Complete and utter nonsense, and factually incorrect to boot. There is tons of proof that progress is slowing and very little it isn't. Also what breakthroughs in synthetic data and "computing infrastructure" will make AGI happen? Why does nobody ever push back on this??
Comments
I'm watching/reading discussion on it and the claims being made are absurd.
Just literally creating an endless recursive loop of synthetic data, leads to persistent degradation in quality.
Oh and they quickly bring up how it is *cheaper*.
Which will of course vary around some sort of point that may or may not be biased or warped.
tangibles inform on why...
-texts in languages other than English must be used to train LLMs, no? How does this affect the dearth of training material?
-English lit has texts that avoid/subvert meaning, eg Surrealist lit. Are these used? How do they affect the models' functioning?
https://bsky.app/profile/did:plc:eaxgzdvtvnsl4ambjx3j5kv6/post/3lcibx6anak2y
I’m wondering what this synthetic data even looks like.
How can you make up fake data? Aren’t you just replicating/copying actual data? Otherwise it’s not factual and will cause more hallucinations??
https://www.wheresyoured.at/bubble-trouble/#:~:text=The%20amount%20of%20data%20it%20needs%20is%20several%20multitudes%20more%20than%20currently%20exists
At first it seemed impossible that all the data we’ve created for thousands of years is running out. But it also kind of makes sense if that’s all the AI has been doing for two years: devouring data.
Maybe we are not all doomed by AI
Truly the future.
“Aha, he says it’s true. I’ve heard others say it’s true. So it is true.”
“Oh. That’s two people just today that I heard say it’s true. Obviously it’s true because why else would so many people be saying it’s true?”
*iterate for years*
Step 2 - 🪄Magic🪄
Step 3 - CONSCIOUSNESS!!!
G rift
I ntelligence