Benjamin Riley

Don't get played

Counterpoint: farts

OMG the sheer comedic value of having a function built in to ChatGPT called "confession," combined with our anti-AI Pope, is too much for me to fathom

Controversial to whom, though? 18 months ago, I could offer a bevy of AI engineers and Big Tech CEOs claiming "scale is all we need" to achieve AGI. Nowadays that specific argument is much harder to find.

Thank you for the kind words!

(whispers) I'm on a mission to revive Rorty to halt AI hype.

Large Language Mistake Current AI models are not on the path to artificial general intelligence

Thanks so much, Eryk. Direct link to non-paywalled version can be found here: buildcognitiveresonance.substack.com/p/large-lang...

There is no Artificial Irony Large-language models are dead-metaphor machines

You're onto something deep I think. I explored the idea of "artificial irony," and why LLMs will never achieve it, in more detail here:

open.substack.com/pub/buildcog...

Ok man, I engaged with you thoughtfully, I think you've just shown your true colors. If you want to turn this direction you'll get blocked.

As this judge rightly notes, what's definitely true is that LLMs do not engage in the metacognitive processes that humans engage in. If we want to label that process reasoning, LLMs do not do it.

There's no way to useful answer that question in binary yes/no fashion because there's no singularly applicable definition of what constitutes "reasoning." There's a document on my website that goes into this briefly (education hazards of generative AI).

The process they use to produce their output is not analogous to human reasoning (as evidenced in part by the fact that humans can reason without using language).

My argument isn't focused on CoT reasoning, it's about whether LLMs are on the path to AGI. It was cited in this case because a lawyer used ChatGPT in filing a brief and (unsurprisingly) the tool made up things that weren't true.

Love that article.

It's really good. I'm gonna track down the full opinion.

As always you make an interesting point, since I don't think my argument is all that controversial to many of the leading AI scientists. But there's a non-trivial gap between the scientific frontier and the public understanding.

I am autoregressing my way to enlightenment.

Thanks Cedar! Appreciate the support. Forever remember how we first interacted on social media (my enduring apologies), what a long strange trip it's been.

Hell yeah it does -- I had no idea y'all had such juice!

Well thank you again for the thoughtful editing!

Large Language Mistake Current AI models are not on the path to artificial general intelligence

Before continuing this debate, might be helpful if you read what I wrote.

open.substack.com/pub/buildcog...

I'm basking in the cool aspect, just for an hour or so.

@reckless.bsky.social @thedextriarchy.bsky.social I had no idea we'd be informing legal precedent! Thanks to you both, again

Last week @theverge.com published my essay exploring the limitations of large-language models. This week, that same essay is cited by a federal judge in Michigan to distinguish the process of human reasoning from what these models do. Very, very gratifying.

Oh wow, thanks for pointing me to this!

"There's a new nostalgia for Five Year Plans that's gripping parts of the Politburo..."

(Pravdabeat, 1991)

I've asked it this question and it keeps using examples of going into the wrong bathroom. I don't know why this exists.

Looks like it's a real AI startup. I think out of China.

What the f---?