Ask Lemmy
A Fediverse community for open-ended, thought provoking questions
Rules: (interactive)
1) Be nice and; have fun
Doxxing, trolling, sealioning, racism, toxicity and dog-whistling are not welcomed in AskLemmy. Remember what your mother said: if you can't say something nice, don't say anything at all. In addition, the site-wide Lemmy.world terms of service also apply here. Please familiarize yourself with them
2) All posts must end with a '?'
This is sort of like Jeopardy. Please phrase all post titles in the form of a proper question ending with ?
3) No spam
Please do not flood the community with nonsense. Actual suspected spammers will be banned on site. No astroturfing.
4) NSFW is okay, within reason
Just remember to tag posts with either a content warning or a [NSFW] tag. Overtly sexual posts are not allowed, please direct them to either !asklemmyafterdark@lemmy.world or !asklemmynsfw@lemmynsfw.com.
NSFW comments should be restricted to posts tagged [NSFW].
5) This is not a support community.
It is not a place for 'how do I?', type questions.
If you have any questions regarding the site itself or would like to report a community, please direct them to Lemmy.world Support or email info@lemmy.world. For other questions check our partnered communities list, or use the search function.
6) No US Politics.
Please don't post about current US Politics. If you need to do this, try !politicaldiscussion@lemmy.world or !uspolitics@lemmy.world
7) No Hit-and-Run questions.
Please don't delete your post for no apparent reason. If you plan on deleting a question later, say so in the post, or if you feel that you have a good reason to remove it, message a mod beforehand. It's not fair to the ones who took their time to answer, and it's not in the spirit of the community.
8) No Bots.
Posts or comments from bots, LLM's, AIs, Neural Networks, Transformers, or Marvin the Paranoid Android are not welcome in AskLemmy. Real humans only please.
Reminder: The terms of service apply here too.
Partnered Communities:
Logo design credit goes to: tubbadu
view the rest of the comments
That's just it, LLMs are not complex, although they might be "powerful" in the sense that we've thrown more processing power at them than any human endeavor in all of history.
To an LLM a word is a unique value. It sorts unique values based on probabilities of those words appearing in sequences based on sample data. It doesn't know what red is. It doesn't know what a dog is. These are just values to it no different than the words which come before or after.
When a human being, or an animal, encounter new stimulus in various forms and context they are able to study and understand it almost instantly. They're able to solve a variation of a riddle they've never heard before, they're able to learn new things.
LLMs cannot do that. LLMs have to be retrained with data in a very specific format from the start, in essence to be remade from nothing, every time new information is added. Furthermore, their statistical approach lacks any permanence in fact, it could give a right answer 7 out of 10 times and then still continue giving the wrong answer the other 3 times, because whether "red" or "dog" was the correct value is a matter of percentages and not any understanding or contextual awareness.
This is very philosophical, but what is understanding of something? Isn't it just relationships between different concepts? LLM currently encode information similarly in their multidimensional vector space. Where a "dog" vector would be close to "animal" "pet" "breed" etc vectors and those in turn to other similar concepts and so on. How different is it from human way of understanding?
Sure, that is true right now. But in not-so-distant future, it is not unreasonable to assume AI would be able to improve itself without as much (or any) human intervention. For example it could streamline training, automatically collect new data from across the world in real time. Be able to access its own weights in the model, increase improvement efficiency possibly exponentially even.
Today it is a statistical model
Tomorrow it is a statistical model
Check out the OpenAI peer review article on "AI Scaling Laws", although it had its math corrected by the staff of DeepMind later on it still more or less correctly predicted the performance of every model since and predicts that we're hitting a hard limit soon as training approaches infinite but returns become smaller and smaller. If you plot the performance as per their charts then the limit after INFINITE training with INFINITE sample data and INFINITE power used then you get about 94% accuracy, which is still less than any one singular human's capabilities. So why train it when you can just go and get a random human?
Clearly this approach doesn't work.
You mentioned a theoretical model which adjusts its own weights? We call that self-sampling which leads to feedback loops and model collapse.
In my OP, I did not mean specifically some LLM, but a hypothetical, scalable-past-human-capabilities synthetic brain.
I would argue that a human brain is a statistical model. Complex, but in the end it's a black box that takes in input from senses, and spits out output in form of interaction with environment. I'm not saying we and LLMs are in any way alike, or that AI is alive. But I do not agree with postulation that a true artificial thinking machine is impossible to build within next 1B years or however long Earth stays habitable for complex life.
You'd be wrong. Humans are a series of logic gates which don't strictly compare values odds. We actually kind of suck at statistics.