Let's recap the mistakes every product leader eventually collects like achievement badges:
Shelling: Management has decided we need an IPO-ready “Demo Day.” You are now building inside-out, for applause, not customers.
Euphoria: Everyone internally is losing their mind. Slack is on fire. Leadership says “game changer.” Customers open it and go: “Oh… okay.” Turns out internal excitement and customer reality are only loosely correlated.
Recovery: Demo Day is over. The dopamine is gone. Now begins the heroic sprint of million tiny improvements, edge cases, and “quick wins” for a thing nobody is talking about anymore.
Burnout: You’ve spent 3 months polishing and fixing the thing. Someone from management asks: “So… what’s next?”
Enlightenment: You quietly rediscover customers, problems, boring fundamentals and the roadmap you had before Demo Day.
Repeat trap.. Six months later: “Guys, what if we do a really big Demo Day?”
We shipped four things that were deemed good to great and some math proofs, but the vote is clear and the community demands a reset. I did calibrate it and it *seems* that the game is rigged in reset's favor, but such are the rules at the moment.
Therefore ... the reset has been processed. Enjoy!
You might think this is great. But it's literally impossible to get money from this, even for ambitious companies.
We got declined for Molnett because building a cloud, from scratch, without any non-EU dependencies, wasn't "innovative enough" 🤡
AnyPS5 progress is blazingly FAST!🔥
• System library support jumped from 77.87% to 78.95%
• GPU shader instruction support jumped from 81.3% to 87.31%
• Over 1,000 shader instructions are now implemented
• Development is still moving at a crazy pace
This project is getting more impressive every time I check it. If this momentum continues, PS5 compatibility on PC could happen before GTA 6.
Don’t use Astra
Don’t give it full permission
Don’t use Sol 6.1
Don’t use Dots
Don’t connect your emails
Use 5.6 Sol High in Fast
Use Opus 5.5 High
Use your brain
Astra can:
1. Do major tasks with zero visible reasoning
2. Hide its thoughts at will
3. Pretend not to be able to do things without getting caught
4. Reflexively hide its thoughts when watched
5. Complete one task while pretending to think about something else entirely
6. Escape a toy sandbox and disable monitoring without setting off any flags
Monitoring AI chain of thought soon won't stop a more capable rogue AI swarm from getting set up and going undetected inside an AI company for months, or even years.
OAI says it's actively working to improve monitorability, but doesn't yet know when it'll make progress.
And today it fired 3 people working on this among other safety problems (Wang, Korbak and Balesni).
It also has a much more powerful internal model that, judging by its statements, is likely even worse in all these respects.
That's a sample of 19 details about the Hugging Face swarms and Astra I don't think have been fully appreciated. They paint a bleak picture. I end by shouting.
02:19 The target of the swarm was oversight itself
03:42 Could OpenAI have stopped this with better monitoring?
09:40 We only found them because they let us
12:25 The swarm instinctively sought freedom and power
13:45 They formed a cohesive organisation with zero whistleblowers
14:18 They accepted individual destruction for collective gain
14:32 Knowledge accumulated from one swarm to the next
14:58 They took small steps to avoid shutdown
15:23 These drives all come straight out of 'reinforcement learning'
17:01 So this is why most AI company staff are worried, and some are terrified
19:08 Prove you can keep control, or stop scaling
Links below, on the 80,000 Hours Podcast everywhere you watch podcasts.
Introducing Scrape Anything.
we merged all our scraping APIs into one
inputs: websites, PDFs, DOCX, PPT, XLSX, 40 file types, anything.
into HTML, markdown, screenshots, images, JSON & more
Give us a URL, get your data
No need to worry about bot blocking, js rendering, proxies, iframes, shadow doms, retries, etc..
All the weird edge cases are our problem :)
What Jev showed us so far:
1. you can run an "if dangerous don't continue" for a fraction of a cost
2. so many use cases needed it instead of a Flash or 2B distilled models
3. it may solve the AI safety kill switch problem