Google's AI summary is still at it.
I just realised that for the base of the word it didn't use my misspelling ā€˜kilogramme' nor its corrected spelling ā€˜kilograms’. Even if you drop the two t's, it's kilograme. <insert exasperated facepalm here>

@billgoats I love the defence of LLMs that takes the form of ā€œbut they’re not made for advanced calculationsā€

Where by ā€œadvanced calculationsā€ they mean counting things the way a toddler has typically mastered.

@billgoats but there is only one t in the metric spelling of kilogram
@billgoats A self-healing system, at least when you ask how many r's in kilogramme.
@billgoats Given how they work, and that the first step is tokenising the input into numbers, this is an unsurprising result. Kilo-gram might be tokenised to (for example) 24253 72525, and it’s just vectors from then onwards. The letters don’t exist.
@billgoats I do wonder if they can fix this with some built-in tool calls, perhaps taking a page from Wolfram Alpha's approach
@billgoats That's why RAM & storage prices are insane? Oh well...
@billgoats The real question in this is why is someone using Google? šŸ¤”