Assoc. Prof. of Strategy, U Toronto Rotman | Chief Economist, CDL Toronto | Co-Founder, AllDayTA | Ars longa, vita brevis, occasio praeceps (especially now)
Toronto, ON, Canada
- Tweets 17,132
- Following 20
- Followers 24,379
- Likes 12,072
Kevin A. Bryan retweeted
It’s not enough to have open weight models for defense. We need all agents run from secure environments like OpenShell (with controls and monitoring) and clear agent credentials for any agent on the web, otherwise the default should be to block all agents. Anonymous agents should all be treated as malicious till we get a workable internet wide credential system like DiD or VC and clearer liability rules between users, enterprises, and platforms.
The point of this is that the balance of having open models help with cyber defense, and the trajectory we’re on being well more fine than people think.
It’s not that open weights won’t be used in cyber attacks. All possible AI trajectories had many cyber incidents imo.
Like many things I do, I’m trying to challenge the fatalistic, largely ai doom / ai safety discourse around cyber and open models, because there’s a credible part of the cyber community that thinks open models are a big part of holding things together in the next 12 months.
1
1
8
455
The problem for the Chinese models: how do you make money w/ open source if you don't have market power in complements a la Ali/Tencent/Xiaomi/Huawei? Not impossible, and Deepseek has unique advantages, but Kimi/Zhipu/Minimax?
🚨China's AI model trade is getting DESTROYED.
Bloomberg reports that MiniMax and Zhipu have erased a combined $150 BILLION from their peaks. MiniMax went from +650% above its IPO price in March to trading BELOW its IPO price. Zhipu has crashed 81% from its June high.
And the squeeze is getting worse. OpenAI and Anthropic have joined the token price war to narrow Chinese models' cost advantage, while MiniMax and Zhipu remain unprofitable. DeepSeek and Moonshot AI could IPO next year, and another 203 million pre-IPO MiniMax and Zhipu shares become eligible for trading in January.
1
206
What should journals do in AI world? I rec'd this to AEA ~two years ago, but more important today: 1) first pass technical review by AI, author makes any corrections they agree with, 2) 1-click replication, 3) open, AI-readable data, 4) formalized proofs (a la MathEconLib). 1/2
1
5
510
Chamley-Judd doesn't miss, kids. (Christophe was actually my Macro 101 teacher when I was 18!)
I don't think the macroeconomics of paying for new programs by taxing unrealized capital gains works.
slowboring.com/p/obamas-forg…
10
567
Fabulous! People think about things like "lists of three", but the reason it is so easy for many of us to pick out AI writing is more subtle, like the "unattested claim" and "inanimate subject, animate verb" patterns. CoT RL artifacts?
We collected matched responses from a range of frontier, near-frontier, and pre-frontier models in chat, agentic coding, and essay-writing settings, ranging from Llama-base to Claude Fable 5 and GPT-5.6 Sol.
2
5
525
Great benchmark...and you have to believe this will be saturated in a year given the pace of improvement.
HLE tested knowledge. ALE tested digital work. PLE tests the work that builds the physical world.
Today we’re releasing Physical-Engineering’s Last Exam (PLE) from Google Cloud AI Research.
All frontier models scored under 50% 🧵
6
969
The great @gideonlewiskrauss with a fabulous article on testing laboratories and how they work. What a fabulous opening! The New Yorker continues to have the best longform nonfiction on the planet...
1
3
1,069
And here is book in question from the review, by Alex Davies: amazon.com/Kobuk-Destroyer-O…
373
Completely agree. The fundamental Canadian problem with economic policy is that they don't believe in the market. It's always "let's just have one more government program." BDC is involved in a *shockingly* high fraction of all Canadian venture rounds, yet look at the returns...
Cut fucking taxes and let the market decide instead of you, who know absolutely nothing about the venture capital sector.
1
2
28
1,726
Kevin A. Bryan retweeted
New from @arcinstitute: using the latent space representation of genomes from single-cell life to find new candidate functional RNAs.
I think this is cool because: (1) it uses deep learning and model interpretability to productive ends on a totally different kind of "knowledge"; (2) it shows how agents may be able to accelerate discovery in vast combinatoric spaces; and (3) it's an example of the synergy we'll hopefully see in the coming years between computational techniques and wet lab biology.
Delighted to share a landmark preprint from our lab, using genome language models and AI agents for biological discovery and genome mining!
The stretches of DNA between protein-coding genes are often called the genome’s “dark matter,” but they are far from empty. They encode RNA molecules that sense metabolites, regulate genes, and control some of molecular biology's most important tools (such as programmable genome editors e.g. CRISPR/bridge)
Much of this biology is unexplored. One reason is that RNAs can retain similar structures and functions even as their DNA sequences change dramatically, making their relationships difficult to recognize with conventional computational tools
21
31
445
44,263
Kevin A. Bryan retweeted
We need to pay Congress people 3-5x more, and also have very strict penalties about accepting gifts, insider trading, or quid pro quo (like harsh felony terms). Hill staff should be paid more too - they’re pretty incredible and doing most of the hard work, esp the technocratic ones in wonky areas like energy, tech, AI, or science policy.
It's unpopular to say, but we now need to pay our politicians more.
Take the US House. Their salaries have been frozen since 2009. In real terms, they've gone from a very respectable $300k to barely half that, $174k. That's much less than many could make in the private sector.
1
4
750
Was discussing this with students today: Ethan's Opus 5.5-written country song about the Bitter Lesson is a surprisingly clear & technically accurate explanation of Sutton's Bitter Lesson and where people are still not "Bitter Lesson-pilled". *The* idea to understand in the 2020s
People asked for me to upload the Bitter Lesson video from this piece to YouTube, so here you are: piped.video/OAwat51S_Sk
1
4
2,093
Really interesting point - as AI improves, you may want to introduce necessary frictions, such as time-based ones, into irreversible actions. I can see how you model this in a mechanism design sense - has anyone done this?
Just want to point out that if this were crypto accounts, your money would be gone forever. But for bank accounts, that take several days to clear payments, this is probably a huge pain for the IT and back-end teams, but for depositors it’s likely a shrug.
3
21
2,089
Econ of science folks - this position via FAS is really interesting if you want policy impact. Essentially, it's a metascience/econ of science position embedded quite high up at places like DOE. They allocate *billions* to R&D, w/ big changes due to AI/Genesis/OSTP doc/etc.
2
22
49
7,923
Nov 9 deadline, for midcareer folks (great sabbatical option!), US citizenship. Btw, for Canada & others, US is actually very good at embedding academics and private sector folks on career leave in specialized parts of the civil service- worth replicating! fas.org/career/metascience-f…
4
463
This is a great demo of where videogen is now. When it comes to fraud, I find a lot of people don't realize that photorealistic, voice-matched video calls are trivial to generate. We will get but don't yet have new "trusted institutions" to handle this...
3
16
1,826
On tech power: folks, use backward induction. Elon did not unilaterally cut Starlink capabilities in Ukraine w/o DC knowing! This was obviously DJT/Hegseth, who don't read anything, asking for it due to the Hubbard situation, not realizing it was critical defensively also.
1
11
1,668