I noticed this a few times I tried using ai to solve a wordle I was about to give up on. It isn’t just worthless, it like can not guess a 5 letter word based on a few rules like “second letter not a, d or g; first letter l, last letter w; word is not " lower”
Its not that it can’t solve it, it can’t even guess slightly correct. I think ai language isn’t as connected to “spelling” as we think, I’ve heard of people using ai to translate instructions to Mandarin and then feeding the Mandarin query in because Mandarin is more “meaning dense” therefore uses less tokens and gets better answers
yeah they do not think nor reason just most likely token prediction, most obv in small models but larger are a bit better at making you believe. emperors new clothes, i bet the house of cards will go down soon enough going to be interesting times at least
All AI companies except deepseek use buttholes as their logos
I feel like they are doing it on purpose. Recently at work we got a bunch of Claude emojii in slack. There is one with a hand rubbing the logo or patting it or something. No one else seems to be noticing that it looks like someone rubbing a butthole lol.
Pics or it didn’t happen
🫨
All = truly all?
There are two r’s in strawberry.
Should ask him which day doesn’t end in “y” next time.
Tap for spoiler
Two don’t, since monda doesn’t end in “y”.
This technology is THEFUTURE if you don’t choke on its dick so hard it looks like you’re trying to fold your own erection between your ass cheeks you’re going to megahell if you don’t understand why we can’t go back to a time without it you’re just an insane Luddite moron
AI doesn’t see a word as a sequence of letters, they just see it as a pointer to an entry in Words table.
Semantic Vectors don’t work that way.
Yeah, if words were actually encoded as 1-hot vectors this would be pretty trivial, but the rest of LLM training would be somewhere between infeasible and impossible. The actual embedding vectors obscure spelling even more.
Side note: last time I checked, current embedding vectors were approximately 40 dimensional… Has that gone up significantly in the last couple of years?
A fair bit. EmbeddingGemma is open weights and allows for 128-768 dimensions.
It’s not as simple as more dimensions = better, due to size, efficiency, and context rot limitations though.
Introducing EmbeddingGemma: The Best-in-Class Open Model for On-Device Embeddings - Google Developers Blog - https://developers.googleblog.com/en/introducing-embeddinggemma/
Shouldn’t it help that it separated them out with underlines? How does this text break down in terms of tokens?
Oh thank God. I was worried that I was really stupid.
Every day except for Tuesday and Thursday are the ancient enemies of Imu and the World Government.
