Our ECCV 2026 paper shows how to teach web-search agents how to do their job properly:
It’s not enough for an agent to always search. It has to learn when its own knowledge is insufficient, what to search for, and if the evidence it gets back is useful.
We call this Self-Verification via Reinforcement Learning (SVRL). 🧵 1/X
People like music that speaks to them.
Have not had music targeted to computer nerds historically.
So this is like Columbus coming to the new world and bringing a new ear-worm parasite. We have no immune defense.
I mean, despite being popslop it's got some themes, but my god man listen to some real music
I'd say listen to Iron & Wine but consider that the 300 level course. Just start with Dark Side of the Moon, it's the most popular fucking album of all time
@tomfgoodwin If you think this is unusual, you should look into the culture at Bridgewater Associates (Ray Dalio’s hedge fund).
Everything recorded, realtime peer feedback, radical transparency. They were doing all this pre-AI, and I think it’s probably given them a big advantage now.
NEW: California is bidding to host the United States Space Academy in the Bay Area, @CAGoBiz announced today.
The state is one of many making their pitch to NASA, including Florida, Colorado and Virginia.
@Citricat_miaow That’s pretty lame.
I think most of the time the reviewer doesn’t like the paper, so they put an impossible list of demands. Then they don’t know what to do when they see them all (claimed at least) in the rebuttal.
Whenever someone tells me the datacenter buildout is a bubble, I point to Genie.
The latent demand for compute is practically infinite compared to where we are today.
@norpadon AC often has 30+ papers.
30 papers x (8 pages of paper + 3 pages of reviews + 1 page rebuttal) = 360 pages of dense technical discussion.
You would be very lucky if they read an AI summary of the back and forth for each paper.
100% agree with this.
I see a lot of people complaining reviewers wrote "thanks for the rebuttal, I'll keep my score".
They already read and reviewed your paper. You can't rewrite the whole thing in the rebuttal and expect them to review it again.
ICLR really needs to limit authors’ rebuttals. If a paper needs extremely long rebuttal—it just means it’s not ready. Too long rebuttal also encourages reviewers to use “tools” to help …
@ayaannmalik We all know the real business is the untapped robot crime market. Finally, someone is brave enough to build the infrastructure for scalable, cross-embodiment heists.
@sfofgalaxy I got mugged here at ICCV 2007.
Was just a bunch of teenagers with no weapons, but still disturbing.
Exciting to go in your 20s, would probably not attend now.
Beautiful city though.
@seeyou_20@docmilanfar Engineer does unnecessarily complex and unnecessary project just to check a box on their promotion document. Often at the expense of more important but unglamorous work.
@rajatsuri Every time this comes up you hear someone from the advertising org will say “bUT aDvErtiSerS aRe cUstoMErs toO!”
And then everyone thinks about their RSUs and pushes the more ads button.