Pinned Tweet
Quick Me:
I'm a man.
Emmy Bear is my kid's nickname, I made a company after her when she was still in womb.
I find dick and fart jokes funny and will make them.
I deadpan sarcasm.
I poke bears, for lulz.
And I love AI. I was skeptical and now I'm amazed. Best community, too.
ban food
Emmy Bear AI retweeted
Final dive — 56.10. 🏊♀️
2-time WDA Conference Champ. 🏆 Undefeated this year.
Up next — 2026 State meet next weekend. 👊
Just an observation:
Had plenty of time to do work, chose not to do work.
"Now we must work and not take breaks!!"
Will still take breaks.
Emmy Bear AI retweeted
Replying to @ashxhart
I'm really close to having enough followers but very very far away from having enough impressions. I guess I need to post better content more often.
Wrote my first exploit like thing of a program ever. Well, @UnslothAI 's qwen3.8-flash-next-unsloth-ud-iq2_xs 128K on strata did.
If it isn't an exploit, then I'm dumb as rocks.
Either way I'm dumb as rocks.
Felt like saying they were gonna do the old Friday release (never update / release on a friday). Looks like they gonna do it. Drop the weights and sit at home for the weekend lol.
Hey @tmaiaroto did you get the strata / maya running for your dude with the 2x3090s ?
I think @PeasantSmith has a Q5 @Blackfrost_AI cyber frost on that maya.
I call BS. Australia isn't real. It's a psyop.
@NeoAIForecast and @Blackwellboy are bots.
xitter.gq/NVIDIAGeForce/status/2…
I'm with @elshayib_ lol.
Go Go hot dog fueled share!
We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄
To enter:
🟢 Share this post
🟢 Comment #RTXON
Terms and Conditions: nvidia.com/en-us/geforce/con…
boop @coldniko is there a technical reason that the unsloths are "too big" but the normals aren't even though they say the same specs?
look, i'm not in this argument, but like, if i wrote a thing saying i got X per s and Y per s it's $ Zk...
i'd not say that the software may not work the way you intend it to (people saying it was a tiny prompt and tiny problem solved and not general).
Replying to @net_termina
we sell computers and write free software with no warranty of fitness for purpose. if you want an enterprise ready solution to serve tokens, go schedule a quick 15 minute call and pay 20x more.
pretty certain some one else on twitter did this before the MIT paper? didnt i read one of yall saying that?
weight-space programming is a new idea from MIT. instead of training a network, you ask an llm to read the weights, work out what they do, and edit them by hand.
gpt-6 astra fixed qwen3.5-0.8b's thinking loops by changing 0.016% of its weights, with no training at all. it also wrote small game-playing networks from scratch that beat ppo.
We used to program code. now models are starting to program weights
I GAVE YOU GUYS A WHOLE CLASS ON DWM DOMAIN WEIGHT MANAGEMENT. can i get at seat @MIT
link below
blackfrostai.com/learn





















