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

Terence Tao on O1

mathstodon.xyz

Post reply on HN
Print view View original

271–280 of 527 posts

Re: Terence Tao on O1

#271

Post by throwaway765123 »

macrolime wrote: post #218
delusional wrote: post #53

Earlier quoted context omitted.

What sort of problems do you solve? I tried to use it. I really did. I've been working on a tree edit distance implementation base on a paper from 95. Not novel stuff. I just can't get it to output anything coherent. The code rarely runs, it's written in absolutely terrible style, it doesn't follow any good practices for performant code. I've struggled with getting it to even implement the algorithm correctly, even t…

This kind of problems is really not where LLMs shine. Where you save loads of time is when you need to write lots of code using unfamiliar APIs. Especially when it's APIs you won't work with a lot and spending loads of time learning then would just be a waste of time. In these cases LLMs call tell you the correct API cells and it's easy to verify. The LLM isn't really solving some difficult technical problem, but sav…

This exactly. LLMs can't reason, so we shouldn't expect them to try. They can do translation extremely well, so things like converting descriptions to 90-95% correct code in 10-100x less time, or converting from one language to another, are the killer use cases IMO.

But expecting them to solve difficult unsolved problems is a fundamental misunderstanding of what they are under the hood.

Re: Terence Tao on O1

#272

Post by freeone3000 »

acedTrex wrote: post #270
viraptor wrote: post #268

Earlier quoted context omitted.

You're listing plain models, so I'm assuming you're using them directly. Aider and similar agents use those models but they don't step at the first answer. You can add test running and a linter to the request and it will essentially enter a loop like: what are the steps to solve (prompt)?; here's a map of the repository, which files do you need?; what's your proposed change?; here's the final change and the test run,…

That just sounds/looks like more work then just doing it normally? what am I missing?

You put it behind an API call and run the loop automatically for every coding query

Re: Terence Tao on O1

#273

Post by viraptor »

acedTrex wrote: post #270
viraptor wrote: post #268

Earlier quoted context omitted.

You're listing plain models, so I'm assuming you're using them directly. Aider and similar agents use those models but they don't step at the first answer. You can add test running and a linter to the request and it will essentially enter a loop like: what are the steps to solve (prompt)?; here's a map of the repository, which files do you need?; what's your proposed change?; here's the final change and the test run,…

That just sounds/looks like more work then just doing it normally? what am I missing?

Depends on the task but if you're going high level enough, it's not more work. Think about it this way: if you're doing proper development you're going to write code, tests and commit messages. Since you know what you want to achieve, write a really good commit message as the prompt, start writing tests and let the agent run in the meantime. Worst case, it doesn't work and you do the code yourself. Best case, it worked and you saved time.

(Not sure if that was clear but the steps/loop described before happens automatically, you're not babysitting it)

Re: Terence Tao on O1

#274

Post by tzs »

cjbgkagh wrote: post #152
bcoates wrote: post #135

Earlier quoted context omitted.

Re: Compensation expectations, I figured out a long time ago that bad programmers create bad code, and bad code creates work for good programmers. If the amount of bad code is no longer limited by the availability of workers who can be trained up to "just below average" and instead anyone who knows how to work a touchscreen can make AI slop, this opens up a big economic opportunity.

One could hope, but in my view perception precedes reality and even if that is the reality the perception is that AI will lower compensation demands and those doing the layoffs/hiring will act accordingly. You could also make the same claims about outsourcing, and while it appears that in most cases the outsourcing doesn't pay off, the perception that it would has really damaged CS as a career.

And like with outsourcing it starts with the jobs at the lower end of the skill range in an industry, and so people at the higher end don't worry about it, and later it expands and they learn that they too are not safe.

What happened a couple of decades ago in poetry [1] could happen now with programming:

> No longer is it just advertising jingles and limericks made in Haiti and Indonesia. It's quatrains, sonnets, and free-form verse being "outsourced" to India, the Philippines, Russia, and China.

...

> "Limericks are a small slice of the economy, and when people saw globalization creating instability there, a lot said, 'It's not my problem,'" says Karl Givens, an economist at Washington's Economic Policy Institute. "Now even those who work in iambic pentameter are feeling it."

[1] http://www.watleyreview.com/2003/111103-2.html

Re: Terence Tao on O1

#275

Post by namanyayg »

viraptor wrote: post #268
acedTrex wrote: post #264

Earlier quoted context omitted.

