The part I'm confused at is it doesn't seem that they are doomed, and end up being very successful companies. But I think this is likely due to lack of competition.
I recently did an internship at one of these big companies, doing ML. I'm a researcher but had a production role. Coming in everything was really weird to me from how they setup their machines to training and evaluation. I brought up that the way they were measuring their performance was wrong and could tell they overfit their data. They didn't believe me. But then it came to be affecting my role. So I fixed it, showed them, and then they were like "oh thanks, but we're moving on to transformers now." Main part of what I did is actually make their model robust and actually work on their customer data! (I constantly hear that "industry is better because we have customers so it has to work" but I'm waiting to see things work like promised...) Of course, their transformer model took way more to train and had all the same problems, but were hidden a few levels deeper due to them dramatically scaling data and model size.
I knew the ML research community had been overly focused on benchmarks but didn't realize how much worse it was in production environments. It just seems that metric hacking is the explicitly stated goal here. But I can't trust anyone to make ML models that themselves are metric hackers. The part that got me though is that I've always been told by industry people that if I added value to the company and made products better that the work (and thus I) would be valued. I did in an uncontestable manner, and I did not in an uncontestable way. I just thought we could make cool products AND make money at the same time. Didn't realize there was far more weight to the latter than the former. I know, I'm naive.
Yeah this is part of the issue with that particular product, the other is the initial capital. But also, the project itself was a bit too authoritarian style creepy so I'd rather not. But I've seen the exact same issues in MANY other products (I mean I could have told you rabbit or humane pin would be shit. In fact, I believe I even stated that on HN if not joked about it in person. I happily shit on plenty of papers too, and do so here)
I think what a lot of people don't understand is that there's criticism and dismissing. I'm an ML researcher, I criticize works because I want our field to be better and because I believe in ML, not because I'm against it. I think people confuse this. I'll criticize GPT all day, while also using it every day.
Mostly capital? Honestly, I have no idea how to get initial capital. Yeah, I know what site we're on lol. But I'm not from a top university and honestly I'd like to focus on actual AGI not this LLM stuff (LLMs are great, but lol they won't scale to AGI). Which arguably, if someone is wanting to compete in that space, why throw more money at a method that is prolific and so many have a head start? But they're momentum limited, throw me a few million and we can try new things. Don't even need half of what some of these companies are getting to produce shit that we all should know is shit and going to be shit from the get go.
I recently did an internship at one of these big companies, doing ML. I'm a researcher but had a production role. Coming in everything was really weird to me from how they setup their machines to training and evaluation. I brought up that the way they were measuring their performance was wrong and could tell they overfit their data. They didn't believe me. But then it came to be affecting my role. So I fixed it, showed them, and then they were like "oh thanks, but we're moving on to transformers now." Main part of what I did is actually make their model robust and actually work on their customer data! (I constantly hear that "industry is better because we have customers so it has to work" but I'm waiting to see things work like promised...) Of course, their transformer model took way more to train and had all the same problems, but were hidden a few levels deeper due to them dramatically scaling data and model size.
I knew the ML research community had been overly focused on benchmarks but didn't realize how much worse it was in production environments. It just seems that metric hacking is the explicitly stated goal here. But I can't trust anyone to make ML models that themselves are metric hackers. The part that got me though is that I've always been told by industry people that if I added value to the company and made products better that the work (and thus I) would be valued. I did in an uncontestable manner, and I did not in an uncontestable way. I just thought we could make cool products AND make money at the same time. Didn't realize there was far more weight to the latter than the former. I know, I'm naive.
reply