//sci/
File: NLP.png (89.1 KB)
89.1 KB
>"I made it up" the field
Showing all 16 replies.
>>
What is exactly is wrong with natural language processing?
>>
>>16996236
didn't the guy who invented that die in a murder-suicide?
>>
The book lexicon is a scifinthriller that takes this to the extreme and has poets who can control others with words. Pretty cool book.
>>
The nexium cult used some of these techniques.
>>
File: cogsci.png (29.5 KB)
29.5 KB
>>16996236
>>
>>16996249
I study computational neuroscience and it is peak
>>
>>16996240
NLP is a bold faced dead end. Language is far too complex to be studied with just numbers.
>>
>>16996267
the numbers are just symbols of light that "compute" in a box. And humans are the same, light in a box producing language.
>>
>>16996236
I've written my Master's thesis on NLP and was quite interested in the topic, then one year later ChatGPT became a thing and has made me into an absolute cynic about anything AI related
>>
>>16996296
shouldnt new advances have excited you, what gives?
>>
>>16996319
That outside of academia GenAI is used for the lamest fucking shit.
On a technical level it's so fascinating how a bigass equation predicting some words can have what feels like a real conversation with you. But to me it's kinda overshadowed by the fact in the process it's ruing the internet, creative fields and society as whole.
>>
>>16996267
Not numbers, space. A lot of NLP methods are just dimensionality reduction techniques from linear algebra. Also LLMs are all about word embeddings.
>>
>>16996332
AI as it's currently known is a shadow ghost of a thought and should not be relied upon for accuracy, veracity, or synthesis.
Memory tech could 40x in density yesterday and they could shove 10s trillions parameters in a model and the entirety of all written and spoken thought into the dataset and the systems will still be this way.
Because its just a prediction model.
Intelligence is not a prediction model. We don't predict that the sequence of words should be equivalent to Bayes' theorem. We know the preceding information is (taken as) true and synthesize new information which tests to also be true.

LLMs will never be able to do this. Yeah the maths and development is cool and all but I instantly degrade anyone who praises these things down to a worm.
>>
>>16996409
"We know the preceding information is true and synthesize new information which tests to also be true".

What you just described is pretty close to statistics, which is centrally about testable information. You should read the book "A Brief History of Intelligence". A defining trait of humans and highly intelligent beings is that they actually simulate/predict the world rather than experience it directly. That is why humans are capable of hallucinations and all sorts of crazy shit.

All vertebrate animals rely on prediction just to move. Giraffes start life on shaky legs because they have to develop a model of how their body will react to inputs. Its all statistical pattern matching, always was.
>>
>>16996413
let me guess the universe is a simulation
there's just no end with people like you
I'm not predicting where my phone just went on the table, I deliberately moved it there.
learning =/= learned
>>
File: thumbnail.png (10.8 KB)
10.8 KB
>dude i learned how to sort a sequence and now i can solved intelligence lol it's just this watered down math course i took as a cs major LMAO

Reply to Thread #16996236


Supported: JPG, PNG, GIF, WebP, WebM, MP4, MP3 (max 4MB)