> I've found that I haven't written a line of code in weeks How are people doing this, none of the code that gpt4o/copilot/sonnet spit out i ever use because it never meets my standards. How are other people accepting the shit it spits out.

You're listing plain models, so I'm assuming you're using them directly. Aider and similar agents use those models but they don't step at the first answer. You can add test running and a linter to the request and it will essentially enter a loop like: what are the steps to solve (prompt)?; here's a map of the repository, which files do you need?; what's your proposed change?; here's the final change and the test run,…

I'm using Cursor and till now the "test run" part is manual, like Cursor doesn't care about testing or actually checking the code it wrote works

Any tips how I could integrate that? Do I need to switch to aider/plandex?

Re: Terence Tao on O1

#276

Post by energy123 »

dartos wrote: post #249
JumpCrisscross wrote: post #246

Earlier quoted context omitted.

The first profession AI seems on track to decimate is programming. In particular, the brilliant but remote and individual contributor. There is an obvious conflict of interest in this forum.

No, the first profession AI was on track to decimate was artists, but that didn’t really happen. AI just destroyed shutterstock.

The large majority of professional writers and artists produce thankless commodity output for things like TV advertisements, games, SEO content. These jobs should be threatened.

Re: Terence Tao on O1

#277

Post by Workaccount2 »

CamperBob2 wrote: post #52
wenc wrote: post #5

Once GPT is tuned more heavily on Lean (proof assistant) -- the way it is on Python -- I expect its usefulness for research level math to increase. I work in a field related to operations research (OR), and ChatGPT 4o has ingested enough of the OR literature that it's able to spit out very useful Mixed Integer Programming (MIP) formulations for many "problem shapes". For instance, I can give it a logic problem like "…

But many of the low-effort comments seem to mostly fall into (1) and (2) -- cynicism rather than cautious optimism. One good riposte to reflexive LLM-bashing is, "Isn't that just what a stochastic parrot would say?" Some HN'ers would dismiss a talking dog because the C code it wrote has a buffer overflow error.

It's understandable that people whose career and lifelong skill set that are seemingly on the precipice of obsolescence are going to be extremely hostile to that threat.

How many more years is senior swe work going to be a $175k/yr gig instead of an $75k check-what-the-robot-does gig?

Re: Terence Tao on O1

#278

Post by Workaccount2 »

skydhash wrote: post #68
WhatIsDukkha wrote: post #26

Earlier quoted context omitted.

I entirely agree about their utility. HN, and the internet in general, have become just an ocean of reactionary sandbagging and blather about how "useless" LLMs are. Meanwhile, in the real world, I've found that I haven't written a line of code in weeks. Just paragraphs of text that specify what I want and then guidance through and around pitfalls in a simple iterative loop of useful working code. It's entirely a lea…

> Much Much more productive world by just knuckling down and learning how to do the work. The fact everyone that say they've become more productive with LLMs won't say how exactly. I can talk about how VIM have make it more enjoyable to edit code (keybinding and motions), how Emacs is a good environment around text tooling (lisp machine), how I use technical books to further my learning (so many great books out here)…

Just be mindful that it is one's interest to push the "LLMs suck, don't waste your time with them" narrative once they figure out how to harness LLMs.

"Jason is a strong coder, and he despises AI tools!"

Re: Terence Tao on O1

#279

Post by mewpmewp2 »

acedTrex wrote: post #269
mewpmewp2 wrote: post #261

Earlier quoted context omitted.

Most gains are from using Copilot, do you use that?

I have it, tried it for a while. I have it turned mostly off new except for rare boilerplate heavy cases. It kept generating annoyingly wrong code. Things with subtly wrong misleading names, missing edge cases, ignoring immediate same file context etc. I found that it slowed me down so i turned it off.

Which language?

Re: Terence Tao on O1

#280

Post by IncreasePosts »

JumpCrisscross wrote: post #246
fnordpiglet wrote: post #149

Rewind your mind to 2019 and imagine reading a post that said “The experience seemed roughly on par with trying to advise a mediocre, but not completely incompetent, graduate student.” With regard to interacting with the equivalent of Alexa. That’s a remarkable difference in 5 years.

The first profession AI seems on track to decimate is programming. In particular, the brilliant but remote and individual contributor. There is an obvious conflict of interest in this forum.

Before it can replace the brilliant programmer, it needs to be able to replace the mediocre programmer. There is so much programming and other tech/it related work that businesses or people want, but can't justify paying even low tech salaries in America for.

So far, there is little chance of a non-technical person developing a technical solution to their problems using AI.

Post reply on HN