• Please review our updated Terms and Rules here

AI - What might people like us use it for?

westveld

westveld

Experienced Member
Joined
Dec 31, 2009
Messages
227
Location
Reed City, MI
The recent developments in ML / LLM are something different, with implications both great and scary.

Not the usual incremental tech improvements (the same stuff just smaller better faster).

I've just been trying out the ChatGPT 4o, o1-preview, o1-mini models.

For me at this point they are a kind of "Super Google" - saving time by pulling stuff together from the results of a search.

Helpful for getting:
- Code examples
- Specs ( eg: what are the different formats of 8" disks? )
- Random starter info on topics I'm unfamiliar with

It would be interesting as a tool to mine old data, maybe models trained on:
- Usenet
- Forums
- Old IC datasheets / databooks

I'd like to have something use a photo of the top of an IC and identify it (OCR text, pin count, chip size) and spit out a title and specs based on a datasheet search.
And if it could figure out house markings too, that'd be great!

Thoughts?
 
I've also been using ChatGPT 4o to get starter info on new topics. Such as: the history of a company and/or a product line, OS compatibility on a given machine, components supported by a machine, history of an OS, and the list goes on.

I agree that it's an improved search experience. And I also enjoy that I can verbally converse with it at any time!

A limitation - currently - is that 4o can't search the web (real time), so I can't ask for example prices of a component/machine/etc. The one time I asked for an image of a machine it produced an AI-generated image... whereas I thought I was going to see an actual one. I'd imagine both of these facets will be worked on but we'll see.
 
Last edited:
I would not trust ChatGPT for accurate technical or historical data.
 
I used ChatGPT to write me an outline for a presentation I did. It was not perfect by any means, but it was good enough to get my slides done and plan out the flow. Saved me more than a few hours work for sure. I find most of these AI tools are a great way to get started on projects, but they seldom produce a final product.
 
Hm, I didn't realize 4o allows you to send an image.
It seems to be pretty OK at photo-OCR of chip markings.
Time to dig into the API I think!
 
Oh, I know! We could start a bunch of threads on the model of "I asked ELIZA about this highly technical question and here's the probabilistic word-salad she served up, can any of you boring fleshbrains tell me if it's anything approximation conventional notions of 'correct' plzkthx?" That'd be great!
 
I use ClaudeAI to write base code for me, then I modify it manually. I've found that if you try to get it to modify or update the code you end up down a rabbit hole. Easier to just do it yourself.

I did read somewhere that you could completely populate a social media platform with AI's such that it just lives in its own world with no humans involved. I'd be interested in what that looked like after a couple million iterations :)
 
DougM said:
I use ClaudeAI to write base code for me, then I modify it manually. I've found that if you try to get it to modify or update the code you end up down a rabbit hole. Easier to just do it yourself.

I did read somewhere that you could completely populate a social media platform with AI's such that it just lives in its own world with no humans involved. I'd be interested in what that looked like after a couple million iterations :)
Click to expand...
I've heard someone describe AI as an 80% tool, in that it gets you about 80% there, but you still have to do the rest. I completely agree with this, and this is truly where the power of AI does come in, especially in industries like education, where the toll allows teachers to spend less time managing their lesson planning and more time actually teaching.
 
I personally don't see a lot of use for these chat-"AI"s. There is a lot of misunderstanding about what they can or can not do out there. I do recognized that they have some valid uses, and some potential uses may improve over time.

I sometimes compare these chat-AIs to a bored research student that does not understand the topic. That can be useful, but it is not something one should rely on by itself.
 
I've found its really useful for getting desktop-level technical support on out-dated operating systems(winXP, '98). Also weirdly Microsoft's AI was happily helped me pirate M$ OSs.
 
I have wondered, given AI is created mainly from the stuff on the web where an ever increasing amount of information is just garbage, where will this take AI to in the future?
I suppose if you feed enough garbale in, eventually something worthwhile may pop out.
 
Garbage in, garbage out.

Its kind of an interesting problem: AIs generate garbage, then build on garbage to produce more garbage. In theory after a point each successive generation is going to start getting worse.

Its like an Ouroboros of garbage.
 
Yes, the result of an AI feedback loop will be rapidly devolving garbage. This is already starting to be an issue for stable diffusion.
 
One way you can tell artwork is AI-generated is when it has JPEG compression artifacts in it, even when it's supposed to be a lossless PNG file. Since the AI algorithms were trained using compressed JPEG images, they think the compression artifacts are part of how the images are supposed to look, and are reproducing that in their output.
 
Just uploading images as the prompt is interesting
 

Attachments

  • firefox_l1rBHNNWYs.png
    firefox_l1rBHNNWYs.png
    214.5 KB · Views: 21
  • firefox_k5mJgY3PfC.png
    firefox_k5mJgY3PfC.png
    154.4 KB · Views: 21
You must log in or register to reply here.
Back
Top