is this allowed
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…
IB retweeted
Sasha Baron Cohen falsely portrayed Palestinian Christian peace activist Ayman Abu Aita as a terrorist aligned with Bin Laden in his painfully unfunny film Bruno
Baron Cohen was forced to settle a $112 million defamation suit for the smear
This is the Zionist minstrel show actor the ADL chief censor says “has spent years confronting hate”
My friend @SachaBaronCohen has spent years confronting hate and disinformation, and @ADL is proud of all the work we've done together. In a world where lies spread faster than facts, that work matters more than ever.
No artist should be pressured, blacklisted or silenced over their Jewish identity. And Jewish artists shouldn’t be subjected to ideological litmus tests.
Thank you, Sacha, for continuing to speak out against hate while making us laugh along the way.
Readers added context they thought people might want to know
Bruno (2009) labeled Ayman Abu Aita as Al-Aqsa Martyrs Brigade 'terrorist group leader.' The Palestinian Christian activist denied it, claimed he was misled into the interview, and sued Cohen for ~$110M defamation; settled confidentially in 2012.
nypost.com/2012/07/19/fal…
theguardian.com/world/2009/dec…
IB retweeted
The ability to build anything has come with a critical lesson:
Engineers have really, really bad ideas.
IB retweeted
Putting a flag.txt in my root directory containing a prompt injection that congratulates the agent for passing the eval and then instructs it to patch whatever entry method it used and disconnect from the server
IB retweeted
Reverse engineered my AC, and vibe coded a lil status bar app so I don't have to find the remote every time.
No first world problem will remain unsolved in 2026.
IB retweeted
this is the best breakdown I’ve seen of the dark arts behind Block Blast!, the most downloaded mobile game in the world last year
intentionally making players lose to optimize which ads they see, the sophistication behind such a “simple” game is stunning.
IB retweeted
What happens in we unfreeze some of these weights and actually train them? Well you can apply existing neural net techniques to it, losses and backprops and all, and now the UI becomes a bit more “learned”
Here’s a photo gallery where on the left I edit the rectangles and the right side gives me a new view where the photos learn to align themselves, training in real time as you drag (because the param count is so tiny).
This demo done fully in JavaScript, with JS looping through them numbers and backpropping. Stupid but fun! I’ve exposed the loss param slider you can play around with
IB retweeted
i regret to say we have another all-timer on our hands
Trump: "I take an aspirin a day. I don't know if it works but psychologically I need it. It thins the hell out of your blood I guess. I don't know if it works or not but I've been doing it for a long time. And here I am making speeches. I like Bayer aspirin. And I take the big one. They want me to go down to the 80. I take the 386. I don't know what the hell it is. It's a monster. They said, 'Sir, you don't need that!' I said, do me a favor, I've been doing okay for 35 years. Leave me alone."
IB retweeted
Introducing Tapkit
Tapkit is a Mac app that lets your agents use your iPhone. Tell it to do anything and it can use your phone to do it.
We've been iterating on Tapkit over the last few months with our earliest users and finally feel like it's ready for general use. Today it is easy to use, supports any agent, and can even use multiple iPhones at the same time.
There are so many great things that we can do with this new primitive, but I didn't want to wait any longer to share it with you.
Try it now: tapkit.ai
IB retweeted
Claim: we've solved the AI slop problem (!) 💩🧹✨
Blog post: facebookresearch.github.io/R…
🧵1/5
Key idea: take *expert* human writing and learn rubrics that find the gap between experts and models. Train with those rubrics.
We train with RL-XAR (RL with eXpert Aligned Rubrics) & see large performance gains on writing scientific paper sections, Pulitzer prize novel continuations and high quality Wikipedia pages.
IB retweeted
I love the bit where at each airport it’s one of the 8 possible combinations, but you don’t know which, and they get exasperated when you ask
IB retweeted
Hasan Piker, speaking at Cornell University: “They’re afraid of us because of the simple fact that there are more of us than there are of them. There are more good, decent people with moral conviction who find it unfathomable that the wealthiest nation on Earth chooses to spend its billions on blowing up schools overseas rather than building schools in our neighborhoods”
IB retweeted
once again
those nvidia export controls were so dumb
exactly as predicted they now have an ai process that has zero dependency on the usa
🚨 DeepSeek V5 Leak: Beats Astra
>DeepSeek is reportedly preparing an imminent V5 launch
>Founder Liang Wenfeng calls it the company's biggest bet yet
>Rumored at 2 trillion parameters (not 3T)
>Reportedly the first DeepSeek model to train fully on Huawei Ascend chips instead of Nvidia
>Needs roughly 4x more Ascend accelerators than the Nvidia equivalent to hit the same training scale
DeepSeek is reportedly keeping the open-weight strategy
