I have just discovered this Ai tech, will give one instance where its a game changer that i have recently used. I am currently starting another bike build, wanting to use parts from other machines and graft them on to the bike, in the old days it was joining forums and asking questions which can take forever. Now its ask "will this fit this and what mods are needed"...you get an instant answer yes or no and if not it tells you why and what you need to do to make the part fit, even tells you what year of production will fit another year...very helpfull the whole Ai engine is now.

I have just discovered this Ai tech, will give one instance where its a game changer that i have recently used. I am currently starting another bike build, wanting to use parts from other machines and graft them on to the bike, in the old days it was joining forums and asking questions which can take forever. Now its ask "will this fit this and what mods are needed"...you get an instant answer yes or no and if not it tells you why and what you need to do to make the part fit, even tells you what year of production will fit another year...very helpfull the whole Ai engine is now.

If it's telling you the truth, which you won't know until you try it and find out it's wrong.
It's replies are still only a best guess based on information, both correct and not, scraped from forums etc..

As an AI skeptic how do you know what it says is correct? It'll have got it's data from scraping the web and there's a lot of rubbish out there in the forums.

Edit: Think of the how to stop your brakes squeaking threads.

As an AI skeptic how do you know what it says is correct? It'll have got it's data from scraping the web and there's a lot of rubbish out there in the forums.

Edit: Think of the how to stop your brakes squeaking threads.

It is a useful tool in the same way as a calculator. You will have some idea of the correct answer before you ask, so that will guide you. The other point is that you can query or argue with it and it will refine its answers as well as citing sources. It is not much different to talking to a very worldly wise mate at the pub; anything mission critical you will double check or challenge.

That having been said, I now consider my current ChatGPT subscription to be well worth it, whereas I stopped paying last year as I was led down too many rabbit holes. It is great to be cautious, so for now just treat it as a VERY efficient search engine and then challenge its output. You can even ask "How can I confirm the accuracy of this answer?".

As an AI skeptic how do you know what it says is correct? It'll have got it's data from scraping the web and there's a lot of rubbish out there in the forums.

Edit: Think of the how to stop your brakes squeaking threads.

Maybe some of the LLMs are better than others, I've never used any of them in anger. But if the summary Google gives you for every search is any guide, I think you're right to be a sceptic.

Ask it some obscure technical question, that only somebody in your field would know. It'll always give a confident plausible looking answer. But the trouble is, when you look closer, half the time the answer is complete ******, cobbled together from bits of other answers to questions that only look similar on the surface. And somebody new to the field won't know which half.

Maybe some of the LLMs are better than others, I've never used any of them in anger. But if the summary Google gives you for every search is any guide, I think you're right to be a sceptic.

Ask it some obscure technical question, that only somebody in your field would know. It'll always give a confident plausible looking answer. But the trouble is, when you look closer, half the time the answer is complete ******, cobbled together from bits of other answers to questions that only look similar on the surface. And somebody new to the field won't know which half.

You can't use that Google thing as any sort of measure. Give one of the big ones a go, I use ChatGPT as said. I have used it for an enormous amount of queries, from analysing blood test results, to writing software and gardening help. None of it in 12 months has been in error.
Recently I have been messing about too; asking what it imagines "Marvin" to look like like it he really existed. That was a fun conversation!

This is just a small selection of some of the subjects I have found it useful for:

That having been said, I now consider my current ChatGPT subscription to be well worth it, whereas I stopped paying last year as I was led down too many rabbit holes

I pay my Anthropic (Claude) subscription every month, the improvement in quality is noticeable.

That thread is following the predictable split:

1. Roofman’s opening position
   He has found AI useful for practical compatibility questions — bike parts, swaps, what fits what, what modifications are needed. His enthusiasm is understandable, but that is exactly the sort of use case where AI can sound most convincing while being dangerously wrong. It may be right, but you often only discover the error when the part is on the bench and does not fit.
2. MysteryG and cheb’s objection
   They are making the strongest practical criticism: how do you know it is correct?
   That is not anti-AI nonsense. It is the central issue. AI gives fluent answers assembled from patterns in available information, including forum posts, adverts, manuals, bad memories, wrong assumptions, and copied myths. For mechanical fitment, a confident wrong answer can cost time, money, or safety.
3. The “forums are full of rubbish too” counterpoint
   This is also true. Old forum advice is often wrong, model-specific, half-remembered, or based on a bodge that happened to work once. AI is not uniquely unreliable; it just packages unreliability with a very polished voice.
4. Your Marvin example
   That is probably the best demonstration in the thread. You were not asking for a safety-critical answer; you were using AI for imaginative comparison, wording, and interpretation. That is a much better fit. The “Wallace and Gromit lunar robot” comparison is subjective, so there is no catastrophic failure mode if the answer is a bit fanciful.
5. The weak claim in the thread
   Any version of “AI will tell you what year of production fits another year” needs heavy qualification. It might, but it should be treated as a lead, not a conclusion. The proper use would be:

