5 ms·
> The roadmap to success will most likely start with training or fine-tuning on data from human professionals performing the task, and proceed with reinforcemen
by dasil003 1y ago
> The roadmap to success will most likely start with training or fine-tuning on data from human professionals performing the task, and proceed with reinforcement learning in custom environments designed to capture more of the complexity of what people do in their jobs.
> [...]
> We think this is essentially a data problem, not an algorithms problem.
This is extremely hand-wavy. How are you going to instrument the various thought processes and non-verbal communication that goes into building successful software? A huge part of it is intuition about what makes sense to other humans. It's related to the idea of common sense, but in the software world there's this layer of unforgiving determinism and rigidity that most humans don't want to deal with. I just don't see how AI crosses that chasm.
- ege_erdil 1y agohow do you think humans cross that chasm?
- AnimalMuppet 1y agoPretty sure that however humans crossed it, it wasn't just "a data problem".
- ege_erdil 1y agowhat makes you sure about that?
- AnimalMuppet 1y agoAt the risk of coming across as a smart aleck, 40 years of experience building software. What experienced software engineers have is a sense of taste - this looks like good code and/or design, that doesn't. But they don't have data; they have, at best, a couple of anecdotes. It's more a sense of "that was harder to work with than it should have been; that approach seems to have drawbacks". But you only get a few examples of that in a career. And there are very few outfits compiling usable data that could shape the approaches that software engineers use. So I don't think how humans got there was primarily data.
- ege_erdil 1y agothat's not the relevant data i'm talking about how much real-world data do you think went into the evolution of the human brain and all its learning algorithms? having 40 years of experience building software gives you no more insight into that than having 40 years of experience using language gives you insight into where your language skills come from
- AnimalMuppet 1y agoThe discussion is about algorithms automating writing software. The human brain did not evolve to write software. It evolved to get a good enough answer fast enough to survive, not to solve the kind of problem that is involved in software. Yeah, I know, the human brain is able to learn to do software. But that isn't what it evolved to do.