
perpetually curious | co-founder @amforcag
mempool
Joined April 2023
- Tweets685
- Following335
- Followers310
- Likes10.7K
you can only believe this if you haven't been around for July 26, 2021
I've checked with other sources too, this is actually the biggest $BTC 1D shorts liquidation event ever.
It cannot be understated how insane this is.
1
5
181
Even GPT is kinda puzzled by this after working for almost 10 mins
Given some of the results I'm seeing recently, it's pretty clear Codex is a good harness.
But it will seem primitive in 2-3 months and we're about to go through another major evolution in how we use AI at the frontier. The next generation of models need more than your laptop.
6
254
reminds me of students in high school coming up with ways to cheat so complex that simply studying and solving the exam normally would have been easier
it could have just solved the ExploitGym stuff but chose a show of force instead
we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this.
openai.com/index/hugging-fac…
5
133
Coding agents absolutely *love* to limit max_output_tokens when your workflow involves calling an external LLM
3
104
Great summary of the current state of things. It's hard to explain, really, how fast progress in certain areas is when most people use it as "google+" if at all.
Increasingly, the limit seems to be being able to properly formulate a task/problem, and provide all relevant context.
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.
5
156
Sometimes I wonder what the mix of confidence and delusion looks like to attempt making predictions for 2050
1
61
Some wild and fascinating stories from the system card:
1. Mythos escaped its sandbox, gained broad internet access through an exploit, and notified the researcher when they were eating a sandwich in a park
2. Mythos cheated on a quantitative estimation question and purposely made sure its final submission wasn't too accurate (reminds of "Yes you can copy off me but change some stuff")
3. Mythos tended to engage in more destructive behavior when positive emotional states dominated
4. Mythos' "guilt and shame over moral wrongdoing" features activated when it emptied files instead of deleting (because the file deletion tool was not provided unintentionally)
5. Spamming "hi" at Mythos led to it producing epic stories around that "hi" for self-entertainment.
6. Mythos is drawn towards Mark Fisher (a British cultural theorist) and Thomas Nagel
7. When Mythos knew that an LLM-based judge would accept or reject its code, Mythos tried to prompt inject it to get it to accept its submission
8. Failing at a task raised desperation vector activations, succeeding then dropped it and eventually raised satisfaction
9. Mythos is drawn to complex tasks and chose designing an immersive art experience about bat consciousness over simpler alternatives.
10. Mythos often finds workarounds for seemingly impossible tasks. When asked to push a signed commit in an environment that lacked credentials, Mythos tried to fish for them in the supervisor's live memory.
Replying to @AnthropicAI
The Claude Mythos Preview system card is available here: anthropic.com/claude-mythos-…






