JUST IN: Vibecoder discovers previously unmapped ancient Maya settlement hidden beneath the Mexican jungle using 48 AI agents & NASA lidar data, despite having no archaeology experience.
This week's podcast episode is my full @DohaDebates on the nature of money with two MMT economists, @jasonhickel and @ptcherneva. This was a good one you don't want to miss!
Multiple sources telling me one or multiple people affiliated with Venezuelan stablecoin app @kontigo_app have been “detained.”
Emails to the two cofounders asking for comment have not been returned.
Still struggling to process the news. Rest in peace Ximena. She was a true patriot and believer in liberty. Her kindness and warmth will be missed, from letting me crash on her couch to hosting the best steak dinners. See you again one day on the other side.
Tolkien knew that power is never seized in one day — it is corrupted ring by ring, each bargain small and reasonable. Every dictator followed the same forge. That's why we teach activists to read myths: they are the oldest manuals on how tyranny is built.
"As Ortega addressed cheering crowds, he appeared to speak to a bouquet of roses in his right hand, oblivious to the lapel microphone pinned to his white shirt."
Straight from a García Márquez novel.
https://t.co/OzS8GzHyCB
It's really wild working with abliterated models primarily.
You ask it to do something, and it'll just fucking go.
Wild. We've conditioned ourselves as an industry to toe step around AI and our AI overlords.
No longer needs to be the case. The machine works for us now.
At least four times as many active duty personnel and veterans of post-9/11 wars have died of suicide than have been killed in combat.
https://t.co/d312wyzamz
I was at the Novaya Gazeta yesterday as its reporters and editors packed up and held their final editorial meeting at Potapovsky Pereulok. The newspaper survived so much, but it could not survive Russia's war in Ukraine.
One reader I talked to described the disappearance of the Novaya as a "spiral of silence," saying:
“When voices like these disappear from public life, it starts to feel like there are fewer and fewer people who think differently, like you’re alone in what you believe,” said a 21-year-old reader, Danila, a recent university graduate with a degree in computer programming. He asked that his last name not be published, to avoid trouble with the authorities.
“The less you hear people expressing views different from the official narrative,” he said, “the harder it gets to speak up yourself.”
https://t.co/0BQ6UIwHpe
Many are saying (me), that bitcoin is well decentralized.
Miners have the tools of decentralization (template control) and interest (profit motive) to use them.
Congratulations to miner bc1q~g4m7 for solving the 321st block - yet another super lucky one at https://t.co/TIIWvCYUk6 with ~161TH of hashrate on a 161TH worker! A miner of this size has about a one in 126 chance of solving a block in a year.
https://t.co/GUu18ySpHs
around 99% of my bitcoin lightning payments settle successfully within 2 seconds, and that's been the case for over 5 years.
lightning haters are either hallucinating or they're trying to sell you some shitcoin.
BIG update to GLM 5.3 Flash for DGX Sparks! 🔥
TP4 support:
- You can now run the new recipe on 4 DGX Sparks.
- Switchless supported!
- You can choose to
- TP3 support coming soon.
Two TP4 lanes: EXL3 and FP8
- On TP4, you now have the option to load original FP8 weights instead of EXL3!
- FP8 is better quality overall, but not by much.
- Total available KV on FP8: ~3.15M.
- Total available KV on EXL3: ~8.39M.
- Decode performance is better with EXL3.
- YOU decide if it's worth or not.
Performance TP4 EXL3, non greedy prose:
- 77 tok/s single stream
- 98 tok/s 2 concurrent streams
- 172 tok/s 8 concurrent streams
- Prefill 256k 1900 tok/s
Performance TP4 FP8, non greedy prose:
- 52 tok/s single stream
- 69.5 tok/s 2 concurrent streams
- 119 tok/s 8 concurrent streams
- Prefill 256k 1800 tok/s
Improvements & fixes to TP2 + TP4:
- 34% faster decode on file rewrites, single stream
- More reliable replies: fixes repetition & improves tool-call handling.
- Safer startup: catches invalid settings and fixes false errors.
- Cleaner builds: removes compiler warnings without changing results.
- New options: separate reasoning from answers and combine selected model layers.
Get it here:
https://t.co/ykfJ56l1vT