Krishna

A new mesh recovery method and model by our lab. My team is currently putting this on our robots, it works really well.

RoboPapers | Substack Chris Paxton & Michael Cho geek out over robotic papers with paper authors. Click to read RoboPapers, a Substack publication with hundreds of subscribers.

Want to learn about the latest in robotics research? Like, the real cutting edge, right from researchers? In addition to our youtube channel, I post notes and information about new RoboPapers episodes on Substack here: robopapers.substack.com

In practice, how do you separate design proposals/decisions from your natural conversation and random plans? Do you keep iterating on plans until you reach something stable (i.e. you’re in full alignment) or something else?

if i had more tokens i could have written a shorter reply

Am I crazy, or was @fchollet active on bsky last year (or the year before)? I wonder why he left and if he would reconsider posting here…

Nobody knows how the whole system works One of the surprising (at least to me) consequences of the fall of Twitter is the rise of LinkedIn as a social media site. I saw some interesting posts I wanted to call attention to: First, Simon W…

this is exactly right and explains why most handwringing about building systems we don’t understand is misplaced. pre-LLM software engineering already consisted almost entirely of building systems we did not understand

I have updated my tutorial on making Vision Language Action models. This tutorial starts with a basic Transformer and walks people through the steps to transform it into a full VLA that uses PaliGemma as the pretrained VLM. Links below.

Loki robotics is cleaning like a person. Doing the work we really want to see automated

An attempt to express how I principally use LLMs.

Rotating the Space: On LLMs as a Medium for Thought
sbgeoaiphd.github.io/rotating_the...

intentional grounding is an incomplete pass, so by definition there is no way to recover the ball

On Programming with Agents - Zed Blog From the Zed Blog: Agents handle typing so we can focus on thinking.

LLMs automate typing, not thinking.

Mikayla Maki shares her mental model for working with agents, and how she puts it into practice:
zed.dev/blog/on-prog...

Pantograph is building a "daycare for robots" where they can learn unsupervised. These low cost robot crabs are part of the solution. pantograph.com/blog/buildin...

a lotta yall still don’t get it

you can use multiple polecats on a single refinery

since LLMs save us time coding we have a lot more time to think

since we're thinking about things longer we realize most of our ideas are bad

and since most of our ideas are bad we don't ship as much code

Sharpa Announces CraftNet - A Hierarchical VTLA Model for Fine Manipul “We manufacture time by making robots useful.” At Sharpa, our mission is to build robots that are useful in our daily lives. We believe robots should help us handle repetitive, tedious, and sometimes ...

Oh cool they blogged about their architecture!

www.sharpa.com/blogs/news/s...

Do you know if it was teleoperated during the demo, or are you suggesting teleoperation during training was slow/methodical and therefore the policy replicating that behavior was also slow?

Can you help me understand what outputs you get with these long-lived loops? I don’t have much experience, but I imagine there would be sooo much content to review. How would you manage massive diffs effectively for production code?

Effective harnesses for long-running agents Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

I’m operating the same way (heavily human-gated), but I don’t think that’s how Boris and other “power” users operate. Anthropic posted about www.anthropic.com/engineering/... and also have a plugin that is a step in that direction: github.com/anthropics/c.... If only I had more time and credits…

The sodium content looks wrong (40 grams?!)

Just published my annual predictions update, tracking from Jan 1st 2018, with new commentary and new ten year predictions. It is long. rodneybrooks.com/predictions-...

The New York Times Digital Get unlimited access to The New York Times, including News, Games, Cooking, Wirecutter, and The Athletic.

In case you didn’t know, many public libraries allow you to redeem access codes for NYT - e.g. I use KCLS/Seattle Public Library and they have 1-day/3-day all access passes. It’s super easy to redeem, takes less than 30 seconds and is free! www.spl.org/books-and-me...

Yeah totally agree - I also think there’s a world where this growing DC demand drives significant investments in renewables and grid infrastructure🤞🏽

I missed what happened… sorry if I attracted the mob, that wasn’t my intent. I appreciate the link(s) you shared!

(2) But thank you for the link, this is a great report and I hope we see more like this. I still think there are valid criticisms / concerns about the future, especially with the lack of transparency from most companies.

(1) 0.1% is a snapshot; total DC emissions projected to hit 1-1.4% by 2030, with AI growing to half that load. Aviation is 2.5% and gets serious policy attention.

As for local impact, Colossus will consume 40% of Memphis’ peak demand; they currently run gas turbines because the grid can’t keep up.

I disagree, the environmental/energy impacts are certainly not made up nor negligible. I think there’s a lot of misinformation about those impacts (from both sides), but dismissing the total energy consumption used to train foundational models seems disingenuous.

I think you’re missing the environmental / energy impacts; that’s a big mode of criticism I’ve seen

That timeline doesn’t explain the job market contraction and increased competition for entry level jobs we saw after ZIRP. Perhaps the large tech companies are ahead of the curve and were preparing for this, but that doesn’t add up for smaller companies and startups