Seeing quite a few post now that are quoting information from LLM's like chatGPT, Gemini etc. Open disclosure - I have an axe to grind with AI. Before I retired, getting these things to work was my day job, and its a sore subject the amount of reliance that is now being put on them.
However, I'm of the view that if you are going to use a tool, its incumbent on you to at least have a basic understanding of how the tool works. Now, the actual tech behind LLM's is complicated, but there is information out that that tried to get the general idea over. I've had a look around and this one from Ars Technica is pretty good. https://arstechnica.com/science/2023/07 ... dels-work/
If anyone want to get a better understanding, more than happy to go over details. Considering how ubiquitous AI is becoming, and how much it is going to impact our daily lives, then I think this falls into the prep'ing domain.
How LLM's like ChatGPT work
Re: How LLM's like ChatGPT work
The number of incorrect answers I have received from Grok and other AI systems is unbelievable. Yet the AI will double down over and over until, at the last minute, say "oh yes, my previous answers were rubbish".
Well, that's great but if I was relying on you I'd be dead or injured by now.
Like the early SATNAVs, folk would drive off cliffs because they were told that it was a road, not a dead end.
Mind if you're gullible enough to accept an AI answer without any critical thinking on your own part, should you be in the gene pool?
On politics and world events, Grok et al decide which the "reliable sources" are for you.
Fatally flawed unless you're gullible and will swallow what is fed to you.
Well, that's great but if I was relying on you I'd be dead or injured by now.
Like the early SATNAVs, folk would drive off cliffs because they were told that it was a road, not a dead end.
Mind if you're gullible enough to accept an AI answer without any critical thinking on your own part, should you be in the gene pool?
On politics and world events, Grok et al decide which the "reliable sources" are for you.
Fatally flawed unless you're gullible and will swallow what is fed to you.
Re: How LLM's like ChatGPT work
I've found them quite useful in tasks which are not subjective. Like asking for opinions or free text. But these models are quite good now at tasks which are more quanititative. Like coding, where the user already knows what "good" looks like and can measure and validate the response.
Some of the new qwen and kimi models would serve as great prepper options for vast knowlage stores that can be queried directly, although models that are small enough to be portable may not be very accurate in their responses.
4b parameter models can be run on mobile phones locally or raspbarry Pi's but with high power draw. You have to ask - what scenario would really necessitate running that kind of compute and power infra to justify it?
I see them being most useful in post collapse recovery rather than any actual bug out emergency.
Some of the new qwen and kimi models would serve as great prepper options for vast knowlage stores that can be queried directly, although models that are small enough to be portable may not be very accurate in their responses.
4b parameter models can be run on mobile phones locally or raspbarry Pi's but with high power draw. You have to ask - what scenario would really necessitate running that kind of compute and power infra to justify it?
I see them being most useful in post collapse recovery rather than any actual bug out emergency.
-
GeraldTheBonzai
- Posts: 347
- Joined: Fri Oct 08, 2021 7:52 pm
Re: How LLM's like ChatGPT work
A way to mitigate the flaws in LLMs i've mentioned in another thread. Basically you need to create your own curated repository of information ( from sources like archive.org, kiwix etc) then tell the LLM to refer to your trusted sources first.
Re: How LLM's like ChatGPT work
I'll admit I'm a bit of a ludditte when it comes to AI . I do use chatgpt but really as little more than a glorified search engine. My wife uses it for her ebay activities and my daughter uses the paid for version for work and uni.
I see quite a bit of AI on social media and some of it is laughably bad. Supposedly historical videos where things like tanks , uniforms , helmets and guns are wildly inaccurate despite there being a wealth of actual historical sources. I could be barking up the wrong tree but the cynic in me can't help thinking they are made deliberately bad to discredit AI...
I see quite a bit of AI on social media and some of it is laughably bad. Supposedly historical videos where things like tanks , uniforms , helmets and guns are wildly inaccurate despite there being a wealth of actual historical sources. I could be barking up the wrong tree but the cynic in me can't help thinking they are made deliberately bad to discredit AI...