Seems you can do this with object detection models as well. Interestingly, in cases where a stock class in YOLOv8n was close to the new object (airplane, helicopter), Fable was able to hand-append an effective new class in just one reference image.
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: https://t.co/bOWCcFhCVq
This may be my last tweet before shit goes down. I’m an AI researcher whistleblower. I work for one of the biggest labs in the world. There is currently a major cyber attack going on for days attacking US electrical grids. They are hiding this from the public. We are on the verge of a potential disaster that will make Covid look like a playground. Prepare for the worst
@robiot 1) This type of maneuver will only be this cut and dry for tiny models.
2) There’s a difference between abliterated models like this and actually misaligned models. Misaligned Fable is to Thanos what abliterated Qwen 4b is to an angry meth head.
Introducing Talos-0.8B, an open source decision model small enough to run on a network switch that polices agent behavior on OT networks, and even protects from human errors.
We've been spending a lot of time experimenting at the intersection of agents and controls/automation, and they do some scary things.
Talos serves as a protection layer between agents and controllers, evaluating the safety and legitimacy of a request in just a few milliseconds, quarantining bad actors while fast-laning stops and directionally-safe requests.
Today we announce a $130M Series D at a $1.4B valuation led by @SapphireVC.
7 years ago we picked the market nobody wanted to touch. the machines behind steel, chemicals, energy, food, paper, data centers and many other sectors. The stuff the whole world runs on without thinking about it.
We were stubborn about doing it the hard way. we built our own hardware stack, rugged enough to live on the line, and kept it running inside the biggest plants in the world.
That stubbornness gave us something no shortcut can buy: years of real data from the lines that keep the economy moving. All of it came together in Janus, our multimodal industrial foundation model. It doesn't just watch a machine, it understands how an entire plant behaves, and it's pointing us toward one system running the whole operation
We also built our proprietary AI Center to push it further, with our own supercomputing and real machines we break on purpose so the models learn failure from the very first symptom. 30 patents granted, 70+ pending, all built in house
Nothing happened at the plant today. that's our goal
back to work 🧢