Jake Van Clief
Region: JP
Friday 23 October 2026 20:02:22 GMT
10443
426
41
35
Music
Download
Comments
InvincibleMD :
i tried 3 separate locally hosted AI models trained on Wikipedia, asked them all what the largest land animal on earth was, one said a lion, one said an otter, and the other said a gorilla. I'd rather search Wikipedia myself
2026-10-24 00:01:35
17
Hibari :
the LLM has more data than Wikipedia. If you train an LLM from scratch JUST with Wikipedia it's not going to be able to even do a simple chat with you because it doesn't have the data on how to answer a chat
2026-10-24 01:31:05
20
Camilo :
llms also need weights which can be thousands of GB in size
2026-10-24 01:00:09
3
Content with Joel :
Ignoring the fact that LLMs are inherently lossy, this is fundamentally wrong. There’s a reason a computer from 1995 can search text archives but can’t run a modern LLM. Even if you spend longer searching, retrieval uses vastly less power than inference.
2026-10-23 21:36:53
12
Brad Arlt :
it's a false dichotomy. the ai can't retrieve data like you propose. a better plan is the slm you propose with Wikipedia tied in with rag. allowing exact information retrieval and human confirmation and human training. the confirmation of the answer requires Wikipedia 😉
2026-10-23 23:46:24
2
karl 🇬🇧 :
If my 9yo wanted to find something then I would avoid him using an AI, even though I would use AI. A book would be even more energy efficient 🙂. Context matters and your presentation is very interesting. I would ask why the issue about data centres? And do we not have enough now?
2026-10-24 14:45:24
0
Artemisa the dogga :
LLMs don’t work like that if you trained based on Wikipedia you would get an useless and stupid ai. But yeh downloading a good local ai could be better
2026-10-26 20:25:14
1
sin :
This is assuming ppl are only using AI for things they would already be doing. That is not the case
2026-10-29 19:08:45
0
NoWarmUp :
Wiki all day, read once and now I know fire making.
2026-10-23 23:22:14
3
s8nsvl :
i would prefer wikipedia becaise i want the proper answer not ais approximation of an answerthat doesnt have thenspecifics i actually need my experience with ai is its useless when you need to perform manual labout and get stuf with a problem, it is actually less than useless. why wont my car start there is no spark. it sends you looking at the fuel system and the computer, whwn the coil is the obvious thing to look at but thats not even part of its answer
2026-10-23 23:55:17
2
carlosmez_1495 :
I'm a diy enthusiast mechanic and used to work on some really strange faults on my own reading tons of data in forums and pages, watching YouTube, translating from various languages etc. making research so exhausting. This past year AI deleted the time and energy waste spent on research I had to go through and now I can get so much more stuff done by simply asking the right questions and it's awesome
2026-10-23 23:02:39
6
Шапка :
Wikipedia really lacks useful lifehacks, so the usage of it/AI trained on it would be pretty narrow. But Wikipedia has too little data - if AI encounters a fact only once or a couple of times, it probably won't be trained to remember it and will rather hallucinate. So I definitely choose wikipedia to at least save some accurate data
2026-10-31 19:37:13
0
LunarMelody :
you have to look at which one is more useful to you; Wikipedia is an encyclopedia, an AI with all of Wikipedia, also has Wikipedia but the AI needs more information on top of that, and needs to be able to generate recipes One takes less functional knowledge to use, the other can act as an instructor but it's not as simple as Wiki or AI, because you can't verify if AI is correct, or have a solution to get better data when it isn't
2026-10-24 03:32:25
0
Hearb :
fix the data over the AI because the AI won't be able to give you accurate data it will just hallucinate and dribble but you know that you just not saying that
2026-10-24 01:01:04
2
joe :
🤣🤣🤣 I install project nomad I can have both Wikipedia and AI offline
2026-10-23 20:24:46
1
___superrookie :
most people cannot afford a computer strong enough to run a local AI model, that's why data centers exist
2026-10-30 22:56:31
0
ll_sergio :
you are right. most efficient is lokal LLM.
2026-10-28 22:33:24
0
🎤❄️Holger❄️🎤 :
Training is power expensive
2026-10-29 07:56:31
0
Ale :
if you'd use the ai for lookup only, your analysis is right. but it can be used for reasoning, offload tasks from your brain, and more
2026-10-27 18:43:02
0
dakillerhairball :
I would just do wiki and skip the ai. and remove 40% of wiki do to most of it is useless at the end of the world.
2026-10-26 07:45:04
0
Gether.To :
Also the amount of data that is not lies on Wikipedia is probably less than half so that changes the picture
2026-10-26 05:31:29
0
makotomusicTikTok :
why would you train an AI model with Wikipedia? Either choice is useless
2026-10-25 07:07:19
0
AISweeties :
Inference is still costly in terms of power vs a search on wikipedia if you run them both locally, granted if you are not good at searching you might have to search for longer and then power consumption becomes a wash, the difference being getting your information directly from Wikipedia would give you a bit more confidence that the AI didn't just hallucinate the answer 😅
2026-10-24 12:02:28
0
Unforseen Consequences :
If done correctly, there's nothing better than having it local. Treat it like you're raising a baby with verifying is reasoning with larger models, and watch it become a professor at any specific task you're looking to take on. And your data is yours, Sandboxed within the system.
2026-10-24 04:14:09
0
HayatiRocks 🇲🇦🇨🇦⚜️ :
Very interesting
2026-10-23 21:43:25
0
To see more videos from user @jakevanclief, please go to the Tikwm
homepage.
