Re: [neonixie-l] Alarm Clocks and ChatGPT

2024-02-18 Thread Tom Nolan
Hi Dylan, That's very interesting. Thanks for trying a different system. Same extrapolation errors. The apology cracks me up. It's especially funny to me because I own a Sony TR-610 and a Philco 850T. It's really difficult for me to understand how the Sony fits into the subject at all; no clock,

Re: [neonixie-l] Alarm Clocks and ChatGPT

2024-02-17 Thread Dylan Distasio
A quick follow up, it would appear Claude2 is also making mistakes. I checked the Sony and it was actually a pocket radio. Here's what happened when I challenged Claude2 on it: You're absolutely right, my previous example of the Sony TR-610 was incorrect. The TR-610 was a transistor radio model

Re: [neonixie-l] Alarm Clocks and ChatGPT

2024-02-17 Thread Dylan Distasio
Just my two cents working with LLMs a fair amount...They are all prone to hallucinations when they don't know the answer. With direct access to the LLM, you can reduce this sometimes by lowering the "temperature" parameter. You may be able to approximate doing this through prompt engineering wi

Re: [neonixie-l] Alarm Clocks and ChatGPT

2024-02-17 Thread Tom Nolan
David Pye and Audry, I guess I'd never asked difficult questions before but I did get helpful answers, nothing as stunningly wrong. Tom On Sat, Feb 17, 2024 at 7:48 PM Audrey wrote: > > Previous to this ChatGPT exchange I had useful experiences with ChatGPT > but this has undermined all confiden

Re: [neonixie-l] Alarm Clocks and ChatGPT

2024-02-17 Thread Audrey
> Previous to this ChatGPT exchange I had useful experiences with ChatGPT but this has undermined all confidence it. I am glad you had this experience, you should not inherently trust the output of any LLM, they are just as fallible as a human, and they word their output confidently. I have asked

Re: [neonixie-l] Alarm Clocks and ChatGPT

2024-02-17 Thread David Pye
ChatGPT is an LLM. If you ask it to write a legal argument, it will produce something that looks like a legal argument. It even has references. However, it was demonstrated that it had invented those references. It produces something that "looks like" a coherent argument but it just confabulate

[neonixie-l] Alarm Clocks and ChatGPT

2024-02-17 Thread OrangeGlow
This isn’t completely nixie related but of all people the folks on this list will most likely understand and appreciate what I have to say. It is about a recent exchange I had with ChatGPT3. [TL;DR: about history of technology] I’ve spent the last couple of years working on a couple of alarm