DARPA suggests turning old C code automatically into Rust

[SiliconWizard]

This one is pretty good:

https://www.theregister.com/2024/08/03/darpa_c_to_rust/

[xrunner]

TRACTOR: TRanslating All C TO Rust

I knew it involved AI as soon as I saw your post. 

[mfro]

"You can go to any of the LLM websites, start chatting with one of the AI chatbots, and all you need to say is 'here's some C code, please translate it to safe idiomatic Rust code,' cut, paste, and something comes out, and it's often very good, but not always,"

Uhh. Not sure if I'd trust the result more than man-made C code...

[Kleinstein]

The idea is getting away from C with it's often unsafe use of pointers and memory handling. A translation of code that uses such features would be tricky.

Translating C code automaticly likely makes the code unreadable for humans, even more than the original C. It would than be essentially impossible to find and fix other errors.
A manual translation to a new language may actually be a good idea. Sometime a new start / rewirte can make the code clearer and even detect old errors.

[tggzzz]

This sounds like one of those concepts which can make simple things easy and difficult things impossible.

Determining the intention of a lump of code is non-trivial.

Spotting examples of the standard C/C++ problems that cause security and other failures is non-trivial.

If those things were easy, they would be in compilers (+ lint, naturally  ).

Waving magic pixie dust (i.e. AI) around won't solve anything interesting.

[DiTBho]

Good use of Tractor: use it to learn Rust
Write C code and learn how it should be in Rust

HappyDays

[nctnico]

Wouldn't it be more usefull to analyse the code for potential buffer overflows and mitigate those?

[iMo]

It seems the gov sector is the easiest to BS audience for the "bleeding edge technology fund risers"..

[tggzzz]

Good use of Tractor: use it to learn Rust
Write C code and learn how it should be in Rust

HappyDays

As the old joke goes, "If I wanted to go there, I wouldn't start from here".

Happy Daze

[Siwastaja]

Wouldn't it be more usefull to analyse the code for potential buffer overflows and mitigate those?

Two problems:
1) this is an already solved problem, with all those actually important codebases already analysed,
2) this doesn't sound investor-sexy in any way, look how boring you made it sound like.

Therefore, for taxpayer and/or investor money, it needs to be graphene hydrogen cloud egge computing AI. And of course, the trend language of the year.

[nctnico]

OK, sorry. I'll go back to my day job at the sales prevention department. 

[tggzzz]

OK, sorry. I'll go back to my day job at the sales prevention department. 

Mine was at the "going down blind alley" prevention department - semi-seriously, of course.

[coppice]

Old C code naturally rusts. No action is needed. Try to build many old packages, and you are in for some serious welding of patches over the rust.

[mfro]

The only real advantage of AI I've seen so far is that it appears BS blockchain talking decreased a lot.

[artag]

How could this possibly be useful ?

If it's good, error-less C code, it won't make any difference. It's not that compilers compile unsafely.

If it's unsafe C code and the translator creates safe Rust, it has to do something different. So it's not functionally equivalent. The C code might DEPEND on being unsafe to work. How do you know the translation is correct ?

At best you end up with something that can be tested (probably requiring fixes) and maintained as Rust. And might be unreadable.

Maybe DARPA's next idea will be arming soldiers with dummy rifles so there won't be any friendly fire incidents.

[radiolistener]

"You can go to any of the LLM websites, start chatting with one of the AI chatbots, and all you need to say is 'here's some C code, please translate it to safe idiomatic Rust code,' cut, paste, and something comes out, and it's often very good, but not always,"

Uhh. Not sure if I'd trust the result more than man-made C code...

I have the same thoughts 

I wouldn't trust to such AI translated code at all, just because I'm often using AI to generate functions. The result, even if it works is usually unusable due to a lot of mistakes which is almost impossible to detect with testing. So it requires to analyze and rework code generated by AI manually. In some cases it may be minor changes, but sometimes AI just unable to generate good code for some tasks and it requires to rewrite it manually from scratch.

