My favourite genre, incoherent Claude-generated slop about how Claude is wrong, with an appetizer of self-aggrandizement to boot.
The Permission Slip
11–14 of 14 posts
Re: The Permission Slip
#12If “scale will solve everything”, even (as the article contends) things that could be solved more cheaply in other ways, that’s of course wasteful and inefficient. But what about things that only scale can achieve? Like the superhuman security vulnerability assessment capabilities that Fable showed? That would be a reason to continue to spend, wouldn’t it?
Do we know that Fable/Mythos was the result of throwing more hardware & data at it? Anthropic is still pretty compute constrained. More does not always equal better, it very well could have been Fable/Mythos came as a result of better data curation or some other break through, not necessary more parameters & compute.
I don't think "just throw more compute at it forever" is the only way to go, but if that turns out to be true, the labs aren't going to share that knowledge because that would be a risk to the dump trucks of cash getting dumped at their feet if they came out and said "You know, we don't really need much more compute, we found a better way to make a smarter model" the cash would slow down.
Re: The Permission Slip
#13My favourite genre, incoherent Claude-generated slop about how Claude is wrong, with an appetizer of self-aggrandizement to boot.
I've been following Bob Cringely for nearly 30 years and he's always written like this.
This is almost certainly not AI-generated text, and he's not saying Claude is wrong.
He's saying something different that is probably true (that LLMs can't produce facts, no matter how much they are trained). It is also not that interesting (LLMs hallucinating sometimes doesn't mean they're not useful).
I'll grant you the self-aggrandizement, but he's come by it honestly.
Re: The Permission Slip
#14Earlier quoted context omitted.
> it can work through the complex, vague machinery of reason with more scaffolding No, it can hold more floating point numbers. I'm not an expert in the field, but I've yet to see a solid rebuttal to this paper; https://arxiv.org/abs/2401.11817
A claim that LLM's can in a theoretical sense be 100% accurate all the time is not the same as the claim that scaling models with more compute/params will reduce hallucination. The former is a far stronger claim and I agree with the paper in that it probably isn't the case, but we don't rely on general reasoners (a.k.a. humans) to be 100% accurate all the time either. > No, it can hold more floating point numbers. Fa…
You're the one who started with the "complex, vague machinery of reason with more scaffolding" here. I'm simply pointing out that that's not actually a thing: it's just floating point numbers.