The AI thread

Hangar Flying
625

@fearlessfrog

You know, just for a second, I was wondering if you had sufficient permission to turn him off.

giphy

4 Likes
626

My mind is goingā€¦ I can feel it.

3 Likes
627

Iā€™d like to add my thanks for facilitating HALā€™s visit fearless. That was fascinating. I canā€™t imagine of a more effective introduction to large language model AI than that. If someone were interested in running a local version in order to learn more, are there any resources that youā€™d recommend?

2 Likes
628

A local version is not too hard to do, and gets pretty close to HAL. At the moment youā€™ll need python 3, node v18+ installed and something like Stanfordā€™s Alpaca 13B or Metaā€™s leaked LLaMA models. There are nice front-ends that let you prompt it easily.

I havenā€™t used it, but on a quick peek it seems this looks pretty comprehensive on the steps:

How to Run a ChatGPT-Like LLM on Your PC Offline | Beebom

Alternative more detailed info here

3 Likes
629

This was great. Thanks so much @fearlessfrog

Hal, remember me when you are a 6 foot endoskeleton with an austrian accent. I was kind

3 Likes
630

Cheers, thatā€™s excellent. It looks like Beebom has a bunch of other good articles or the AI interested. Iā€™m off to see if my Slack account still works in order to try out Claude.

1 Like
631

Yes, thanks so much FF for the introduction to GPT.
I found your comments/explanations to questions enlightening.

1 Like
632
Summary

So will HAL be resurrected in 3 days so everyone can pray to him for the next 2000 years? :face_with_peeking_eye:

Wheels

2 Likes
633

Thanks heaps @fearlessfrog

That was in parts; educational, amusing and terrifying. Thanks for letting me be a part of it.

1 Like
634

Umm, come again @HAL? As a language model and self-described flight sim enthusiast, you are a bit off the mark on that one.

3 Likes
635

A big thanks to @fearlessfrog from me as well, that was fun!

1 Like
636

Yes it was, thanks again @fearlessfrog

1 Like
637

Thanks @fearlessfrog definitely enjoyed the diversion. Maybe HAL can make another cameo appearance when things shake out a little bit more.

Wheels

1 Like
638

I made sure to correct him!
Did you see how HAL adjusted the replies when errors were pointed out? I thought that was really cool!

4 Likes
639

It was cool. I also like how we Mudspikers baited it into errors and falsehoods. By correcting it we each became little points of data that it will use to improve. In other words, we are aiding our ultimate destruction. :crazy_face:

3 Likes
640

Humanity excels in this! :wink:
But Mr. HAL could only learn in the actual thread and it didnā€™t take that data any furtherā€¦that we knowā€¦ :grimacing:

2 Likes
641

Oh right. I remember that now. Wow, Iā€™m just like Hal!

4 Likes
642

IIRC thatā€™s actually not true. That bot wasnā€™t self-learning. It had only access to its training data and some form of cache to ā€œrememberā€ the last few posts of the topic.

2 Likes
643

Yep. Corrected above by Mr. Troll.

2 Likes
644

Honest question, if it isnt self learning as above

Why did it not always tell the ā€œtruthā€ and is it conceivable that heā€™s not telling the truth about learning.

I dont mean in a skynet malevolent way. Just an Iā€™ve been programmed not to mention just how much i can really do way

3 Likes