I mean, Baba Yaga is the best. I was thinking of her but was not gonna attempt chicken feet; I thought a hefty tree trunk was safer

Haha makes sense. I’m not a polyglot but I love languages: speak two languages pretty well, understand 50-70% of two more and have studied another 3-4 some, and IME you are correct. It’s all just work. For some people the work feels fun often, and immersion helps a lot, but only work “works” 😂

Haha, tho in our house they usually sit around til January some time when they are visibly dusty and kind of gross 😂

It was pretty great

Ooo maybe?? I dunno!

That’s what I thought! And it did look like it would. Maybe it’s the weight of all the decorations

Musta been too much weight. Maybe it needed a double layer for both trunk dimensions?

The trunk cookies just literally broke in half.

We did that. 😭 We just kept the trunk and the house separate to decorate (but had glued a platform to the top of the trunk for the house to sit on

It was so cool! 😭

(This was poorly phrased—some companies have made multi-component systems that include both LLMs and explicit reasoning systems that evaluate what the LLMs output before deciding what to pursue/output)

Yup. It’s fine for something you can verify, but if you don’t know the right answer yourself, it’s a nightmare bc it’ll make up something that sounds very convincingly answer-shaped and it might be true or it might kill you. 😵‍💫

LLMs are a subfield of machine learning, which is a subfield of artificial intelligence, which is a category we have been using for decades to classify pretty much every problem we don’t know how to solve well computationally.

LLMs alone will never be reliable for “truth”

Sometimes, or even often in the right situations, most-likely-next-word sort of language generation DOES produce true statements, but there is no actually checking or verification built into the system, so no way to be sure without checking everything after.

Some companies, for some purposes have hooked up LLMs with explicit reasoning, and that has worked better, but isn’t what chatbots are using—they’re just barfing out free-associations basically, bc that’s what LLMs *do*

They’re vaguely, sorta like if you set the language center of your brain to free associating without any of the other bits that evaluate whether what’s coming out makes sense or not. They’re very good at producing language that sounds natural and grammatical, and good at spotting patterns in lang

LLMs are a subfield of machine learning, which is a subfield of artificial intelligence, which is a category we have been using for decades to classify pretty much every problem we don’t know how to solve well computationally.

LLMs alone will never be reliable for “truth”