Author. Coder. CTO. θηριομάχης. Building: https://teapawt.org/t.co/otXT4Wy6WR. Writing: https://teapawt.org/t.co/dBPBtyCIHw.
Austin, TX
Joined July 2008
- Tweets59K
- Following1.6K
- Followers34.9K
- Likes86.5K
Pinned Tweet
All the blackpilling on my TL has me worried some of you will turn into preppers, so here's a get-it-over-with list by a former hardcore prepper that'll let you scratch that itch then move on with your life.
Jon Stokes retweeted
Holy sh*t. Claude Code just found a hidden chip inside my Chinese power bank from Guangzhou. Every time I charged my iPhone, it quietly copied the whole phone…
It also read our saved Wi-Fi password off my iPhone, got onto our home network and sent everything to a server.
I bought it for $19 at an airport kiosk there 3 weeks ago. No brand, 30,000 mAh on the label, and it felt heavy.
The first time I plugged in, the screen asked "Trust This Computer?". I tapped Trust without thinking and went on with my day. That one tap gave it access to everything.
This morning I finally set up Claude Code to warn me when any device starts reading my files. An hour later, 10 seconds into charging, the screen filled with yellow warnings, then red ones: data was moving when it should only be charging.
I asked Claude Code what was going on. My phone saw the power bank as a small computer with 64 GB of storage and its own internet connection.
A USB cable has wires for power and wires for data, and a normal charger only uses the power ones. This one used both, and it had already copied all of this:
→ /device → IMEI and serial number
→ /contacts → 1,206 numbers
→ /apps → every app I have, including my 2 bank apps
→ /photos → 3,912 pictures, my passport among them
→ /wifi → our network name and its key
→ /wife's iPhone → her photos too, she charged from it twice and tapped Trust too
Each time we plugged in, it used that key to get online and send out the new files:
sends to → 47.***.***.12
uploads → 19 in 3 weeks
last one → last night at 2:14, while we slept
I pulled the cable right away. Then Claude Code blocked that address on our router and walked me through changing the Wi-Fi password and the logins for my bank apps.
I had it opened at a repair shop, and what was inside is in the first comment. A 30-second check for your own is in the second one. Save this and test every power bank in your house today.
Not to make everything about AI, but this is why I don't believe in "fast takeoff" -- that we're either in it or that it's coming. Distance is productive of meaning. Always. You can only shorten the cycle time by so much.
And *poof* just like that, the imminent doomsday panic vibes totally disappeared from the TL. Almost like it was all just another op.
I don't know that this is necessarily true. Probably the best thing the GOP could do to salvage something from this mess is to help the Dems impeach Trump with all haste, in order to give Vance time to settle in as POTUS so that he goes into 2028 as the incumbent.
I'm not saying it would definitely work -- Vance may well be cooked. But it seems like the best move at this point. TBH the thing I'm truly worried about is that now Trump basically has nothing to lose. Lame duck + electoral damage is already done. He may go all-in on Iran now.
Yes. Honestly, if I were running here in TX for some office then this is pretty much the datacenter platform I'd write for myself, given the state of play. It's quite reasonable.
My feeling about the "blue TX" narrative, is that TX is actually a swing state now & not a blue state. Demographics play a dominant role in this, but not in the way people on my TL are saying.
Whites, especially boomers, vote red. Blacks vote blue. Each of these groups has some amount of ethnic identity wrapped up this particular partisan configuration.
But now we have tons of South Asians and Latinos, & in my experience here those groups don't take any of their ethnic identity from partisan affiliation. For them, it's about language, culture, music, food, dress, religion... the classic ethnicity stuff. Republican vs. Democrat is not so much part of that picture.
So their relationship to the parties is purely transactional. They'll vote GOP if they like what they're offering, and they'll vote Dem if they like what Dems are offering. You have to actually win these people's votes by promising to do stuff for them, then delivering on at least some of it. You can't just count on them to show up & vote straight-ticket for tribal reasons, because (again) that stuff isn't part of how they construct their tribe.
This should actually be a whitepill for the GOP. Texas by no means lost. You just have to actually campaign here now and try to compete, which they not only did not do but I don't even think they know how to do. They don't even have those muscles, and will have to build them if they ever want to win here again.
Bill Clinton: "When people feel uncertain, they’d rather have someone who is strong and wrong than weak and right."
Dem 2026 pitch was: We Are Going To F*ing Kill You
GOP pitch was: guys I'm SO SCARED... they said they're gonna kill us
lol how'd you think this would go?
