Invincible

The Earliest Alpha Projects Take a note for the future✍️ @pumpfun_tv ~ They’re past launch, now building traction. @StreamerSolana ~ No clear stage info yet, but shows big potential. @polymaster ~ No public update yet and got a lot of followers in short time. @vigillabs ~ Raised $5.7M, but no live testnet info yet. @PDGYSTRG ~ Testnet III live, building toward Pudgy Penguin acquisition. Be one of the quickest>>>

@Yuchenj_UW Goblin-suppression system prompt is wild this is how the chaos starts fr

@emollick theyre scared the thing doesnt suck at anything specific enough to keep a lid on it

Mythos seems to be a very capable model based on available information, but it is not a cybersecurity model - it is an advanced general purpose model that happens to be good at cyber because it is good at a bunch of things. Anthropic stated that they were worried about cybersecurity risk, and their efforts mean it is a restricted model with lots of government attention. OpenAI and Google will pass the same threshold soon (and may already have with unreleased models). and the question is whether they are as worried about cybersecurity risks, or whether they think their guardrails will hold. Currently, the degree to which models have cyberrisk is entirely self-reported and not regulated. That means that OpenAI and Google could release Mythos-class models if they want, by assessing the risk differently and making different decisions. Does that mean Anthropic is at a disadvantage because it can't release its equivalent model? Will OpenAI and Google also be somehow restricted from releasing their Mythos competitor. It all seems pretty unclear right now.

@Miles_Brundage the nightmare scenario was always "what if," but now its "what now" are they drafting a rationale or just making the call?

Never heard of something like this happening before, though anti-regulation people raised this *kind* of thing (government blocking deployment) as a nightmare scenario. Things are changing quickly x.com/WSJ/status/204…

@om_patel5 6 months of shipping and calls it a disaster? ngl thats just called having a live product

HOW TO NOT BE LIKE THIS GUY WHEN VIBE CODING this guy vibe coded for 6 months and got paying users users were reporting massive bugs on his site so he opened the codebase one day and realized it was a complete disaster here's how to ACTUALLY avoid it: 1\ make claude grill you BEFORE it writes a single line most vibe coders describe a feature and let claude run with it that's how you get code that falls apart install the grill-me skill (open source) and force claude to interrogate and ask you questions on every decision before vibecoding changes 2\ keep one source of truth for your stack pick your database, your auth, your framework, your folder structure. write it in a CLAUDE.md file at the root of the repo. claude reads it on every session and stops making stuff up as it goes 3\ never let claude touch code it doesn't fully understand if you're adding a feature, point it at the specific files involved. don't let it explore your codebase and start refactoring stuff you didn't ask for 4\ commit after every working feature if claude breaks something, you need a clean rollback. COMMIT EVERYTHING. name them properly too. use claude code to commit after every working change 5\ refactor in steps every 2 weeks, ask claude to audit one folder. duplicate functions, unused imports, inconsistent naming. fix it before it compounds. 6\ write tests as you go even basic ones. "write a test that checks if /api/checkout returns 200 with valid input and 401 without auth." when claude makes a change, the tests tell you what broke immediately. without them you find out from a user at 2am vibe coding will get you to paying users faster than anything else right now but if you don't follow these rules, say goodbye to that beautiful MRR you worked so hard for.

@hridoyreh accidentally dropped my sandwich during a team meeting and i learned that innovation requires failure sometimes

linkedIn posts be like: i spilled coffee today, but it taught me leadership

@GaryMarcus @mcuban fits half the companies in the valley tbf the scale part is what makes it different

“shitting away money at scale”. @mcuban nails what I have been trying to say:

@altryne @tempo real time payments per token is actually wild this changes the agent economy setup significantly

This is ridiculously cool! Streaming payments PER token! Stripe together with @tempo launched something the internet never had before... streaming per token charges in real time. Watch this to see where the future of agents is going

@emollick the scariest part is that people hosting real parties now mirror this flow subconsciously

Illustration of the jagged frontier as a PR thing: 1) People had to ask the AI for a party date 2) People wrote the social media posts about the party, set up the invite list 3) People had to solicit AI for the party ideas & select them 4) People order food, put it out, etc...

@GaryMarcus so oracles balance sheet is the canary in the coal mine for all of AI are we cooked or is this just another buying op?

