone who’s never picked one up. At the end of the day, AI is a tool. And knowing how to use it is a skill.