Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AutoDev: Automated AI-driven development by Microsoft

arxiv.org

Post reply on HN
Print view View original

161–170 of 232 posts

Re: AutoDev: Automated AI-driven development by Microsoft

#161

Post by osigurdson »

polycaster wrote: post #86
osigurdson wrote: post #80

Earlier quoted context omitted.

>> Most dev jobs are mostly repetitive plumbing Those jobs should go away. Basically, the elimination of anything boring is ultimately a net good for humanity.

Hey over there. I’m very much grateful for the privilege of this boring job, which is not 100% of my job, but a huge part of it. Grateful because it allows me to feed a family of four. I’m sure in your Musk’esque utopia without boring work is place enough for all mankind. But please, don’t forget to draft a bridge that will bring us all over there and not just a bunch of filthy rich Silicon Valley assholes. Because t…

That is a fair assessment. I am probably parroting Musk here a little. However, your main issue is access to food and resources, not boring work. I can't see why the price of everything goes to zero if there is no cost to make it.

Re: AutoDev: Automated AI-driven development by Microsoft

#162

Post by mkleczek »

93po wrote: post #139
mkleczek wrote: post #61

Earlier quoted context omitted.

I am quite confident my job is not in danger. Sure - scaffolding and boilerplate can be automated using ML - but problem solving - not at all. The reason is simple - to find a solution is to create a program that has a specific set of properties. This - by Rice's Theorem - is not possible to realise using a Turing Machine. So as long as the solution search space is small (ie. we are talking about very similar program…

The problem is software development is 95% stuff that has been done a million times before. If you disagree you’re probably not exposed to enough of the industry to see this.

That would mean 95% of problems are already solved.

I don’t agree with this - the space of problems to solve is as big as that of solutions (both are programs in the end).

I don’t think suggestion about lack of experience deserves any response.

Re: AutoDev: Automated AI-driven development by Microsoft

#163

Post by osigurdson »

Sindisil wrote: post #111
osigurdson wrote: post #80

Earlier quoted context omitted.

>> Most dev jobs are mostly repetitive plumbing Those jobs should go away. Basically, the elimination of anything boring is ultimately a net good for humanity.

Neat idea. That solves everything. Oh, one quick thing. I'm sure it's nothing, but I'm a bit slow. How do you get new experts if no one gets to do the junior work that gives them the experience to become an expert?

I guess in the ideal world, everyone just does what they want, since everything will be so cheap. Enjoy chess? Study it and play against other humans. AI will of course crush you in any game. Enjoy accounting, radiology or programming? Same thing.

Re: AutoDev: Automated AI-driven development by Microsoft

#164

Post by mkleczek »

Smaug123 wrote: post #82
mkleczek wrote: post #61

Earlier quoted context omitted.

I am quite confident my job is not in danger. Sure - scaffolding and boilerplate can be automated using ML - but problem solving - not at all. The reason is simple - to find a solution is to create a program that has a specific set of properties. This - by Rice's Theorem - is not possible to realise using a Turing Machine. So as long as the solution search space is small (ie. we are talking about very similar program…

The world contains a great many arguments invoking Rice (or equivalently the halting problem) to "prove" non-mathematical statements, and they are almost all specious. a) the mere fact that there is no fully general procedure to do some particular task doesn't imply some human can do that task any better than the best procedure we can find, and b) even if you believe there's something noncomputable about the human mi…

Until a computer can solve the problem “how to redesign this particular database design (tables/partitioning/indexes/queries) so that it can handle expected load in next 5 years” I can safely assume it does not hold :)

Re: AutoDev: Automated AI-driven development by Microsoft

#165

Post by margorczynski »

osigurdson wrote: post #161
polycaster wrote: post #86

Earlier quoted context omitted.

Hey over there. I’m very much grateful for the privilege of this boring job, which is not 100% of my job, but a huge part of it. Grateful because it allows me to feed a family of four. I’m sure in your Musk’esque utopia without boring work is place enough for all mankind. But please, don’t forget to draft a bridge that will bring us all over there and not just a bunch of filthy rich Silicon Valley assholes. Because t…

That is a fair assessment. I am probably parroting Musk here a little. However, your main issue is access to food and resources, not boring work. I can't see why the price of everything goes to zero if there is no cost to make it.

Because there's a finite amount of resources so almost always you'll run into scarcity?

