“tiny models behaved the most like people” is a scary sentence in 2026
Cara Leong@craa.bsky.social · 2d agoQualitatively, while people sometimes misremember facts, models rarely do. Instead, models forget too much, which is why people and user sims get different questions wrong. (Tiny models behaved the most like people here! 📉)
The replies seem to believe that the number one priority in fighting a fire at the library of Alexandria should have been minimizing water usage
Internet Archive@archive.org · last wk.An update from Brewster Kahle on today's outage.
actually, 4 is a very hard to guess number, and is secure from modern attacks
No holes if your plane doesn't come back at all
Pavel@spavel.bsky.social · last wk.The invincible airplane
do think there's a danger of looking at extremely cool shit like a car driving upside down then doing a kickflip before landing, and convincing yourself it's actually lame because the guy funding it is a white supremacist
Yahoo Finance@yahoofinance.com · 2w ago"Flying cars are real!" 🚀 Elon Musk and Tesla showcase the new Roadster.
i found the butterfly neuron! this is fun. it is neuron: "stage3.block2 · post-product coordinate feature 295"
Eris ⌬@eriskii.net · 2w agoIsopod neuron
What has been will be again, what has been done will be done again; there is nothing new under the sun.
Evan@evansmithhist.bsky.social · 2w agoToday I found out that Umberto Eco crafted a version of this meme in the 1980s…
XKCD: I wrote xkcd.com/221/ "Random Number" as a cautionary tale. Tech Company: At long last, we have created the Random Number from classic web comic Don't Create The Random Number.
mr. TIM@timkellogg.me · 2w agowow, i had my doubts, but this is true in every single LLM i can find
We really lost something by upgrading image generators. Plugging in shit like “my happy family at the beach” gave you horrifying things like this and it was actually great
the answer is no, they do not know if many of these results are just slop, and have already withdrawn 3 of them
ponder@ponder.ooo · 3w agodoes openai even know if these results are just slop 😭
We byteified Qwen 3 8B & Llama 3 8B to create Bwen 8B & Blama 8B. Both come close to matching their source models' performance in our evaluations. Bwen 8B also outperforms Bolmo 7B across our aggregate evaluation suite. buff.ly/9BSybFo
Still issue with one finger, easy to identify it's AI If you know Cyriak: look what they need to mimic a fraction of our power
werewolf ver mitzvah 🗿@ver.ooo · 3w agoWelp Gemini can do gifs now