101–110 of 149 posts
AlphaFold is optimization, not thinking. Propaganda 'r us.
Sure, but AlphaFold is still probably the most impactful and positive thing to have come out of "Deep Learning" so far.
Earlier quoted context omitted.
Not sure why this is downvoted. The comment cuts to the core of the "Intelligence vs. Curve-Fitting" debate. From my humble perspective as a PhD in the molecular biology /biophysics field you are fundamentally correct: AlphaFold is optimization (curve-fitting), not thinking. But calling it "propaganda" might be a slight oversimplification of why that optimization is useful. If you ask AlphaFold to predict a protein t…
I think if you watch the actual film you'd find they don't claim AlphaFold is thinking.
"Oh, machine learning certainly is not real learning! It is a purely statistical process, but perhaps you need to take some linear algebra. Okay... Now watch this machine learn some theoretical physics!"
"Of course chain-of-thought is not analogous to real thought. Goodness me, it was a metaphor! Okay... now let's see what ChatGPT is really thinking!"
"Nobody is claiming that LLMs are provably intelligent. We are Serious Scientists. We have a responsibility. Okay... now let's prove this LLM is intelligent by having it take a Putnam exam!"
One day AI researchers will be as honest as other researchers. Until then, Demis Hassabis will continue to tell people that MuZero improves via self-play. (MuZero is not capable of play and never will be)
Earlier quoted context omitted.
Parent said "entertainment use cases" are a complete waste, not all uses of images and video. I don't agree, but do particularly find educational use cases of AI video are becoming compelling. I help people turn wire rolling shelf racks into the base of their home studio, and AI can now create a "how to attach something to a wire shelf rack" without me having to do all the space and rack and equipment and lighting an…
If AI can produce movies, video and art better aka “more entertaining” then humans than how is it a waste?
Earlier quoted context omitted.
The chatbots and image editors are just a side-show. The real value is coming in e.g. chemistry (Alpha fold etc all), fusion research, weather prediction etc.
ML is used in weather prediction since the 80s and is the backbone of it since almost a decade. Not sure what are LLMs supposed to do there.
Earlier quoted context omitted.
I'm an hour into it, unconvinced. The illusion that agency 'emerges' from rules like games, is fundamentally absurd. This is the foundational illusion of mechanics. It's UFOlogy not science.
Why is it absurd? Because believing that would break some deep delusion humans have about themselves?
Look around you, look at the absolute shit people are believing, the hope that we have any more agency than machines... to use the language of the kids, is cope.
I have never considered myself particularly intelligent, which, I feel puts me at odds with many of HN readership, but I do always try to surround myself with myself with the smartest people I can.
The amount of them that have fallen down the stupidest rabbit holes i have ever seen really makes me think: as a species, we have no agency
Earlier quoted context omitted.
Not sure why this is downvoted. The comment cuts to the core of the "Intelligence vs. Curve-Fitting" debate. From my humble perspective as a PhD in the molecular biology /biophysics field you are fundamentally correct: AlphaFold is optimization (curve-fitting), not thinking. But calling it "propaganda" might be a slight oversimplification of why that optimization is useful. If you ask AlphaFold to predict a protein t…
If there's one thing I wish DeepMind did less of, it's conflating the protein folding problem with static structure prediction. The former is a grand challenge problem that remains 'unsolved' while the latter is an impressive achievment that really is optimization using a huge collection of prior knowledge. I've told John Moult, the organizer of CASP this (I used to "compete" in these things), and I think most people…
Edit to clarify my question: What useful techniques 1. Exist and are used now, and 2. Theoretically exist but have insurmountable engineering issues?
As a brit, I found it to be a really great documentary about the fact that you can be idealistic and still make it. There are, for sure, numerous reasons to give Deepmind shit: Alphabet, potential arms usage, "we're doing research, we're not responsible". The Oppenheimer aspect is not to be lost, we all have to take responsibility for wielding technology.
I was more anti-Deepmind than pro before this, but the truth is as I get older it's nicer to see someone embodying the aspiration of wanton benevolence (for whatever reason) based on scientific reasoning, than to not. To keep it away from the US and acknowledge the benefits of spreading the proverbial "love" to the benefit of all (US included) shows a level of consideration that should not be under-acknowledged.
I like this documentary. Does AGI and the search for it scare me? Hell yes. So do killer mutant spiders descending on earth post nuclear holocaust. It's all about probabilities. To be honest: disease X freaks me out more than a superintelligence built by an organisation willing to donate the research to solve the problems of disease X. Google are assbiscuits, but Deepmind point in the right direction (I know more about their weather and climate forecasting efforts). This at least gave me reason to think some heart is involved...
Earlier quoted context omitted.
If there's one thing I wish DeepMind did less of, it's conflating the protein folding problem with static structure prediction. The former is a grand challenge problem that remains 'unsolved' while the latter is an impressive achievment that really is optimization using a huge collection of prior knowledge. I've told John Moult, the organizer of CASP this (I used to "compete" in these things), and I think most people…
I'm really curious about this space: what types of simulation/prediction (if any) do you see as being the most useful? Edit to clarify my question: What useful techniques 1. Exist and are used now, and 2. Theoretically exist but have insurmountable engineering issues?
If your goal is to bring a drug to market, the most useful thing is predicting the outcome of the FDA drug approval process before you run all the clinical trials. Nobody has a foolproof method to do this, so failure rates at the clinical stage remain high (and it's unlikely you could create a useful predictive model for this).
Getting even more out there, you could in principle imagine an extremely high fidelity simulation model of humans that gave you detailed explanations of why a drug works but has side effects, and which patients would respond positively to the drug due to their genome or other factors. In principle, if you had that technology, you could iterate over large drug-like molecule libraries and just pick successful drugs (effective, few side effects, works for a large portion of the population). I would describe this as an insurmountable engineering issue because the space and time complexity is very high and we don't really know what level of fidelity is required to make useful predictions.
"Solving the protein folding problem" is really more of an academic exercise to answer a fundamental question; personally, I believe you could create successful drugs without knowing the structure of the target at all.
Earlier quoted context omitted.
Not sure why this is downvoted. The comment cuts to the core of the "Intelligence vs. Curve-Fitting" debate. From my humble perspective as a PhD in the molecular biology /biophysics field you are fundamentally correct: AlphaFold is optimization (curve-fitting), not thinking. But calling it "propaganda" might be a slight oversimplification of why that optimization is useful. If you ask AlphaFold to predict a protein t…
It seems that to solve the protein folding problem in a fundamental way would require solving chemistry, yet the big lie (or false hope) of reductionism is that discovering the fundamental laws of the universe such as quantum theory doesn't in fact help that much with figuring out the laws/dynamics at higher levels of abstraction such as chemistry. So, in the meantime (or perhaps for ever), we look for patterns rathe…
Earlier quoted context omitted.
I'm really curious about this space: what types of simulation/prediction (if any) do you see as being the most useful? Edit to clarify my question: What useful techniques 1. Exist and are used now, and 2. Theoretically exist but have insurmountable engineering issues?
Right now techniques that exist and used now are mostly around target discovery (identifying proteins in humans that can be targeted by a drug), protein structure prediction and function prediction. Identifying sites on the protein that can be bound by a drug is also pretty common. I worked on a project recently where our goal was to identify useful mutations to make to an engineered antibody so that it bound to a sp…