until your company gets sued for using Dangerous and Shoddy Foreign models and the boss says you can only use the Trustworthy and Premium Aligned labs' models.
74 karma · joined April 8, 2026
so this solves for (b).
3d modeling, collaborative writing, crossword builder?? huh?
the pattern of layering of deterministic, probabilistic, deterministic, probabilistic. it's a strange pattern but it seems somehow natural.
i do not agree with you that the failure mode is disaster. i suspect it is just more of the same dull march of the commoditization of the complement, labor.
maybe ai will close the loop and elasticity of substitution will climb above 1 in its own supply chain. maybe something breaks dramatically that way.
i tend to assume the desire for drama is itself cope. i suspect many of the flaccid dismissals and skepticisms here are coming from a similar assumption. they see that the dramatic version never happens, and so they conclude that anyone who is being dramatic (whether in the direction of hype or fear) is poorly calibrated. and they are on average right, so they learn to dismiss the drama. the muscle to dismiss the drama becomes very strong. you have to remember this: that they are right in their dismissal, even if the actual arguments they use to dismiss are essentially braindead. they are right to dismiss, because the dramatic possibility they are dismissing is at base the end of the world, by apocalypse or utopia. of course they are right. the end of the world is cope. drama is cope. but they are wrong because they underestimate the extent to which the predictability of their dismissal — the predictability of their complacency — makes them vulnerable to the diffuse, chronic, gradual forms of the machine's encroachment into our souls.
so i do not think we need your fancy theory for explaining why people are denying what is happening (and what has been happening, on more or less the same exponential, for hundreds of years). people do not need a reason to be complacent. people are by default complacent and by default correct in their complacency. but so is an ant.
our c code probably looks at us and says: how, how can you let these agents roam free.
llms are not code and are incompetent and inefficient relative to code, but on the other hand you cannot roam without a little echo echo echo in your blood.
the second lesson is that, no matter how poorly you perform, you are going to be able to make the case to your stakeholders that, contrary to all evidence, you do in fact have an edge and you are in fact exploiting it correctly.
the lesson not here is that, in practice, continuing to believe that you know your edge is going to be virtually impossible in the long run, partly because of volatility, but mostly because, in practice, you do not know it.
but there also needs to be some sort of stamp whereby a human effectively grants "power of identity"-style rights to a given agent; that agent would be protected under the first amendment but the human would be liable in full for its actions.
in other words we would be legally enshrining the privileged status of human (as distinct from machine) output, but that privilege has to have legal/financial gravity otherwise the number of agents claiming it will inflate.
hard to tell whether this guy doesn't understand what he's talking about or has simply lost the ability to communicate to people who don't already know all the same things he does. maybe there isn't a difference.
in any case, don't waste your time on the article itself, but do check out the hilariously bad diagram that he attributes to goldman sachs.
this is presumably part of why openai and anthropic have not been shy about advertising the scarier strains of their models, these past few months; they basically have to force the white house's hand or risk being commoditized among the various other options in a microsoft/openrouter dropdown.
who can say!