Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

Предыдущая публикация
OκULTRA XP🧿, writer for The Economist

how do you reconcile the claim that it is spicy autocorrect with recently demonstrated capacity (in "generic chat" specimen, not domain-optimized) to solve Erdős problems? What is your boundary between intelligence and "autocorrect"?

240
Uncle Eagle

Humans have been talking about Erdős problems online since the 90s... BBS, Forums, Reddit, Youtube, ect... LLMs have gobbled up ALL the Human written content (accurate or not) & utilized an order Matrix to organize it into tokens... Than it outputs the probable next word. Its an Autocomplete...

Обсуждение

1 прямых ответов · 8 сообщений

OκULTRA XP🧿, writer for The Economist

it solved a novel problem tho. Like, what's the supposed boundary between statistics based next word prediction and intelligence if the former is sufficient to solve math problems not in the training corpus ?

Ответ для OκULTRA XP🧿, writer for The Economist
Uncle Eagle

Bc Humans hadn't gotten around to seeing the patterns w/ higher dimensionality doesnt mean the pattern didnt exist. Machine Learning is VERY good at pattern recognition. I mean fuck, it found the pattern recognition in Human language. Its not thought or intelligence. Its good at finding patterns.

Ответ для Uncle Eagle
Uncle Eagle

Most humans have lost the ability to comprehend just how much data has been stolen by our Oligarchs. LLMs dont do math, just an API to a calculator to do the math & than match the output to more autocorrected best guess of next words again. Matching hundreds of existing work efforts to search thru.

Ответ для Uncle Eagle
Uncle Eagle

Im not shitting on value of Machine Learning & LLMs. That doesnt change that LLM/ML isnt thinking. Just pattern matching in a similar way of RegEx, just in an order matrix; using tokenizers. Its fucking impressive. Not magic. Oligarchic, Capitalist Hype doesn't make ML intelligence, just a tool.

Ответ для Uncle Eagle
OκULTRA XP🧿, writer for The Economist

I really don't understand what is "thought" or "intelligence" supposed to be for you, like, what exactly is an artifact supposed to functionally do for you to believe that it is "intelligent"? Also I think existence of construction of attention heads and self-attention in modern LLMs is >

Ответ для OκULTRA XP🧿, writer for The Economist
Ответ для OκULTRA XP🧿, writer for The Economist
Uncle Eagle

Hey comrade, if you believe the machine learning model that by definition is just an autocomplete, is better or similar to what Humans do already. Sure, make the argument. Sell the idea. I understand the math behind ML, & professionally study ML Its an autocomplete, maybe Humans are too (shrugs)

Ответ для Uncle Eagle
Uncle Eagle

I believe that ML IS a step closer to understanding Intelligence. But its missing a bunch of steps. Oligarchy is the problem. These tools are ready for Humanity scale. Treat LLMs like Autocomplete, bc chatbots are... "Language is primarily a tool for communication rather than thought"

gwern.net