Am still learning about waiting lists, very much not my area of expertise, but my understanding is there are BIG data quality issues here. Giving people greater visibility to these problems could actually create more frustration for individuals, not less.

Could be a safeguarding type thing? If a constituent (for some reason) emails their MP and situation is an emergency, does MP's office have some sort of duty of care to pass on to correct service which only really works if said service can then phone the individual straightaway? Dunno

Maybe something more fundamental than training? Would make naive sense that models just predicting what the next word should be give most weight to the most recent word it spat out, less to the word before that, and so on backwards, so low/zero weight given to words 20 or more ago => repetition?

2/ Almost as significant is the clear message being sent by two dates associated with the Black community being deleted, despite both being federal holidays: parks are no longer free to enter on Martin Luther King Jr. Day (third Monday in January) and Juneteenth (June 19).

www.fjc.gov/history/judg...

Southern District by any chance?

www.bbc.com/news/article...

Is there any info on where this hypothetical action might take place? Only because I think the last time Trump tried something similar in UK courts he.... lost www.theguardian.com/us-news/2024...

This might feel like splitting hairs but I think the language here is super important if we are to understand and use LLMs appropriately. If we try to project intelligence on to these things based on misunderstanding the tech, that's on us (though the AI hype train has a lot to answer for) /end

At one point the article reads:

"'Mere presence plus an ambiguous reaction under stress falls short of proving shared intent beyond a reasonable doubt in my view' ChatGPT explained"

My issue is with the word "explained". ChatGPT does not understand, it is not intelligent, so cannot "explain". 4/n

I believe we also need to be much more careful with the language we actually humans use around LLMs. I've even done this wrong myself in this thread by using the word "they" to describe LLMs - "it" is probably more appropriate since anthropomorphising these models is part of the problem 3/n

LLMs are just text generators. They do not weigh up evidence. They do not understand the law, because they don't understand. They have no concept of "fact", "right" or "wrong", because they have no concepts. They just splurge out the most statistically likely reasonable words. 2/n

A smart admin would have moved to drop this "case" on Tuesday before the polls closed