   “AI says X part from model years A–B may fit. Now check parts fiche, dimensions, mounting points, wiring, service manual, photos, and ideally someone who has done the swap.”

6. The strongest pro-AI argument
   It is excellent for shortening the first stage of research. Instead of spending hours searching through contradictory forum threads, it can give you a structured starting point: likely compatible models, terms to search, part numbers to verify, likely traps, and questions to ask.
7. The strongest anti-AI argument
   It has no embarrassment reflex. It will not naturally say, “I don’t know; measure it.” It may invent certainty where a human expert would ask for the VIN, year, market, engine code, casting number, or a photo.

The sensible forum answer would be something like:

AI is useful as a research assistant, not as an authority. For letters, summaries, explanations, first-pass research, and organising thoughts, it can be very good. For welding settings, structural design, vehicle brakes, steering, lifting gear, gas safety, electrical safety, or mechanical compatibility, it should be treated like an enthusiastic apprentice: useful, quick, sometimes insightful, but requiring supervision.

That is probably the line I would take if replying there. The sceptics are right about verification; the enthusiasts are right that it can save a lot of time. The error is letting either side pretend it is either magic or worthless.

I use it quite a lot, my latest Freelander project would not have been possible without it

I recently went through a recruitment process, multiple times it asked about AI use, saying yes caused your application to be withdrawn, saying no but actually using it meant removal from the process

After the interview I asked why the strong stance, basically they had had multiple people who were amazing on paper but turned out to have entirely invented qualifications, experience and job history, interviews were a waste of time and money as they could no answer basic questions so they took a hard line

That’s the key, asking the correct question in the correct way. I use one AI to write the prompt for another. Asking ChatGPT in free text what I want to achieve and getting it to write the prompts for Claude, or Codex, or CoPilot and so on.

I asked AI for a sensible answer on a forum and it gave me this:

AI is useful as a research assistant, not as an authority. For letters, summaries, explanations, first-pass research, and organising thoughts, it can be very good. For welding settings, structural design, vehicle brakes, steering, lifting gear, gas safety, electrical safety, or mechanical compatibility, it should be treated like an enthusiastic apprentice: useful, quick, sometimes insightful, but requiring supervision.

If it's telling you the truth, which you won't know until you try it and find out it's wrong.
It's replies are still only a best guess based on information, both correct and not, scraped from forums etc..

Yes i did think that, but it does all the work for you on one page, starting point and go from there.

I have just discovered this Ai tech, will give one instance where its a game changer that i have recently used. I am currently starting another bike build, wanting to use parts from other machines and graft them on to the bike, in the old days it was joining forums and asking questions which can take forever. Now its ask "will this fit this and what mods are needed"...you get an instant answer yes or no and if not it tells you why and what you need to do to make the part fit, even tells you what year of production will fit another year...very helpfull the whole Ai engine is now.

It’s good isn’t it. I use it like my personal assistant

That was a fun conversation!

Define fun.

It’s good isn’t it. I use it like my personal assistant

Exactly. Where I would google something in the past and then spend 15 minutes visiting a dozen websites and combining the answers, I now just ask ChatGPT.

The one point I would make is that it's essential to apply critical thinking to what it responds with. Just as you would when the guy down the pub tells you with absolute certainty that Devon is the world's largest producer of mangos, you need to be able to take a step back and ask yourself "does that seem likely?".

If you do that then it's an incredibly useful tool. My list of things-I've-asked-ChatGPT-in-the-last-48-hours is as eclectic as @ajlelectronics' list.
The other great thing about it is that you can have a conversation to arrive at the right level of information / advice, which you can't do as easily with plain google.