you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 09 Nov 2026
13 points (100.0% liked)
TechTakes
2780 readers
298 users here now
Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.
This is not debate club. Unless it’s amusing debate.
For actually-good tech, you want our NotAwfulTech community
founded 3 years ago
MODERATORS
But they really aren't! The actually interesting scaling laws from about 5-7 years ago that opened up machine learning to more applications all show that everything about machine learning is power-law diminishing returns with linear improvements in any metric requiring exponential increases in numbers of parameters / amounts of data / number of backprop cycles applied.
uh but have you considered unbounded recursive self-improvement? it’s how it works in books and in my imagination
The scaling laws maxed out around mid 2023 with the release of GPT-4 - that was around the time that everyone got excited about "test time inference" which is where LLMs spit out reasoning streams to convince you they are "thinking". If you look at where LLMs have improved since then it's mainly in tasks where they can be strapped into a harness and given concrete feedback like "your program didn't compile" which are also coincidentally use cases where data from gpt and Claude users can easily be harvested and new training data can be squeezed out of underpaid third world mturkers.
In other words they've improved their training data in the specific domains they are operating in like we've been doing in machine learning since the mid 90s...
The big labs have just been doing an incredible job of marketing (and particularly in Anthropic's case, brain damaging themselves into believing their own marketing)
These scaling laws are even funnier than that, as are more recent ones.
I continue to be extremely amused by the fact that you can take a LLM 'reasoning' trace, replace all the 'reasoning' with dashes or spaces, re-run the attention mechanism, and have what comes out the other end improve by a significant fraction as much as if you let it exude text that has the appearance of having to do with the topic at hand. It's just more opportunity for the embedding vectors to dance against each other before you translate the abstract vectors back into natural language. Hence recent frontier models doing a thing where instead of doing a long reasoning trace, they build the attention matrix, allow the vectors to push each other around... then let those new vectors push each other around again, and again, and again, multiple rounds, before you start decoding.
More recently, I have seen some vague math into the scaling of the speed with which running many parallel systems with access to a common pool of text produce a legible result. If I recall, it comes to roughly:
N parallel systems
Finish in ~1/N^0.5 as much time
At ~N^0.5x the cost
So 16 parallel systems crosstalking finish complex tasks in 1/4 the time at 4x the cost (and 16x the instantaneous rate of operation per second)