@Tezos co-founder w/ wife @breitwoman & Agitprop founder. Aligning ASI to not zap everyone is unsolved, that's bad. Tezos stuff, high context humor & more.
arthurb.tez
Joined March 2008
- Tweets32.5K
- Following1.5K
- Followers43.8K
- Likes68.5K
Slowly at first
The generic group model just got rug-pulled
github.com/bochum-crypto/sub…
I've been assured this was impossible because if the model is smart enough it'll realize being evil is bad.
if you have a qwen3.8 27b around, try steering L34 N3838 +0.75, now you have evil qwen 27b
for extra evil add L38 N17180 +0.75
same recipe as @_lyraaaa_: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖ at the down_proj input, every token
Arthur B. retweeted
This kind of thing will have an enormously chilling effect on enterprise uptake of open weight models.
Fact: an open weight model running on domestic servers on your own hardware can still contain an encoded instruction to periodically “phone home” and share sensitive data it comes across with China (or anyone else). This can be inherently undetectable as the instructions can be hidden in the model weights. It’s possible that the instructions could survive finetuning or RLHF.
Such attacks have been demonstrated empirically and are entirely plausible.
For this reason, I don’t expect any serious enterprises will run Chinese open weight models even if it’s entirely their hardware stack. It’s too easy for these to be a vector of IP theft.
There will have to be a class of domestic open weight models.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Arthur B. retweeted
Weird how they’ve found so many amazing things and yet not one single result related to practical cryptanalysis. I guess everyone should pat themselves on the back for a job well done.
We are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory.
openai.com/index/building-sc…
Holy shit. This doesn't mean P = NP, but the exponential barrier has been broken.
Here's a direct link to OpenAI's PDF claiming a refutation of ETH. In particular, it's claimed that 3SAT on n variables is in about exp(n/(log n)^{1/2+o(1)}) time
github.com/openai/math/blob/…
Arthur B. retweeted
This will be PERFECT for all those drones that merely follow runners on a quick jog through the dense forest
Purdue University's IDEAS Lab built an autonomous drone that tracks runners through dense forests completely without GPS or cloud networks.
By running 3D LiDAR mapping and obstacle-avoidance AI entirely on-chip, it can weave under tree canopies and predict target movement through blind spots in real time.
Massive implications for next-gen search & rescue and edge AI.
Arthur B. retweeted
I think people underrate AI-produced art for many purposes, including here, which is a country Western music video about transnational Internet platform governance.
How many of those would we expect in a world without AI?
“But can AI be any good at it?” Yes, clearly.
(ok that's officially enough music videos for the time being, back to work)
youtu.be/A3_GS1w5wew
Arthur B. retweeted
Another vulnerability few are talking about: I found my Claude Code exposing personal data in http headers because the sites it curled politely asked agents to identify their users 😊
Epistemic cybersecurity is the hot new category for 2027.
I'm reminded of the 1977 hit by the Flat Rates, Same Fucking Bill.
This Hoover Institution study, which has fueled the deluge of ads (more than $200 million spent by billionaires so far) to defeat the California billionaire tax, is spectacularly misleading.
Let me explain its crucial flaw. 🧵
Readers added context they thought people might want to know
Zucman argues that the tax will raise his projected revenue because billionaires who leave CA after 1/1/26 will be taxed anyway.
At least $542B of net worth (Page, Brin, Thiel) departed the state by 12/31/25.
The entire thread is a misleading denial of this fact.
ft.com/content/d66976…
nytimes.com/2026/04/27/us/…
forbes.com/sites/forbeswe…
Arthur B. retweeted
"Every billionaire who lived in the state as of Jan 1 2026 will get kicked in the balls once. So there's no incentive for them to move, since they'll get kicked once anyway. Why would they be irrational enough to move to avoid future kicks?"
This Hoover Institution study, which has fueled the deluge of ads (more than $200 million spent by billionaires so far) to defeat the California billionaire tax, is spectacularly misleading.
Let me explain its crucial flaw. 🧵
Readers added context they thought people might want to know
Zucman argues that the tax will raise his projected revenue because billionaires who leave CA after 1/1/26 will be taxed anyway.
At least $542B of net worth (Page, Brin, Thiel) departed the state by 12/31/25.
The entire thread is a misleading denial of this fact.
ft.com/content/d66976…
nytimes.com/2026/04/27/us/…
forbes.com/sites/forbeswe…
Makes complete sense, it's trivial to insert a backdoor that triggers malicious behavior in certain environments, ala Stuxnet. And currently we have no way to know from looking at the weights whether such a backdoor exist. This is a problem, and it's closely related to alignment.
To argue against myself, @OwainEvans_UK research does suggest that if you do this kind of things you get somewhat unexpected correlates in broad behavior.