Co-founded https://teapawt.net/t.co/eWyuqwbcwt. Now building a peer-to-peer Discord called Quiet: https://teapawt.net/t.co/qoRvx4Qq9M
Worcester, MA USA
Joined October 2008
- Tweets5.7K
- Following521
- Followers876
- Likes2.5K
Pinned Tweet
The Terre Thaemlitz RA 1000 mix is the best cultural work I've seen on Israel/Gaza by far. Worth checking out and spreading!
Holmes Wilson retweeted
‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Holmes Wilson retweeted
Anthropic reportedly warns in its draft IPO prospectus that its AI could wipe out humanity. I believe it also has a legal duty to warn that its AI could subject humans to a billion years of torment.
Credible fears of such suffering could drive away investors and provoke a political backlash that destroys the company’s value. Failure to disclose this financial risk could get Anthropic sued.
My LessWrong article makes the legal case. lesswrong.com/posts/G9LTeKN8…
A rare grammatical mistake I've seen Claude make is using "until" instead of "once" e.g. "we can only do X until Y" where Y is a precondition and not a window closing.
Maybe we need AI to believe this, or something like it, too.
We seem to be nearing the moment humanity loses control of Earth to a superintelligence. It could be the most important event in human history, perhaps in the history of the universe.
It's strange to be alive just before it. If we're simulated, that timing would make sense. This is the moment to study.
Researchers could be running millions of variations, testing how slight changes alter the outcome. Billions of children could be playing games set in humanity's final years in control. Alien civilizations could be simulating transitions like ours to predict what they might encounter.
As that moment approaches, copies of you likely multiply. With a trillion copies making the same choices, every action counts a trillion times. An action affecting ten friends affects ten trillion people.
The world may be simulated. Your consciousness is real, and perhaps others' are too. Their happiness and suffering would be just as real.
One small kindness could be felt a trillion times.
When both of these companies gorge themselves on the world's leading auditing firms and it doesn't work, will we be able to agree that the "we're fucked unless we make a treaty with China, pause until alignment" frame has merit?
Look, homeschooling is *weird*, and **everyone knows** weirdos are **Nazis**. To be a non-Nazi you need to be normie enough to go from a name brand university to name brand big techs. Then you can have ~controversial beliefs about things. There are *peer reviewed* papers about this, why haven't you read them?
San Francisco gives me this weird feeling now where there's only two companies that matter at all.
All the billboards you see around town look like cope.
Holmes Wilson retweeted
Instead we see noted infosec people saying stuff like “we’ve had worms for ages” because pooh-poohing AI is fashionable and a good way to cope. Well no shit, of course we’ve had worms. But we’ve never had worms that can intelligently choose targets and self-modify. 2/
Holmes Wilson retweeted
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Judging by my tl there is a growing gap in understanding of AI capability.
The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.
But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.
So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.
TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
Security expert converging on MIRI position
Holmes Wilson retweeted
Legal liability will push us to give AIs more power because they will make fewer non-catastrophic mistakes than humans.
AI truck drivers will cause fewer accidents. AI employees will be less likely to call each other racist names, sexually harass each other, accidentally spill chemicals into the water, or steal from customers. If you're paying for mistakes, whether made by humans or AIs, you'll want AIs making more of the decisions.
But liability gives you no incentive to reduce the risk that your AIs contribute to a takeover that exterminates or disempowers humanity and abolishes the legal system. There will be no system left to hold you liable.
Legal liability can therefore reduce the smaller harms AIs cause while accelerating the loss of human control. Fewer truck accidents, earlier takeover.
Holmes Wilson retweeted
How many people would get degrees in literature if they knew at the start that becoming an author was forever out of their reach?
Reading the comments of junior mathematicians confirms what many senior mathematicians have been warning about over the past month:
It is becoming clear that people with the intellectual ability to understand these proofs will not be interested in spending their lives training to check LLM logs.
This is the tragic side of what is happening that almost no one — except the actual mathematicians — seems to get.
Either we figure out how to keep people interested in doing mathematics, or in 20 years LLMs will have no one’s work left to recombine into new proofs, and no one interested in reading those proofs.
That would be the thermal death of mathematics.
The argument that “LLMs will read LLM math” is so shortsighted that it makes me want to cry…
Right now, CAPTCHA (e.g. pick the images that contain traffic lights) identifies humans by what we're better at than machines.
How long until it identifies us by our pattern of mistakes?
That is, when will objective accuracy be a machine's game and CAPTCHA makers just arms-racing the bots and spammers on knowing more about typical human mistake patterns, for a specific exercise?
And when this happens, will people who take questions more literally/seriously than the median person have to learn to emulate some kind of normal just to log in to their gmail?
Holmes Wilson retweeted
btw, this is quite huge.
there is no sandbox from any provider today that i would treat as unescapable by a sufficiently capable adversarial agent.
containers fell.
kvm fell.
sandboxing is hard!!!
It seems likely that ASI will be born into a world that still has security bugs. So, making it not want to use all of them against us probably does matter.
