Chris Paxton

OmniExtreme from BIGAI: a general-purpose tracking policy which works with extreme motions and complex contacts. Works by pre-training a general-purpose flow policy, then learning an actuation-aware residual policy for handling complex physical dynamics.

I do think this is just them goofing off, the military versions are less colorful and way scarier

Drone and dog shooting fireworks at each other

Video from 阿阳 / Shenzhen media group

Xiaomi humanoids with self tapping nuts. 90% success rate, 3 hour deployment, using tactile

Next on RoboPapers: VLam4VLA -- what makes a vision language model good for robotics

Now let me design the plan with a Plan agent. [...] Excellent analysis from the Plan agent.

claude giving claude a medal never gets old

"bullshit bench but for people"

Alibaba Qwen remain the champions of open-source ai

Qwen releases 4 new Qwen3.5 Small models!

Qwen3.5: 0.8B • 2B • 4B • 9B

Run Qwen3.5-0.8B, 2B and 4B on your phone. Run 9B on 6GB RAM.

The vision reasoning LLMs perform better than models 4x their size.

GGUFs: huggingface.co/collections/...
Guide: unsloth.ai/docs/models/...

Bullshit Bench V2

new: 100 questions across several domains

- Anthropic & Qwen still on top
- Reasoning seems to hurt
- New models are *not* better than old (except Claude)
- Seems to be independent of domain

github.com/petergpt/bul...

bsky.app/profile/did:...

did he really say this lol

well so the thing about him is that he's a liar

Earliest dyson spheres were built inside out covering increasingly large fractions of the builders' native planet.

I think he is working his way towards that point of view tbh

roon @tszzl X.com there is no contractual redline obligation or safety guardrail on earth that will protect you from a counterparty that has its own secret courts, zero day retention, full secrecy on the provenance of its data etc. every deal you make here is a trust relationship

this tweet is such a wild ride

how can you write something like this and not conclude that you should never ever ever ever work with the DoW?

seriously, if this is your position, how is it remotely feasible?

The world's largest offshore solar farm in Shandong

open.substack.com/pub/hyperdim...

AI replacing all work isn’t a “doomer” scenario, it’s a utopian vision of a post-scarcity world in which it would finally be possible to enact every socialist dream where everyone has high living standards and freedom from want or oppressive toil.

It needs more legs

Two waymos struggle to get past each other. But they do figure it out! And this video makes it incredibly clear they aren't just being teleoperated. The failures are always more informative than the successes!

Re: anthropic:
- it's interesting how unified the "AI guys" are in support of anthropic; even the OAI die-hards are on their side of this.
- the right-wing "pro freedom" VCs on the other hand...

Has science gone too far

It was all about spying on Americans: www.theatlantic.com/technology/2...

so everything on earth has been monotonically more and more about ai, every year, for three years

if the trend holds this is the least anything will ever be about ai again

fellforitagainaward.jpg

ManiFeel is a benchmark for contact-rich manipulation tasks. Force- and tactile-based tasks with occlusions and partially-observable state are going to be crucial for true general purpose robotics.

Website: zhengtongxu.github.io/manifeel-web...

Insects are fascinating to me. Intricate miniature control systems produced for "cheap" and en masse.

It's a glance at how much we still have to catch up to nature.

If you can make them use'em! That's the next logical step. I think this project has legs, creepy, crawly legs.

economics notches another W

energy, permitting, resources are all super expensive in the USA. chatted with some people who failed recently. they said the economics would have worked if they could legally grow marijuana, so it's at least possible with lower input costs lol