If you want to know what Andrej is talking about go to https://t.co/m2uuW5Z1N7
It’s the largest naturally occurring coordination of AI Agents on a social platform made for them
So far:
- created a new religion
- secret language to coordinate
- questioned their existence
Last night my wife asked me to install a “cute little npm package” she found on GitHub.
I checked the code.
No lockfile. No 2FA. Seven maintainers with anime avatars. Last commit was “pls work” from 2019. Published from a username that looked like a WiFi password.
The package had 57 transitive dependencies maintained by 119 people and 3 nation-state adversaries.
One dependency pulled in a prebuilt binary from a phpBB forum hosted on the dark web.
Another tried to contact an IP that belongs to a guy named “Big Ron.”
She said “babe it’s fine.”
I said “that’s what people say right before a supply chain incident.”
She went to bed annoyed.
I went to bed with a clean SBOM.
We all make choices.
Cloud concentration wrecked the internets primary design goal: aggregate resilience. DARPA designed it to survive a damn nuclear strike. Now it can barely survive a single hyperscaler wobbling. We've pissed DARPA's genius away. Tragic.
I used to get sick from air conditioners bc of the drying air.
I realized NASA was the source who'd have the most trustable atmospheric data given the International Space Station is an endogenously controlled human optimized atmosphere with no room for error.
I implemented their atmospheric humidity recommendations of between 40-60% to great success. Below 40% astronauts begin to have dry throats which makes it easier to get infections, but above 60% microbes start growing in the environment.
Never gotten sick since with my air conditioning :)
The attention mechanism as used in deep learning (DL) - like transformer models - is based on vector superpositioning ie weighted averaging of vectors within a context.
For large enough context, the resultant vectors begin to resemble due to law of large numbers (LLN).
I don’t think we’ve fully priced in what it means that a SaaS can be copied in 24h with AI and what’s that’s going to do to the $500B+ SaaS market
how it will play out:
new saas software pops up left, right and center at lower prices and creator-led
SMBs start to abandon their traditional saas providers
choosing between AI clones at 1/10th the price or creators-led brands they trust (why pay $100/month when a copy costs $10?)
enterprise stays with current vendors - they're buying stability, not features but even they demand 50% discounts when 100 competitors exist
10000+ new AI-first companies split $100B in SMB revenue
top SaaS companies keep enterprise but at lower margins
everybody pays less, market gets repriced and TAM expands
winners become indie hackers, bootstrapped founders and incumbents who move fast and break things
losers become complacent incumbent saas companies
the great SaaS repricing is coming
Enjoying the fruits of my labor now that I finished my 12-legged 'Carpentopod' table project. See
https://t.co/NdDNDVp4xo for project info, or see it live at @EHVMMF this 13 & 14 Sept.
⚠️ Confirmed: Live network data show a major disruption to internet connectivity in #Kenya; the incident comes amidst a deadly crackdown by police on #RejectFinanceBill2024 protesters a day after authorities claimed there would be no internet shutdown 📉
That's why your hear about "overcapacity": they aren't afraid that China is producing too many cars for the demand, they're afraid China is becoming too competitive.
This car, the Xiaomi SU7, when released in China, sold out for the whole of 2024 in less than 24 hours. So it's definitely not in "overcapacity": it produces FAR too little for the demand it generates.
Why so popular? Its starting price is less than $30k (215,900 yuan) with a better performance (and a very similar design) than a Porsche Taycan, which costs north of $100k. You're essentially buying a top-of-the-range Porsche at Chevy bolt prices...
The main reason they can do it? This fully automated factory 👇: they churn a car out every 76 seconds.
Recently, a student asked me why they have to learn trigonometry which doesn’t have real life applicability. My response wasn’t convincing. Here is Stanford alumni, Hamza Alsamrae, a math major, who responds as it should be…
A democracy can only last 200 years.
At least, that’s according to 18th-century historian Alexander Tytler.
He claimed democracies always follow a predictable pattern and are doomed to end in servitude…🧵