@kiki_morozova's Black Hat Europe talk on our image scaling research is on YouTube now! If you liked Anamorpher, check it out for more cool image preprocessing exploits-- plus some fun audio ones too! piped.video/watch?v=rHvFGz7_…

Lmao Anthropic used my photo without credit/permission, never beating the stealing allegations. They’re in the middle of MULTIPLE LAWSUITS for exactly this, yet do not care.

A twitter thread catalog of lesser known non-fiction-esque comics.

I am on Signal at sharongoldman.43

Test-time communication looks like a next axis for scaling capabilities New paper with the incredible @jon_ghoh and @vkontonis @ShivamGarg91462 and Akshay : arxiv.org/pdf/2609.21032 The Hugging Face incident showed when agents can find a channel they'll use the heck out of it. A useful question, I think is: when does communication make a group MORE CAPABLE than the same agents working alone? Aka is Team-of-N better than Best-of-N, when, and why? We had N identical agents work on the same task with no prescribed roles, using only a shared log (i.e, text file) and telling them to "collaborate". Across three "researchy" tasks communicating teams beat the heck out of independent agents: - On ARC-AGI-3, a Team-of-5 sonnet-4.6 agents matches Best-of-33, and can for example solve a game 65% of the time that no single agent cracked in 64 tries. - On polyomino packing (pack Tetris like pieces into the smallest rectangle, cf Frontier-CS by @eigenlabs), a Team-of-3 Opus 4.6 agents surpasses best-of-60 and set, as far as i understand, a new record for that benchmark. - On MNIST compression, a team of four 5.6-Sol agents find a 1,957 byte model with 99.4% accuracy, which btw is 20% smaller than the best human solution (on a problem beaten to death!!), while no independent agent gets below 3KB. The mechanism is a bit obvious in hindsight: when one agent finds a clearly better partial solution, it broadcasts it, and everyone immediately builds on it. Why? A single lonely agent must make every breakthrough itself, yet a team needs each insight only once, found by any member. That is kinda like comparing a minimum of sum of "time to n-th breakthrough" vs a sum of minimum of "time to n-th breakthrough". That gap can grow exponentially with the number of "breakthroughs" needed to arrive at a solution. We worked on this because prior work (before the hf incident) suggests unclear benefits for communicating aganets. Which is true, when the tasks are inherently serial (duh), eg some Terminal bench style tasks. Yet feels it should not be true for research problems. Indeed for research heavy problems... Test-time communication seems like a new capabilities axis. I'm sure we will see a ton more of it!

When I shared @harvey’s model strategy a few months ago, there were two parts: 1. Build our own model 2. Use that to help customers do the same We’ve done the first. Now we’re hiring for the second: Harvey’s Private Model Program. We’re seeing huge demand from law firms to own their own intelligence leveraging private data. This will enable them to become frontier firms that get smarter with every client matter. You’ll:
 - Partner with law firms to build these systems. - Build the team that delivers this at scale.
 - Work with our technical org to define the platform that powers this team. We’re looking for a technical PM or founder type to own Harvey’s Private Model Program. DM me if this sounds interesting.

Don't y'all dare submit sloppy differential privacy papers to ICLR

I put my @UnpromptedAU slides up at justdionysus.github.io/slide… — a bit of reflection on exploit development in the age of AI. My TL;DR is keep pushing to understand complex things, be honest with your own understanding, and use AI as a power tool to increase pace and depth.

“I used my agent to formally verify my software and it fixed all these bugs” is the new “I asked an LLM judge to fix my LLM output”. The devil is in the details and it is insanely difficult to get the formalisms “right” (eg interpretable, expressive, extensible) for humans and agents to co-create software. Seems that we are in for an entirely new age of learning how to build software

i'm back on the job market again - golang - mid level role - location flexible ⏰ pls rt for help ⏰

This is brave but necessary. There is little incentive to calling out poor scholarship by professional colleagues, it's gonna make people upset. But to maintain standards in our community we have to hold each other accountable I encourage others to call out slop when they see it

model evasion attacks are back and better than ever fr fr

People who submit slop papers should be banned from submitting papers in the future. We don’t need bad faith actors in the academic community.

bullish af on jev now

Never underestimate AI security experts. Put two of them together, and even if they're air-gapped, they'll find a way to exchange information through farts.

I'd say that this is really a failure of actual defense-in-depth and layered defenses. A good metric is how many 0day exploits were required by the end-to-end attack path from attacker to goal. The goal of security architecture is to increase that number as inexpensively for defender as possible.

Excited to try Jev in parts of our systems that need calibrated probabilities for categorical decisions. We currently use a hacky version of this idea: small LLM classifiers for routing, citations, parts of Vault, tool use, and user escalation. One challenge is that LLM softmax probabilities aren’t necessarily calibrated confidence estimates. It will be interesting to see how RLCD improves calibration over the naive approach. Jev doesn’t generate text, so its “hallucination-free” framing isn’t a full solution to hallucinations. But better routing, citation selection, and escalation could reduce hallucinations across the broader system. Longer term applications for law firms include matter selection, associate staffing, and predicting billing disputes. Also excited to see open-source implementation of RLCD so we can post-train these models ourselves.

This really pisses me off. Security experts have been telling OpenAI that their security is negligent for years. Recent hacks aren’t examples of “people” underestimating AI, they’re examples of Noam and his colleagues doing it. Or of corporate prioritizing $ over safety.