Lars Marowsky-Brée 😷Tell me again how #GenAI will extract meaningful trends from and answer queries about your data set.
I can also see this going great for coding, programming languages and computers are known to be very forgiving and tolerant
Theuni@larsmb It's fantastic. Here's the FastGPT version of it. It greatly demonstrates that it doesn't have the ability to count / compute but only combines textual artefacts.
@larsmb And obviously it immediately fucks up any arbitrary counting. And then it gives a reference to some site that obviously doesn't have that "word" in it ...
Vlad 🇺🇦@vladimir_lu @larsmb I'm assuming you're ironic, but there is a point worth discussion. Mathematical models are intended to cover all general cases. Specifically in computer science we've been bitten by exactly the idea of "nobody will ever put this in" (see y2k bug). We've been through this. And on a more anthropological level good math (and physics) gets extended and remains valuable even in shifting contexts. LLMs are garbage in that way.
Barry Rowlingson@theuni @vladimir_lu @larsmb It might get asked "How many C base pairs are in the DNA sequence CCTGAGATCTAGGAGGGCATCCGC?"
@geospacedman @vladimir_lu @larsmb There are no C in DNA sequences. This is an often made confusion as there are only G, T and A in DNA as the C looks much alike the G.
Tom Ritchford@theuni @geospacedman @vladimir_lu @larsmb
Assuming you aren't joking:
https://en.wikipedia.org/wiki/Nucleic_acid_sequence
"The possible letters are A, C, G, and T, representing the four nucleotide bases of a DNA strand – adenine, cytosine, guanine, thymine"
@theuni @geospacedman @vladimir_lu @larsmb haha!
People don't recognize satire on the internet, endless studies have shown...
Peter Bindels@geospacedman @theuni @vladimir_lu @larsmb statistically you are over 99% likely to have GATTACA in your DNA.
Shark Attak@geospacedman @theuni @vladimir_lu @larsmb
This sounds like the premise of a bad sci-fi horror movie..
This sounds like the premise of a bad sci-fi horror movie..
Magnus Ahltorp@theuni @vladimir_lu @larsmb The whole point of LLMs is to *not* do things “correctly”. The LLM is doing it’s job perfectly here, it’s just that the expectation is wrong.
@theuni Oh, I wasn't actually aware that someone else had found this before! Nice. I had actually stumbled across it independently.
But it is very funny to me that this is the level of technology that certain people want to ram into everything and are burning 100+ Terawatt hours per year by now.
Sure, GPTs, LLMs, ML in general, very cool stuff. But also, as ready as a wet noodle.
Florian Idelberger
Georg Arne Spenden
IBBoard
echopapa
Subjekt_91
Tartas1995
Father EnochI'm sorry for my ignorance but I thought THE WHOLE POINT of #computers was that they could count.
For three to four decades I was told computers were a one trick pony, with the trick being able to count faster than humans ever could.
WHY are we making computers that can't count?