[GoLUG] LLMs can't reason

[Ron]

Kevin Chadwick wrote on 2025-06-20 09:25:

>> It was a joke.
> 
> I guess I need to lighten up, sorry.
> 
> p.s. I did see it but couldn't be bothered to fix it on my phone 
> with crappy> capacitative touch

All's good!

A trick for input on phones that I heard people rave about for years 
before trying myself is swiping or gliding:

Press the first letter of the word, drag finger near each subsequent 
letter whilst maintaining contact, and >90% of the time it gets the 
correct word, *never* a typo.

Might get the wrong word though, but it's... slick! And fast.


Took me a day or two to get used to it and I'm never going back.


Also, sliding left or right on the space bar moves the cursor, which is 
also very handy.


Both settings supported by Gboard and others (Swype was first to 
implement swiping. Founded in 2002!?! Wow, I wish I'd started using it 
much earlier.)



> The AI lies wind me up a bit. I hate lies, especially when it's 
> rewarded with huge sums of money.

Agreed, the tsunami of lies and bullshit is the curse of our times.


But I'd argue that humans confidently spewing BS is a much larger 
problem - they vote, for example. And should be able to differentiate 
beliefs from facts.


We've seen it on this very list, in other messages on this very thread. 
It's disturbing.


And, AI can be handy - it's really helped me with some thorny SQL 
queries creating functions in PostgreSQL's weird syntax as recently as 
today.

Documentation can be lacking, or obtuse, or written for a level of 
expertise that makes them hard to digest, or don't quite explain an 
unusual and specific situation.

DuckDuckGo's built in ChatGPT bot is really handy in such cases.



> .ca, eh. I'm actually visiting Canada soon.
Nice - here's hoping you have a great time!  Where will you be visiting?