51–60 of 69 posts
I still consider biological life as the best ‘robot’ because it can create more of itself. As long as robots are incapable of recreation I don’t see the threat. One could say all maschines today are infertile.
What about computer viruses?
Earlier quoted context omitted.
You mean exactly what 23andMe tried to do, and failed miserably at.
We are still early. Eventually you'll be able to change your race, gender, add reptile eyes, regrow limbs etc. Has to start somewhere. Need more data.
Reading their blog post I wonder if an LLMs is really the best way to do this. If I got it right, they used the LLM to enumerate potential protein DNA sequences. Does that really need an LLM? Enumeration is not novel, nor are LLMs particularily good at it. If you want to computationally parallelize the search in a large enumeration space it would be much easier to simply, well, do that instead of taking a detour via…
I worked on a predecessor to LLMs - HMMs for protein modelling. They were, and still are for most people the best way to model protein sequences. It's usually done as prediction, rather than generation (IE, you use the model to classify an unknown sequence into a known category, rather than asking the model to generate new instances of a category). HMMs for proteins are a bit stuffy, and they model local changes well, but struggle with long-range interactions that LLMs seem to excel at (for example, an HMM will do a good job of letting you stuff a few more residues into a protein in a localized region such as a hinge, but are not so great at modelling groups of residues that are located far-apart in sequence space but close in protein space).
One detail of the bitter lesson is, imho, that statistical parrots are better than they "should" be, probably for the same reason that mathematics is unexpectedly proficient in modelling physics: to some degree, the models recapitulate the true latent space of the underlying system well enough to generalize outside the original observation space.
Reading their blog post I wonder if an LLMs is really the best way to do this. If I got it right, they used the LLM to enumerate potential protein DNA sequences. Does that really need an LLM? Enumeration is not novel, nor are LLMs particularily good at it. If you want to computationally parallelize the search in a large enumeration space it would be much easier to simply, well, do that instead of taking a detour via…
It's not an English LLM, but a "protein" language model, where tokens represent amino acids or nucleotides. Learning a transformer language model on such data simply learns a distribution over sequences of tokens. It's a fine approach conceptually that in many ways is the "right" way or most elegant method, and not a stretch at all.
If statistical distributions can model higher level polypeptide structure, then it could be useful.
Earlier quoted context omitted.
We are still early. Eventually you'll be able to change your race, gender, add reptile eyes, regrow limbs etc. Has to start somewhere. Need more data.
Exactly. We should start building global database connecting DNA with medical history.
What can possibly go wrong if we let ChatGPT edit our DNA?
Earlier quoted context omitted.
You mean exactly what 23andMe tried to do, and failed miserably at.
We are still early. Eventually you'll be able to change your race, gender, add reptile eyes, regrow limbs etc. Has to start somewhere. Need more data.
Earlier quoted context omitted.
It's not an English LLM, but a "protein" language model, where tokens represent amino acids or nucleotides. Learning a transformer language model on such data simply learns a distribution over sequences of tokens. It's a fine approach conceptually that in many ways is the "right" way or most elegant method, and not a stretch at all.
I enjoyed the feeling when I made this connection talking with a startup doing this a while back. It's just a different "language" and although it's not a given that LLMs can operate in it, it's a reasonable thing to try, and it turns out they can.
I suspect we will need to move from sequential modelling to graphical modelling to level-up again, though.
Reading their blog post I wonder if an LLMs is really the best way to do this. If I got it right, they used the LLM to enumerate potential protein DNA sequences. Does that really need an LLM? Enumeration is not novel, nor are LLMs particularily good at it. If you want to computationally parallelize the search in a large enumeration space it would be much easier to simply, well, do that instead of taking a detour via…
Imagine an AI learning from photos/videos of a person and their DNA sequence? And also a list of diseases, health records, etc. Then asking it for predictions while giving it feedback afterwards so it can tune itself. You could even guarantee privacy. That would be some really useful data.
You mean exactly what 23andMe tried to do, and failed miserably at.