slopnt

what do you mean by this?

that has actually been proved to be the case. the weight tweaking being done by abliteration does change token distribution in some cases precisely making LLMs produce tokens it wouldn't have otherwise (and therefore less reflective of training data)

bun was probably proposed by an LLM whose training preceded August 20

It really really loves mentioning what something "is not" right after it just said what it is. I don't like this pattern and see it more and more in READMEs, probably part of recent Claude RL. I don't think its as prevalent in the zeitgeist as the emdash, "Its not X its Y" and the emoji bullet list.

Maybe by then there is no need for employees, and ASI is unleashed! /s

Given the fact that the article proves this can be accomplished with very little training, I believe it could at least be a possibility that this could happen unintentionally (or at least without their creators' knowledge) via adversarial training data.

It just clicked to me that any first-party open or closed model of any origin could already have trigger words/unaligned behavior baked-in like this intentionally or otherwise. Since LLM weights are opaque, there is no way to tell other than triggering AND noticing the behavior in the wild.

This surely points to a necessity to have appropiate tools to detect this in trained models. Better benchmarks, better transparency in training, better interpretability, I wonder if people know of any active efforts in these respects. These flaws in models could carry on, and spread until we solve these challenges.

yes, but that does not prevent other applications from existing or being successful this year

They have to have decision models already in production. Part of their business is detecting bots, DDoSers and spammers.

I like reduce in principle since it generalizes a simple concept pretty nicely. I don't use it that much in practice since its alternatives just require less brainpower. It competes against using local mutable state with a loop or iterator combinator which I would argue are easier to wrap your head around (i.e. loop with variable/map with closure). I would argue its one of those cases where something is just harder to do/understand in functional vs imperative programming.