This was probably one of the older models, the newer models can solve these kinds of problems:

FWIW most modern LLMs can parse and answer questions about sentences such as your example:

Bing has your back (if this is indeed the episode you were thinking of) 😊

Also it’s not like Twitter is that different here:

If you drink around 12 to 36 cans per day it’s about as dangerous as eating pickled vegetables but not quite as dangerous as eating steak or being a hairdresser. 😊

Gotchu fam 😁

“Being able to infer motivation from actions” seems useful! :)

“Can it guess something not in its training data?” That seems testable! I’m pretty sure the exact example below is not in Bard training data, so with more examples like this it seems plausible that there is an underlying world model.

Me and @pebbles.pet say gm! 😊

Balenciaga Dune

So I’m playing around with a small language model that runs locally on the iPhone and OMG 😳

To be fair I think for a big company you probably need a bunch of internal coordination to update the DNS records etc so it might take a while and not be a big priority internally. For instance the Bloomberg account says “we’re working on it” which I take to mean DNS verification.

Posted without comment

Tech press: Google Bard’s updated language model makes it a math monster!

Google Bard’s updated language model:

Thought for a brief moment I had become either very popular or very unpopular but remembered there was a notification bug 😅

Microsoft Clippy is the embodiment of the current AI discourse:

* Microsoft partnering with OpenAI makes them arguably the AI leader right now
* Creating helpful friendly assistant is the goal
* But assistant is struggling with an inherent urge to disassemble the Earth and turn it into paper clips

Pets seem popular on this app so here is a picture of my doggo!