LOL *NSFW* *The new Demote thread*

If you have Instant Answers turned on (which is different to the AI ******** they're doubling-down on, and has its own on/off toggle in the General Settings tab), you get the right answer no fuss.

 

Works for other "magic" numbers in other conversions, too. There's a simple reason, too: LLMs are large language models, not large number models. They don't work in numbers, they work in tokens - and the tokens assemble into words, not numbers. Oh, sure, some of 'em look like numbers, but they're still words.

The word "1250" is associated with inches in the training set, so it returns tokens which reassemble into an answer shape as though it were inches - regardless of what you actually asked for. Better still, it churns through way more computational resources than any other method of figuring out the answer... and still gets it wrong.

The Future!

 

Had a few AI things pop up in my feeds over the last couple of days, starting with the artist behind Kryptos - a cryptographic statue with an as-yet unsolved side - getting flooded by people who reckon they've solved it in a morning thanks to Delphic oracles AI chatbots. Spoiler: no, they have not.

A February study out of Stanford confirms what we all knew: the web is being choked to death by AI slop.

AI chatbots are completely useless at answering news-related questions, and the more you pay the worse they get. Big shout-out to Pedo Guy's Grok, which was wrong 91% of the time.

Oh, and their training data may have been poisoned by a Russian disinfo campaign.

"But," you say, "I only use it for code stuff, where it's obvious if it's wrong!" Yeah, until it bricks your system.

THE FUTURE!

 

You can also search LibGen directly... and you can download stuff too.

 

(Source.)