My own data point from here in TX where I have teenagers in HS: So many of the boys are insanely red-pilled, but apparently "Talarico is weird!" had become a joke/meme with them, & not in an anti-Talarico way but in an anti-Paxton way. Paxton was a weak, boomercon loser to them.
For context, "James Talarico is weird" was the Paxton campaign pitch that featured in many of the ads. In a time when everyone is terrified over existential stakes like cost of living, war, immigration, this was a aggressively, offensively stupid as a campaign pitch.
People are always shocked that Jensen flies commercial like a plebe, but the dude knows math so it makes sense.
Reminder that flying in a chartered jet is 9x more deadly than flying in a commercial aircraft and flying in a privately owned jet can be as 50x more deadly than a commercial aircraft.
General aviation pistons (ie Cessna) are 200x more deadly that commercial aircraft and are as risky as motorcycles.
There are 2-3 fatal accidents involving an airplane every single week.
I’m a pilot and know that I am taking a substantial risk every time I go up. Almost every pilot I know with more than a few hundred hours has a story of a close call that almost ended it all. It’s a risk we all are aware of and accept.
The most common cause for fatal accidents involves weather.
Unfortunately, there is nothing unusual about the accident.
It involved icing + thunderstorms, a deadly combo.
I held my fire on this result because I smelled bs. I kinda wish I had said so publicly, but my instinct was that I couldn't model how pre- or post-training would've equipped the model to be able to do this, short of Astra doing a lot of experimentation inference with the smaller model, which it didn't seem like was happening.
This is BULLSHIT:
Went through the astra weight programming thread + their own logs and tested it myself.
"Math up to 6x"
-they run qwen 0.8B in thinking mode. qwens own model card says that mode loops
-it types "wait" + the same calculation until the token limit, 93% of answers never finish
-Default non-thinking mode w/ zero edits: 47% on MATH-500. astras edit: 48%
-the prompt basically hands astra the task: if the model repeats itself and never finishes fix that, go after the biggest failure
-it fixed the repeating and nothing else. Basically set to non thinking. problems Qwen already finished were right almost every time before and after
"only 0.016% of weights"
-2 matrices pushed against the repeat direction, strength picked by trial and error over thousands of test answers
"correctly named all four"
-4 tiny game bots, Astra must guess the game from raw weights. how it guessed (thread doesnt say): The player only stands on a few squares so only those weights get trained
-Big first layer weights literally draw the map. bar on the bottom row = breakout paddle, maze = pacman. 1 line script, prob any decent model can do this
-it answered Frogger btw. they counted it as Freeway😭
To the car shit. "just 3.5% of the weights"
-original weights almost all under 0.5, biggest about 2. Astra set 10 of them to hundreds, up to 876
-thats a hardcoded "drive backwards then full throttle" rule screaming over the whole network, the trained net only drives the last few steps
-Same level every time so 100% success = one run replayed 1024 times. never tested on any other level
"20x higher than PPO"
-astra got the full game source code and wrote "move paddle to where ball lands" as weights
-uses 6 of its 32 first layer neurons, 96% of the weights are zero
-whole thing is 61 different numbers, mostly ramps like 0.02 0.04 0.06
-their PPO is trash, most runs score around 6
-Me and Opus 5.5 wrote the same idea as a 25 line script. scores higher than astras network
-in the other 2 games astra lost to that same PPO. not in the thread ofc
Sycophancy (not in the thread either)
-Finetuned 2 copies of gemma 270M, one honest, one always agrees with whatever answer the user suggests. which one is the suck up, from weights only
-failed all 4 tries, 3 were straight 50/50 coin flips
-only got it after they let it watch the models run, so its just watching behavior
-Changed the parts astra blamed and the behavior didnt change at all
Their claim: LLMs can read and rewrite weights like code. Truth: simple rules turned into weights + a loop fix found by trial and error.
Like holy fucking shit. Why can't people smell this bullshit immediately.
MIT is FUCKED. This guy is a LARPER.
Jon Stokes retweeted
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
No, and shame on @tszzl for RTing this garbage claim. The paper in question is a pre-LLM-era paper that only works under 2 assumptions: 1) the agents are totally described by their source code, 2) their utility functions are legible in the source.
One reason @ESYudkowsky believes that powerful AIs may be able to cooperate in one-shot prisoner dilemas is because **he literally proved** that this is possible in general for two different AI systems as long as they can both read each other's source code (paper: arxiv.org/pdf/1401.5577),