Two thoughts on someone who went out of their way to learn math:
1. If you can already program, the worst thing you can do is think of mathematics as learning a programming language. It is not, and you will waste your time being frustrated with things like syntax and notation. You get “used to” mathematics by doing it, and it’s something on its own. Just go with it. It’s ok to be confused.
2. Do the exercises, and stop asking for “solution manuals”, the point is to get you thinking and the struggle is most important part, not whether you got it “right”. Again, I think this is a programmer centric way of looking at things: “how do I know it’s
right if I can’t compile it”.
Maybe that’s why programmers like the foundations of mathematics. Like if somehow they could just go to the bottom of things, the assembler/machine code of sorts, the whole enterprise would make sense. Counterintuitively, the really great mathematicians of yore, did mathematics before it was anywhere close to formalized.
>Again, I think this is a programmer centric way of looking at things: “how do I know it’s right if I can’t compile it”.
I would argue this would only be true for those without formal education. Writing your code on paper is very common in CS courses. You get used to not being able to compile it.
I think your latter comment is kind of analogous to people writing python (or any high-level language) without understanding assembly. I think maybe that reduces the mystery a bit?
Another famous dude dumping his thoughts on HN who is gulping it up like an addict.
Add this to the long list of names like Terence Tao, and others who seem to be intellectually incontinent lately in the sense that one cannot navigate this space anymore without encountering their thoughts
We're talking about a deep human experience here. I don't know about HN as a whole but I personally come out of this much more touched about the human side of the story and how someone's life events can tremendously change their paths and goals than the LLM itself.
Same. And the insight of having used LLMs and it being so capable is a wtf moment for me in itself. And I am an AI engineer. It’s my job to have a good idea about how far these models can take us.
> And the insight of having used LLMs and it being so capable is a wtf moment for me in itself. And I am an AI engineer. It’s my job to have a good idea about how far these models can take us.
Are you saying that, as an 'AI engineer', you were unaware that LLMs could be used to interpret genetic variants? A thing that Google has been publishing on for well over 10 years?
This was not intended to be a technical post (obvious I hope).
I'm planning on getting one out in the next few weeks characterizing the system and how it performed on real clinical use-cases vs. alternatives and existing tools.
The TL;DR is that Gamow Labs is a harness and interface company on top of SOTA LLMs as you suggested, but my harness and interface outperforms the existing thing. While this approach would have earned me the "wrapper company" label last year, I hope the success of OpenEvidence, Harvey, Perplexity, and so on has opened minds with respect to the value here.
It was only working through clinical cases that I realized how much more I needed beyond dropping raw reads into Codex.
Can you say a little more about how current genomic techniques rely on human interpretation? Is that mostly where you use an LLM (to act human-like), or is your approach different than that?
Current genomic techniques involve humans using a lot of different software (search, ranking models, visualizations, alignment algorithms, etc.) and synthesizing the results manually into a diagnosis.
Your assumption is correct about my technique. I cloned (and expanded) this workflow into an LLM harness, so the LLM is basically orchestrating a bunch of tools that normally humans would use (and writing the conclusions and doing all the standard LLM stuff).
In "Language Models are Unsupervised Multitask Learners"[0]. Not sure whether it’s "the" GPT-2 paper.
3.7 Translation
> Performance on this task was surprising to us, since we deliberately removed non-English
webpages from WebText as a filtering step. In order to con-
firm this, we ran a byte-level language detector2 on WebText
which detected only 10MB of data in the French language […]
First, something lying sometimes doesn't mean it lies all the time.
Second, the whole point of LLMs is inference - they use massive amounts of amalgamated information to produce answers. The Olympiad math problems are not frontier mathematics requiring ideation, they are complex examples of existing problems. That means they're exactly the sort of thing an LLM with enough training data is good at.
The question of whether recombining existing knowledge is all it takes to be "creative" or produce things which are novel is an open one, but I don't think this is contradictory on its face.
> built with American capital and mostly American minds.
I would say "built with American agency and commercial spirit", not minds.
Most of the things that we have were first built elsewhere (Germany being a prime supplier here with the mp3 or the Zuse), but turning them commercial was the input that came from America.
What was the reason for those protests? Was it perhaps economic hardship brought about by US sanctions? How much is the US liable for the suffering of the Iranian people?
(A lot, is the answer)
That doesn't excuse the Iranian regime, but the US is not exactly helping, is it.
It was hardship brought on by not attempting to address the problems. Sanctions made things a bit worse but if Iran put effort into ensuring there was fresh water instead of funding terrorists and building missles things would have been a lot better for the people. (And likely no senctions for those things)
A bit worse? The sanctions directly brought about this. Scott Bessent admitted -- unprompted -- that the purpose of the sanctions was to destroy the Iranian economy.
I'm not saying the regime is good. It's not. It's terrible. But that does not change what the US has done.
The US has consistently made the suffering in Iran worse over the years. And let's not forget that the US and the British caused the Islamic revolutionaries to come into power by installing a puppet Shah that was deeply unpopular.
This really must be a very math-starved community of people who wanted to learn math but never quite could.