I would love to train an AI model with millions of old electronics service manuals so you could just ask it about anything, like a veteran radio or TV technician. How to repair, adjust, and suggestions even if it doesn’t know specifics. That knowledge is getting rare these days
Conversation
Notices
-
Embed this notice
Benj Edwards (benjedwards@mastodon.social)'s status on Monday, 06-Mar-2023 12:31:53 JST Benj Edwards
-
Embed this notice
Thomas 🔭✨ (thomasfuchs@hachyderm.io)'s status on Monday, 06-Mar-2023 12:31:52 JST Thomas 🔭✨
@benjedwards If you like to get slightly wrong to very false instructions on how to repair high-voltage equipment, go ahead
-
Embed this notice
Thomas 🔭✨ (thomasfuchs@hachyderm.io)'s status on Monday, 06-Mar-2023 12:44:25 JST Thomas 🔭✨
@benjedwards fair enough
-
Embed this notice
Benj Edwards (benjedwards@mastodon.social)'s status on Monday, 06-Mar-2023 12:44:26 JST Benj Edwards
@thomasfuchs My hypothetical scenario involves a fantasy LLM that is accurate
-
Embed this notice
Thomas 🔭✨ (thomasfuchs@hachyderm.io)'s status on Tuesday, 07-Mar-2023 06:11:05 JST Thomas 🔭✨
@chucker @benjedwards There aren't service manuals available for everything anymore.
A LLM could infer instructions to repair things from more vague descriptions of problems and even for models not directly covered by scans.
Alas, a LLM will also make up shit and do it very confidently.
-
Embed this notice
Sören (chucker@norden.social)'s status on Tuesday, 07-Mar-2023 06:11:06 JST Sören
I’m probably overthinking this… but if you need to feed it those manuals anyway, wouldn’t it be simpler to stuff a bunch of PDFs in a document store with full-text search (and run OCR where needed)? Takes fewer computing resources and yields higher quality
-
Embed this notice