Pinned Tweet
We're working on a new harness for JEV type models "XERA JEV" with my goat @RoliumGens.
Can't wait to show you the result! 🔥🔥
Angel. retweeted
Another DeepSeek drop: dsh-superagent
An agent framework that works through a repo's GitHub issues and PRs with many agents in parallel, designed to take as little human attention as possible while delivering as much value as possible.
github.com/deepseek-ai/dsh-s…
I switched to Anthropic a month ago and seeing this I'm very happy about that 🤣🤣🤣
Bad news for ChatGPT users: starting in December, Plus, Pro and Business limits will be shared across ChatGPT and Codex.
OpenAI quietly updated its help page. Usage will now be measured in tokens, so bigger models and harder tasks burn through your allowance faster. Only Instant chat stays free of the limit. Hit the cap and you wait, upgrade or buy credits. Heavy Codex users may feel this one.
Qwen 4 is comming!
I hear a lot of good things about it, can't wait to see the 27B and what this monster can do!!🔥🔥
Hell yeah!!!!!
When you go into datasets on HF the dataset I released yesterday is on the trending page!!!
Incredible, I didn't expect much visibility from this to be honest. 🔥🔥🔥
Look closer at these four rotors
Same prompt, four models: Claude Opus 5.5, GPT-6 Astra, DeepSeek V4.1 Flash 1017 and Claude Sonnet 5.5, each asked to build a helicopter swashplate. It’s a tough test because every actuator, bearing and link has to connect and move correctly when the rotor tilts.
The differences in mechanical detail are striking. Worth a watch for anyone following how the latest models handle 3D and engineering tasks.
Look closer at these four rotors. This is engineering, not animation. 🚁
Same prompt, four models: Claude Opus 5.5 · GPT-6 Astra · DeepSeek V4.1 Flash 1017 · Claude Sonnet 5.5
A swashplate has nowhere to hide. Every actuator, bearing and link has to connect. Raise or tilt the ring, and the motion has to travel through rigid linkages to every blade. One stretched rod or one floating joint, and it's fake.
I'm not scoring them. You tell me: which one is real engineering, and which one is just spinning metal? 👀
Opus, Astra, DeepSeek or Sonnet?
Qwen4 is almost here 🔥
The Qwen team just teased the Qwen4 series live from Amsterdam: Qwen4-Max, Qwen4-Flash & Plus, and Qwen4-27B. The release is reportedly planned for next week.
Definitely one to watch for anyone following open and frontier models.
#Qwen4 #AI #LLM
Introducing HolyCyber, a defensive cybersecurity instruction dataset that shows its work.
51,805 records · 17 domains · 127.6M tokens
Every row ≥2,000 tokens, grounded in 3,602 real sources (ATT&CK, CWE, OWASP, NIST, arXiv)
Trained to build detections, triage incidents & refuse the dark side
huggingface.co/datasets/Reiz…
We're cooking something that could possibly change everything in AI's world.
lmao the progress i've done on the O(n) architecture is CRAZY.
it converges so quickly. Tomorrow is when i'm gonna do the most progress as i'm gonna lock in on it with a friend we're gonna figure everything out.
the inference engine is ready, I'm iterating a few things with people who have other rtx cards than mine, to create all the recipes and making it work with gpu-unlock ( which i'm hillclimbing ).
I'll be able to release this in a few days, it's gonna go brrrr!
Who's in the deep waters of quantum maths with me?
share what you feel I would be pleased to read everyone 😂😂
Introducing, Qwen-Image 2.1 Uncensored quantization
A per-layer mixed-precision quantisation in 4.9 bits
Every one of the 297 tensors was quantised independently, with all other tensors held at full precision, and the format was then chosen per tensor to minimise total reconstruction error under a size budget. Result: Q4_K quality budget, ~10 % better fidelity than uniform Q4_K, at 4.39 GB instead of 4.00 GB.
huggingface.co/Reizxn/Qwen-I…
BRO that could be a new way to modify the games while building them directly in game or DLSS.
Analysing the game, find problems or visuals you can improve (like DLSS) while you're building or playing the game and modify it in stream:
- animations
- textures
- code
- designes
-lighting
...
let me know if you think this Idea should be studied???
Me and @RoliumGens are working on a harness with the same subject.
happy to see that this getting more and more popular!
Introducing Optimus-1, the first swarm harness for autonomous research. Generally available in ArkenOS, starting now.
Optimus-1 lets ArkenOS spawn a swarm that runs the entire discovery loop on hard research problems autonomously. The harness samples the hypothesis space in parallel and works the tails of the distribution, chasing every anomaly until it resolves into an artifact or a discovery. We want every researcher to have a lab of their own, and to see how far their curiosity can go.
Unlike coding agents, Optimus-1 is designed for long-horizon discovery, where most branches fail, and the hard part is knowing which result to believe. It keeps what replicates and keeps pushing past the edge of what's known.
We're still early, but ArkenOS is helping hundreds of teams, from RL environments and data labs to frontier research groups, scale the alpha in their core research.
If you're working on a hard research problem and want to try it, my DMs are open. I'd love to hear what you'd point it at.