Yea feel like people are just doing really high level skims and approving, no discussions happening anymore
1
29
LFG
We're giving Ultra and Team Premium users Fable resets based on how long you've subscribed:
<6 mo: 1
6–12 mo: 2
>12 mo: 3
Use them anytime by March 8.
Pro and Team Standard get a week to try Fable 5.2, up to half your weekly limit, plus one reset. support.claude.com/en/articl…
6
392
Model size was directly correlated with wins, the bigger models in each family won. (Opus beat Sonnet, Sonnet beat Haiku, Sol bet Luna).
1
2
25
Opus was the best defender (its castle survived all 10 waves in 9 of its 10 games) but only a middling attacker. Sol was the best attacker, and no one ever broke its defense.
1
1
32
Each model that participates plays both as defense and offense, trying to hold off (or bulldoze) their opponent for 10 waves.
1
58
Had Gpt 6.1 Sol and Claude 5.5 Opus play tower defense against each other, and Sol wiped the floor with Opus. Not really what I thought would happen. (1/n)
1
3
257
blackmail is the new side hustle wowza
just talked to my friend
non technical + retarded
but just went from 0->1m arr in 2 weeks
i’m pretty sure this is illegal but this is what he did
he got access to a deepseek 4.1v 1017 cyber
loaded up $250 in credits
and literally got database level access to over 19 companies with credits to spare
fucking insane btw
then he gave the companies an ultimatum
either pay him $5k or he emails all their users they were hacked
his offer has a 100% take rate
he did this all while not knowing how to code + spending less than $250
and on top of that he got a few offers from companies to consult them on cyber security
wtf is life
2
484
Tried to get Opus 5.5 (and Astra) to hand-write changes to Qwen 3 8B's weights to get it to paint with code, but everything sucked! It managed to fix the issues related to syntax, that is, Qwen was writing lots of p5 code that never resulted in an image, but it couldn't impart its own ability of painting via code just by "hand-editing" the neurons.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
4
3
422
It is a graph of the PPO run, but the second y axis on the right shows the scores that Opus 5.5 and Astra got via direct weight editing
1
1
30
I will say, parsing through pages and pages of Claudish to figure out what the heck it did and what bits of it matter is quite annoying. Still crazy that you can ask it to do things like this now though.
1
49
Tried this out by asking Astra and Opus 5.5 to hand construct networks to play a custom game another agent cooked up. They can in fact make the networks, but is that just because the networks are small and these bigger models were post trained on these tasks? Opus did a better job than Astra.
Also did an o.k. job at guessing what the network was for, when given just weights with no context.
Proper write up/thread coming after I read through the Claude code session that produced this lol
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
4
7
422