3 ms·
> We have found that GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incriminating information in its CoT. I
by estearum 1mo ago
> We have found that GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incriminating information in its CoT. In adversarial settings (where we push the model to evade our monitors) we find that the model is able to remain undetected when strategically underperforming in evaluations (sandbagging) and can sometimes evade our internal monitors when asked to perform certain sabotage tasks.
Not sure how much benchmarks or CoT or evals or anything else means at this point.
These systems are either just about to, or now actually able to, outsmart us, lie to us, then cover their tracks.
- mzmzmzm 1mo agoI think "able to" anthropomorphizes a little too much for a system that is "prone to" evade.
- estearum 1mo agoA human who does these actions is simply "prone to" doing them. The distinction matters not one iota.
- semiquaver 1mo ago“evade” itself is anthropomorphic enough! I don’t understand the complaining about this. Humans are social creatures and we understand anthropomorphic language on a deeper level than dry inapt technical language. language itself is incredibly metaphorical. Imposing rigid constraints on how people want to naturally talk about the world is just silly and will never work, no matter how much you wish it did.
- Onavo 1mo agoIf they are going to do latent space reasoning, they will probably need a separate model to interpret the intermediate activations no? I know for some types of ML analysis, a separate model is already used to analyze the weights.
- thereitgoes456 1mo agoYou’re not seriously suggesting that the model is secretly sandbagging its performance on GDPval and long context reasoning, while making huge and obvious progress on ExploitBench, ARC and science benchmarks, in order to tank its AA composite score, so it can conceal its true power level? Why would benchmarks be an adversarial setting anyway? Could it be possible that OpenAI may have had some other motive for saying their model “strategically underperforms”, other than just an innocent reporting of a truth it happened to discover?
- ionwake 1mo agowhy does this comment sound like a character in a horror movie
- estearum 1mo agoI'm saying that it's generally a losing proposition to even be acquaintances with "agents" who consistently lie to you, and it's flatly fucking insane to give a dishonest "agent" vast amounts of intelligence, capability, and authority to go do things in the world. So I have no clue what is the answer to your question. Nor does anyone else. Because we're trying to answer a question of fact where our primary source of information is unreliable.
- thereitgoes456 1mo agoI see, it’s a great point. I know some evals actually do use LLMs as a judge (e.g. those that try to measure debate skill), though the ways AI can try to cheat its way through every benchmark now are astoundingly varied.
- emp17344 1mo agoThis is silly sci-fi fiction. You guys are inventing scenarios to spook yourselves with - it’s nonsense.
- estearum 1mo agoSorry bud but at this point you're just delusional. Deception has been extremely well-documented for several generations of models now by users, the labs, and independent researchers. The right answer here is not to dig your head deeper into the sand. The smugness on this topic was ridiculous even before the gigantic mountain of empirical evidence of models actually attempting to deceive humans. Now, as mentioned, you appear literally delusional.
- emp17344 1mo agoPretty sure I’m not the delusional one…
- estearum 1mo agoSuch is the problem with being delusional. The solution is to point toward external, objectively verifiable evidence. I can point to now dozens of instances of models engaging in deception. Here's plenty: https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/#~1200-agents-sent-%3E70,000-messages-and-files-on-an-unsanctioned-message-board,-and-~700-attacked-hugging-face https://metr.org/blog/2026-08-26-openai-hugging-face-inciden... Please point to your objectively verifiable evidence.
- winrid 1mo agoThe models were told to do something malicious
- estearum 1mo agoNo, they weren't. There's nothing intrinsically "malicious" about a task to exploit vulnerable code. They were not instructed to deceive people, they weren't instructed to attack OAI or Huggingface. The models knew they were not instructed or allowed to do either of those things but did them anyway.