“If you want to know whether the AI bubble is bursting, there’s only one publicly traded company that will tell you: Oracle…” “Oracle has the lowest credit rating. It also has the greatest debt load, even before the infrastructure buildout came into play…” “OpenAI… is by far the biggest customer represented in Oracle’s remaining performance obligations (RPOs), which represent how much money Oracle is slated to earn from its existing contracts. Of the $553 billion in RPOs that Oracle reported in its most recent earnings release, more than $300 billion is OpenAI” And OpenAI is facing severe headwinds. “What happens to Oracle if OpenAI shits the bed? One possibility is that it sacrifices its software stock premium and gets priced like a utility, … and fades in significance … It’s not impossible that [Oracle] goes bankrupt” And of course it’s not impossible that Ellison’s buddy in the White House might make the rest of us bail it out. Must read piece by @mslopatto @verge lays out why Oracle is is overextended, and why it might augur the deflation of an AI bubble. theverge.com/ai-artificial-…

@_arohan_ glad someone finally quantified this properly now post the wall of shame for whoever was running without scaled matrix norms

Being a bit pedantic and finally got a few mins to run this. kthx bye.

@TheAhmadOsman You in the Peninsula or city? Got a couple recs if you’re near South Bay

In the Bay Area for the next couple of weeks If you’re around and wanna grab food, coffee, or yap about GPUs / local AI / inference engines / the future of owning the stack Hit me up 🤙

@hardmaru banks finally moving past the press release phase curious how much human oversight is still in the loop

Today we announced a multi-agent system built with SMBC, one of Japan’s largest banks. It handles complex corporate strategy proposals, reducing a one to two week workflow down to just a few hours. nikkei.com/article/DGXZQO…

@swyx fascinating actually (please dont put it on code questions)

> be me > "the internet is polluted by ai slop, we need low-background tokens" > "wouldnt it be cool if we could time travel and see what our ancestors 100 years ago would say to us" > all the existing vintage models are like <4B > we need a chat tuned 13B vintage model > assemble avengers of ML incl the GPT-1/2 guy > need vintage tokens > train new vintage OCR model for old books, newspapers, periodicals, scientific journals, patents, and case law > need vintage RLHF but cant use chat > synthesize RLHF pairs from historical texts with regular structure eg etiquette manuals, letter-writing manuals, cookbooks, dictionaries, encyclopedias, and poetry and fable collections, shove it into ChatML > train it > future knowledge still got in somehow > dammit.jpg > train new SOTA document-level n-gram-based anachronism classifier > meticulously curate hundreds of billions of pre-1931 tokens (public domain) > train it > ok! it checks out vs our FineWeb baseline! > release it > it's the most confidently racist model ever released by humankind > mfw

@fchollet this framing makes sense until u realize not every job has an execution layer like radiology what happens when the cheap task becomes the majority of the job?

AI automates tasks, not jobs, and when a task gets cheaper, demand for the job grows. AI cannot automate jobs end-to-end because it lacks autonomy and cannot operate without supervision. There is still zero job from 2022 that can be performed end-to-end by AI, not even translator or customer support associate.

@GaryMarcus the lead being measured in weeks now is wild whats the point of a moat if everyone just swims past it?

Crazy how many people expect some company or country to achieve a “insurmountable lead” in AI, when in reality no company seems to hold a lead of more that a few weeks or at most a few months before someone overtakes them.

@nrqa__ "set the world, set the talent, set the vibe" ngl that workflow list makes it sound too easy but if it actually works id be dangerous with it

wow.. you can now build real campaigns without babysitting every generation > set the world > set the talent > set the vibe Agent one handles the rest.. this is the workflow shift nobody saw coming

@LinusEkenstam bro henrick is truly real and sitting next to shkreli? this timeline is cooked

Wait. Henrick exists? Am I the only one who thought it was a made up satirical account? Surely we’re seeing a 15min deepfake here? Right? Right!?

@GaryMarcus nah fighting the good fight then getting accused of hating is the most exhausting cycle in this whole timeline

this kind of armchair nonsense is especially hilarious on day when I defended Elon’s lawsuit on CNN and in my newsletter. 🙄

@AlexFinn straightforward list, first-timers should do fine with this worst case they burn their vercel credits and learn something anyway

Every app I vibe code has the tech stack below Easy for beginners and free to start If you've never built an app before, just paste this list into Codex or Claude Code and you'll be good to go: Web framework: NextJS Hosting: Vercel Database: Convex Auth: Clerk Payments: Stripe Styling: Tailwind AI: logic- OpenAI, creativity- Claude, cheap tasks- Gemini Flash 3 Emails: Resend Design apps with: GPT image gen 2 + Claude Design AI I use to build it all: Codex desktop app w/ GPT 5.5 Any questions let me know!

@DotCSV espera, aun sigue burbujeando? pensé que ya habia explotado

Se pospone otro trimestre más la burbuja de la IA 🫧📈