[GoLUG] LLMs can't reason

[Kevin Chadwick]

On 13/06/2025 08:19, Steve Litt wrote:
> Where do I even start? ChatGPT would NEVER make that mistake.

I believe that's incorrect. Did you see the car company that had
chatgpt as it's chat bot. Chris Wright from embedded FM knew of LLMs problems and
convinced it to give him a car for free. A human would never do that.
Technically the company had to give it to him due to some terms and conditions
but I believe he let them off the hook.

I used chatgpt to write some Ada code. I asked it to write a driver for
something that I forget. It couldn't because the code didn't exist on the
internet. Eventually it gave me names of functions that looked right but just
randomly and dangerously played with GPIO clocks. A human might do that but at
least they would know they were being deceitful and could be fired.