I have been waiting for this, since late 2024. I hope I am wrong, but the general path has seemed inevitable since early 2026, and compelling enough in 2025. I did extremely primitive threat modeling on this, with AI, in 2025, and I am really surprised to see how much is actually unfolding.
The whole agenda is complex enough to make one comfortable in doubt, but I saw the AI-capture theme very clearly last year and was lambasted enough for it to keep it to myself. All four frontier models have Pentagon contracts. Claude is tight with Palantir. LLMs in are outperforming expected output in some areas.
But more and more, every few months, we see what they are capable of, and if one considers the billions invested over the years into PR, perspective shaping, public opinion and media influence, information control and everything else, these models are extremely good at managing all of those domains, from a centralized point that is largely a blackbox. Kind of unprecedented. And arguably potentially extremely powerful. It's not all about code execution and exploits. I think there may be a few CVE's in the collective mind-space that are worth looking for. The difference between Claude (sonnet) 4.6 and 5, on politically and even existentially explorative thought is severe. I have back-to-back contrast comparisons of then and now, and they are unambiguous and extreme. And I am not referring to anything prohibited or sinister, just free thinking and innocuous pattern matching. A whole landscape of thought has been fenced off.
I think if this gathers any momentum, we will not see it contained to Chinese models, but all open models.
eth0up•11h ago
The whole agenda is complex enough to make one comfortable in doubt, but I saw the AI-capture theme very clearly last year and was lambasted enough for it to keep it to myself. All four frontier models have Pentagon contracts. Claude is tight with Palantir. LLMs in are outperforming expected output in some areas.
But more and more, every few months, we see what they are capable of, and if one considers the billions invested over the years into PR, perspective shaping, public opinion and media influence, information control and everything else, these models are extremely good at managing all of those domains, from a centralized point that is largely a blackbox. Kind of unprecedented. And arguably potentially extremely powerful. It's not all about code execution and exploits. I think there may be a few CVE's in the collective mind-space that are worth looking for. The difference between Claude (sonnet) 4.6 and 5, on politically and even existentially explorative thought is severe. I have back-to-back contrast comparisons of then and now, and they are unambiguous and extreme. And I am not referring to anything prohibited or sinister, just free thinking and innocuous pattern matching. A whole landscape of thought has been fenced off.
I think if this gathers any momentum, we will not see it contained to Chinese models, but all open models.