I actually thought the opposite - that he seems to be seriously thinking about AGI from a broader intellectual standpoint than most ML researchers. With that said, I was a little confused when he said evolution was more optimized for locomotion/vision than language. Like yes, language is super recent, but communication in general is not.
Yup, Dwarkesh needs to broaden his intellectual scope, and the Sutton interview completely exposed the echo chamber he's been inhabiting. There is no certainty in science, and I don't think building 'AGI' will be any exception.
But there is no way to know who is truly the 'best'. The people who position and market themselves to be viewed as the best are the only ones who even have a chance to be viewed as such. So if you're a great researcher but don't project yourself that way, no one will ever know you're a great researcher (except for the other great researchers who aren't really invested in communicating how great you are). The system seems to incentivize people to not only optimize for their output but also their image. This isn't a bad thing per se, but is sort of antithetical to the whole shoulder of giants ethos of science.
My practical interpretation of the EMH is more that easily accessible, public information is already priced in. But non-obvious insights may not be simply because the volume of people trading on that information will be smaller.
The answer stems from McCarthy’s deeply disparaging view of modern society, which he considered lost, divorced from nature, history and tradition and heading toward social collapse and apocalypse. “Cormac considered contemporary fiction a waste of time,” said Dennis, “because contemporary writers no longer have a legitimate culture to feed their souls.”
One fundamental challenge to me is that if each training run because more and more expensive, the time it takes it to learn what works/doesn't work widens. Half a billion dollars for training a model is already nuts, but if it takes 100 iterations to perfect it, you've cumulatively spent 50 billion dollars... Smaller models may actually be where rapid innovation continues simply because of tighter feedback loops. O3 may be an example of this.
I don't think he got taken for a ride. Rather, he also wanted to believe that AlphaChip would be as revolutionary as it claimed to be and chose to ignore Chaterjee's reservations. Understandable, given all the AlphaX models coming out around that timeframe.
"Topology is all that matters" --> bold statement, especially when you read the paper. The original authors were much more reserved in terms of their conclusions.
To call this memory seems like a stretch. By the logic of the article, every daughter cell has 'memory' of the parent cell because some proteins from the parent cell are present in the daughter cell. I would be curious to see p53 complex concentration as a function of cell generation/mitosis time to show how durable this 'memory' actually is.
while potentially interesting work, very shortsighted and premature to say this is a "GPT" moment in biology. ML people in bio need to think hard not only about what they are doing, but why are they are doing it (other than this is cool and will lead to a nice Nature publication). Their basic premise (learning from DNA is the next grand challenge in biology) is shaky. Imo, the grand challenge in biology is determining what the grand challenge is, and that is a deep scientific/philosophical question.
if you just want count/location, super resolution techniques (https://www.science.org/doi/10.1126/science.ade2676) and proximity labeling (https://www.biorxiv.org/content/10.1101/2023.10.28.564055v1) may be a good starting point. cryo may be able to help with that if the direct electron detectors get better (as my understanding (and in my experience), cryo-et data is quite noisy). my guess is that multiple techniques will have to be combined and processed with sophisticated computational pipelines to make this a reality. each technique provides some information on the state of the cell, so the computational question becomes if can you figure out how to integrate information between these techniques to get the specific information you want. it may be too early to tackle this, but who knows...
Sure, nothing fundamentally prevents us from repairing it from a physical standpoint. The practicality of it is the key question. While the machine analogy is partially useful, genes/proteins/cells are dynamic, adaptive, and can exhibit stochastic traits. This fundamentally contrasts them from traditional machines, and is the reason (imo) we suck at making effective therapies to even treat diseases where we think we know what is going on.