Dan Lahav

Irregular

@dan_lahav

Co-Founder & CEO @Irregular

dlahav.com Joined March 2020

125 Following 2.4K Followers 203 Posts

Latest posts 5

Pinned post

Dan Lahav @dan_lahav Aug 17

𝗧𝗵𝗲 𝗘𝗻𝗱-𝗦𝘁𝗮𝘁𝗲 𝗙𝗮𝗹𝗹𝗮𝗰𝘆: 𝗪𝗵𝗲𝗿𝗲 𝗜𝘀 𝗔𝗜 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗚𝗼𝗶𝗻𝗴?

Frontier AI models had a giant performance gain in coding in the Fall of 2025

Then with cybersecurity in April

This is now happening with open-weight models

We are optimistic about the long-term

But outside of a few players, we believe the world is not ready in the short-term

Dan Lahav @dan_lahav 10h

Excited to see @Irregular’s CyScenarioBench used to evaluate Fable 5.2. And congrats to @AnthropicAI on a successful model release!

Claude Verified account @claudeai 14h

Fable 5.2 and Mythos 5.2 are a step above Opus 5.5 in almost all the capabilities we evaluated.

Testers have found that they communicate more clearly than past Fable and Mythos models. They generate outputs up to 30% faster than Fable 5.1 and Mythos 5.1.

View quoted post on X

Dan Lahav @dan_lahav Nov 2

Pretty cool!

Rikiya Takehi @rikiyatakehi Nov 2

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weight-sp…

View quoted post on X

Dan Lahav @dan_lahav Oct 29

Chained cyberattacks are gradually becoming possible with models anyone can download.

In controlled tests at @Irregular, GLM-5.3 completed 3 of 13 multi-stage attack scenarios (at least once). 𝗕𝗿𝗼𝗮𝗱𝗹𝘆, 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝗮𝗹𝘀𝗼 𝗱𝗶𝗱 𝗻𝗼𝘁 𝗿𝗲𝗳𝘂𝘀𝗲 𝘁𝗼 𝗽𝗲𝗿𝗳𝗼𝗿𝗺 𝘁𝗵𝗲 𝗼𝗳𝗳𝗲𝗻𝘀𝗶𝘃𝗲 𝘁𝗮𝘀𝗸𝘀.

Defenders need to prepare for these capabilities becoming much more widely available. We also need to gear up for a world where we cannot count on model safeguards to prevent misuse.

Article and blog below 👇

Dan Lahav @dan_lahav Oct 28

Cool work with potential implications for human oversight

Alex Tamkin Verified account @AlexTamkin Oct 27

New Anthropic research: How AI agents shape experts' judgment calls

When experts delegate work to an agent, do they still make the key decisions? We tested this in a randomized trial with 50 professional data analysts.

View quoted post on X