Very well the wages might fall much quicker than the costs so for a handful it will be beneficial, for the rest not so much.

Re: AutoDev: Automated AI-driven development by Microsoft

#166

Post by onthecanposting »

_therealtoogy wrote: post #143

Maybe ignorant, but if AI can get to a point of fully automating SWEs, hardly any white-collar knowledge based job is safe.

I think this is OK. By extension, the services those knowledge workers provide become trivial costs. World class medical advice doesnt cost $500/hr with a month-long wait. Getting a freeway interchange designed doesn't cost the public $20M and take 5 years. Business logic for large organizations would be handled by something other than a labrynth of excel spreadsheets.

Re: AutoDev: Automated AI-driven development by Microsoft

#167

Post by Terretta »

_therealtoogy wrote: post #143

Maybe ignorant, but if AI can get to a point of fully automating SWEs, hardly any white-collar knowledge based job is safe.

Not sure all jobs in software are white collar. Some are blue.

Consider a hierarchy:

- coder: translates requirements into code

- developer: comes up with software to meet specified goals

- engineer: decides what why and how to do, buy vs. build, do this with humans or software, approach and materials ... understands the multi modality stuff of which the outcome is made and deploys it effectively

"Devs" or "Engineers" who are actually "Coders" are at risk, as that work is not really "white collar". Turning requirements into code is piece work on an assembly line, blue collar at a keyboard.

LLMs are already better than most of those, even though it's engineers here who are saying LLMs don't cut it. Both things are true.

What engineers might want to wrap their head around is using LLMs as apprentices, leverage, force multipliers, so a "team lead" is leading a team of junior coders, LLMs that type.

Re: AutoDev: Automated AI-driven development by Microsoft

#168

Post by sobasically »

PodgieTar wrote: post #34

I guess I'd be interested to see how this performs against the same benchmark Devin was using. It's hard to deny that this isn't impressive. But I think there's two interesting parts to it. Claude 3 Opus already scored around 85-86% on these benchmarks, without an "AutoDev" style agentic approach. And all the same problems with HumanEval remain, the limitations in terms of what style of problems are chosen, and real…

They still use “agents” to make Opus. This is fancy syntax sugar for “while ! EOF; read next chunk of data of size N, do XY or Z with it”

It’s recursion and memoization to avoid fractalness all the way down. We keep trying to make these language bubbles that mean something but they mean nothing to the grand churn of the universe. The effort to so strictly and specifically codify a generalized, endless, mechanics of reality is a wacky hallucination humans keep diving into

Re: AutoDev: Automated AI-driven development by Microsoft

#169

Post by throwuwu »

dudeinhawaii wrote: post #158

My hopes (and fears) about AI don't seem to match the reality which is that, until we reach AGI, humans are the sole source of creativity and novelty. I think about it like this, would ChatGPT invent Google Search if we had ChatGPT in 2000? Probably not. LLMs seem to exist in this realm of "as smart as an average human with a really big encyclopedia". They're confidently wrong, invent fantasy to defend bad reasoning,…

What makes you so sure that problem solving and invention aren’t just engineering challenges that we can solve by combining LLMs with well designed algorithms? The way I see it, we’ve just discovered something fundamental like steam power or electricity and we’re currently in the very inefficient stage of brute force solutions like mine pumps driven by condensing engines that needed tonnes of cheap readily available coal and arc lamps running off an entire room full of galvanic piles. In other words, we’re just throwing shit at the wall and seeing what sticks; but stick it will and then we’ll quickly be on to locomotives and lightbulbs.

Re: AutoDev: Automated AI-driven development by Microsoft

#170

Post by Smaug123 »

mkleczek wrote: post #164
Smaug123 wrote: post #82

Earlier quoted context omitted.

The world contains a great many arguments invoking Rice (or equivalently the halting problem) to "prove" non-mathematical statements, and they are almost all specious. a) the mere fact that there is no fully general procedure to do some particular task doesn't imply some human can do that task any better than the best procedure we can find, and b) even if you believe there's something noncomputable about the human mi…

Until a computer can solve the problem “how to redesign this particular database design (tables/partitioning/indexes/queries) so that it can handle expected load in next 5 years” I can safely assume it does not hold :)

Ah, now you're talking price! How many years do you bet before this happens - three? Five?
Post reply on HN