AlphaCode
Hello, what do you think of what is said in this article:
"Creating solutions to unforeseen problems is second nature in human intelligence – a result of critical thinking informed by experience. The machine learning community has made tremendous progress in generating and understanding textual data, but advances in problem solving remain limited to relatively simple maths and programming problems, or else retrieving and copying existing solutions. As part of DeepMind’s mission to solve intelligence, we created a system called AlphaCode that writes computer programs at a competitive level. AlphaCode achieved an estimated rank within the top 54% of participants in programming competitions by solving new problems that require a combination of critical thinking, logic, algorithms, coding, and natural language understanding.
In our preprint, we detail AlphaCode, which uses transformer-based language models to generate code at an unprecedented scale, and then smartly filters to a small set of promising programs."
https://deepmind.com/blog/article/Competitive-programming-with-AlphaCode
"Creating solutions to unforeseen problems is second nature in human intelligence – a result of critical thinking informed by experience. The machine learning community has made tremendous progress in generating and understanding textual data, but advances in problem solving remain limited to relatively simple maths and programming problems, or else retrieving and copying existing solutions. As part of DeepMind’s mission to solve intelligence, we created a system called AlphaCode that writes computer programs at a competitive level. AlphaCode achieved an estimated rank within the top 54% of participants in programming competitions by solving new problems that require a combination of critical thinking, logic, algorithms, coding, and natural language understanding.
In our preprint, we detail AlphaCode, which uses transformer-based language models to generate code at an unprecedented scale, and then smartly filters to a small set of promising programs."
https://deepmind.com/blog/article/Competitive-programming-with-AlphaCode
The entire concept of competitive programming... PFFFFFFFFFFFFFT!
Competitive coding tends to discourage writing good, easily maintainable code.
Beyond that small glimpse of my disdain you really don't want to know what I think of that drivel.
Competitive coding tends to discourage writing good, easily maintainable code.
Beyond that small glimpse of my disdain you really don't want to know what I think of that drivel.
Ok, I get it. And I think it's like asking an artificial intelligence to write poetry is silly. But if it's a machine that can write code for itself, does it really need a human to maintain it? Because the end goal is not to create code for humans, but to make a machine that learns and transforms its own code.
(laughs) To read the book of Shoshana Zuboff one is not far from being at the mercy of some machine...
but ok thanks for the insight.
but ok thanks for the insight.
Last edited on
You wanted opinions, I gave one.
You don't like what I have to say. Waaaah!!
Ciao.
You don't like what I have to say. Waaaah!!
Ciao.
While we need to take new approaches if we want breakthroughs, and this could eventually lead to it, I am not impressed. Now, if you said that 100% of the entries worked 100% correctly and the scoring was simply performance and style, then being in the top half is impressive. If its a typical coding site where people submit total garbage that does not work for all inputs or has whatever flaws, then this is like winning the special olympics -- exciting for you, but not really for anyone else.
That said, my hopes for AI have fallen a lot over the decades. Even as it gets better (in part by throwing todays hardware at it, which is yesterday's supercomputer) and we make new inroads, the results are what they have always been. And that is that we can teach a computer to get about 85% correct results on just about anything that has testable results. That last 15% has eluded us for 40 years for the bulk of problems. Solve that, and you have made some progress. Stuff like identify all the objects in a picture... get me 99% accuracy over thousands of images on that, however you do it, THAT is progress.
I am not down on the idea or their research, mind you. We need a new approach, because the old stuff isnt working. Unfortunately, so far, the new stuff is also not working :)
That said, my hopes for AI have fallen a lot over the decades. Even as it gets better (in part by throwing todays hardware at it, which is yesterday's supercomputer) and we make new inroads, the results are what they have always been. And that is that we can teach a computer to get about 85% correct results on just about anything that has testable results. That last 15% has eluded us for 40 years for the bulk of problems. Solve that, and you have made some progress. Stuff like identify all the objects in a picture... get me 99% accuracy over thousands of images on that, however you do it, THAT is progress.
I am not down on the idea or their research, mind you. We need a new approach, because the old stuff isnt working. Unfortunately, so far, the new stuff is also not working :)
Topic archived. No new replies allowed.