Oddbean new post about login | logout Im not an expert but the LLM analyzes the text broken down into little tokens like words or parts of words or letters and then the llms are trained on statistical relationships between the tokens in the data it analyzes. This gives tokens a kind of preference in relation to other tokens (preference is just a word I used not a technical term) so when you use the llm you aren't seeing humanities knowledge youre seeing a statistical representation of humanities mode of expressing its knowledge.
Still too advanced for me. “AI just regurgitates unoriginal word salad, comparable to Kamala Harris after 1 bottle.” There.
I use it mostly as an alternative to search engines.
In line with my theory that the present doesn’t really exist, and humans exist in a predictive hallucination, i assume ai works the same way, decryption Token is just an interesting word choice besides codex or meme or symbol or character 🥠🧧
I like the word token because it has less normative or symbolic meaning beyond what it is because the AI isn’t thinking this token is part of that word just factoring in that token A tends to be near tokens Z and T etc. In fact there’s no thinking at all. Also, I agree. The world and time is an illusion and we live in a holographic universes projection-dream.
I picture AI training like massive game of plinko where each puck bends the rods it comes into contact with a little bit until you have a plinko game that isn’t predictive but tends to produce similar meme outputs for similar meme inputs
I picture AI having about as much intelligence as the CIA 🤣
I think that only happens if the predictions are wrong But define wrong
I don't rely on AI for right or wrong. For me, it's more like having autistic friend I can bounce ideas off of and then explore for myself from source materials. Also, wrong is defined as any conclusion of opinion tightly held. This is why I like bitcoin, no opinions in the math of it.
So far found it really useful to deal with things normal search engines would take an age to deal with. Things like - give me a list of _ who knew _ & what worked on ? It’s incredible for stuff like that
Totally. Also, sometimes I'll ask questions like "where did I leave my keys" and "why is my wife mad at me." Less results there, but in lieu of friends, it can help.
Yes 🤣 or how can I deal with the thought of killing my co worker
I recommend unleashed.chat or venice.ai for that question. The answer is obviously leave the thoughts in your head and kill the thoughts instead, lol.
Thanks John - 21 sats for the Therapy session
It's funny though, I did have this idea of making myself available for sats just to have a conversation, with anyone, about anything. Not debate, just give me a few sats and I'll chat you up about whatever your interested in. Ironically, this would be more valuable to children and to lonely older folks so it wouldn't be much of a business and is probably more appropriate for professionals.
Someone has made that app
Probably, but I was thinking in person. People don't see eachother enough. One of my favorite things to do is chat people up at bars when I travel for work. If it doesn't work out, it just sends, if it does it's a perfect relationship. You have convo. then leave eachother never to meet again or to ever let the other down or forget a birthday or whatever. I collect transient friend memories.
Right is joint validation Wrong is partial or none
Pretty accurate…the entire calculation is also derived from a relatively simple formula, the cosine similarity, in case anyone was curious