4 ms·
I only started believing in the singularity this year. When AI which could write code was demonstrated to the public at large, it became a short mental leap, co
by b800h 4y ago
I only started believing in the singularity this year. When AI which could write code was demonstrated to the public at large, it became a short mental leap, combining the tech with, say, genetic algorithms, to get to a runaway state. So either a singularity or a sigmoid curve which might as well be parabolic from our perspective?
- epicureanideal 4y ago[flagged]
- BoGoToTo 4y agoAlternate take: Regulation should be turned up to 11 and AI should be nationalized so that the benefits go to society instead of a handful of billionaires
- czbond 4y agoI don't think it should be nationalized. However, I do think the Treasury should just print the dollars, and buy OpenAI, etc and all profits go back to the American people.
- B1FF_PSUVM 4y agoDoes calling dibs stop at the border?
- 0xcde4c3db 4y agotheyrethesamepicture.jpg
- czbond 4y agoHa - in my view I always thought nationalized was "take over without recompense". So maybe they are the same picture that way.
- CatWChainsaw 4y agoMore train derailments, more building fires, more wealth hoarding, more medical malpractice, yaaaaay! :)
- jmull 4y ago> short mental leap It looks like climbing a tree to reach the moon to me. AI like chatgpt aren’t writing new code, they’re outputting code following their inputs, which is code found on the internet. That’s useful to an individual, but by its nature can’t go beyond what’s already there. If you hunt around on SO, you aren’t going to find code to generate an AI that can code a better AI. Because no one’s created that code. That’s how current AIs work too, so they aren’t going to be able to do it either, no matter how large and capable the models become.
- cma 4y agoAI already found that matrix multiply speedup.
- b800h 4y agoI disagree with this take, fundamentally, because both innovation through iteration and through cognitive leap are both possible with algorithms that are complementary to the likes of ChatGPT. The addition of mutation, recombination and selection to the code for a system with a variety of different types of neural net available will allow them to develop increasingly rapidly. In short, I don't think people are being imaginative enough about this stuff, given the number of additional ways in which it can be improved.
- visarga 4y agoHave I got a link for you: > Evolution through Large Models They are using a diff-LM trained on GitHub commits to generate mutations to a population of problem solutions expressed as code. https://arxiv.org/abs/2206.08896 https://arxiv.org/abs/2206.08896 The non intuitive bit is why are these models surpassing the original programming? It's because they got a secondary source for learning - the feedback they get from generating solutions and testing them. Learning through feedback, be it reinforcement, evolutionary or gradient based, is what can push AI forward. AlphaZero started with no human knowledge by pure self-play and beat all humans in a few days of practice. It learned from game outcomes as a feedback source. AI can learn from many things, not just from humans. But if it doesn't require humans it usually requires massive computation, simulation or search, maybe even real world testing which can get expensive.