In my opinion just 1-10% of AI generated code is usable the rest needs to be refactored or rewritten from scratch.

[peter-h]

"AI" doesn't actually exist.

Look up how ChatGPT really works. But 99.9% of people think it is "AI".

There has been minimal progress in generic AI since the 1970s. Lots of vertical applications, sure. Especially in applications involving pattern recognition.

[coppice]

"AI" doesn't actually exist.

Look up how ChatGPT really works. But 99.9% of people think it is "AI".

There has been minimal progress in generic AI since the 1970s. Lots of vertical applications, sure. Especially in applications involving pattern recognition.

Well, what they have are elaborate pattern matchers, like a comms equalizer on steroids, and using a similar underlying principle - the correlation quality of a convolution. That is a key part of natural intelligence. Its just not the whole thing. I wonder why the physics Nobel prize didn't go to Norbert Weiner.

[SiliconWizard]

Interestingly, I think the concept that is shown in the following site (that I linked to in another thread) also applies here with LLMs : https://tylervigen.com/spurious-correlations

What LLMs do (and CNNs in general) is pretty much that: a bunch of correlations on large sets of data, with the goal that it would yield something "intelligent". Makes you ponder.

[coppice]

Interestingly, I think the concept that is shown in the following site (that I linked to in another thread) also applies here with LLMs : https://tylervigen.com/spurious-correlations

What LLMs do (and CNNs in general) is pretty much that: a bunch of correlations on large sets of data, with the goal that it would yield something "intelligent". Makes you ponder.

We all realise that people need constant reminders that correlation does not mean causation, because the human mind is wired as a pattern matcher, and keeps hooking onto many false correlations. The snag in 2024 is that line has been repeated so many times, dumb people are now claiming that when things correlate that means there is NO causation. The difference between IQ and wisdom/common sense/whatever you want to call it appears to be mostly that IQ is largely pattern matching, and wisdom is the ability to filter out the dumb matches.

[tggzzz]

Interestingly, I think the concept that is shown in the following site (that I linked to in another thread) also applies here with LLMs : https://tylervigen.com/spurious-correlations

What LLMs do (and CNNs in general) is pretty much that: a bunch of correlations on large sets of data, with the goal that it would yield something "intelligent". Makes you ponder.

We all realise that people need constant reminders that correlation does not mean causation, because the human mind is wired as a pattern matcher, and keeps hooking onto many false correlations. The snag in 2024 is that line has been repeated so many times, dumb people are now claiming that when things correlate that means there is NO causation. The difference between IQ and wisdom/common sense/whatever you want to call it appears to be mostly that IQ is largely pattern matching, and wisdom is the ability to filter out the dumb matches.

I don't think anybody claimed Igor Aleksander's 1983 WISARD had IQ

It was designed, like modern ML systems, to do pattern matching based on training sets. It worked well in the lab when trained on pictures of tanks and cars, but failed miserably on Luneberger Heath. Nobody could tell why. Eventually they figured out the training set was car adverts and tanks in N Europe, and it was discriminating between sunny days and cloudy days. It is rumoured that Aleksander's colleagues wouldn't recognise his existence on sunny days.

Modern ML systems are more sophisticated. They recommend for/against medical treatment based on the font labelling the scan.

[coppice]

I don't think anybody claimed Igor Aleksander's 1983 WISARD had IQ

It was designed, like modern ML systems, to do pattern matching based on training sets. It worked well in the lab when trained on pictures of tanks and cars, but failed miserably on Luneberger Heath. Nobody could tell why. Eventually they figured out the training set was car adverts and tanks in N Europe, and it was discriminating between sunny days and cloudy days. It is rumoured that Aleksander's colleagues wouldn't recognise his existence on sunny days.

Modern ML systems are more sophisticated. They recommend for/against medical treatment based on the font labelling the scan.

You keep quoting that ham fisted botched up case like it represents something significant. It just represents what incompetence can achieve on a good day.

[tggzzz]

I don't think anybody claimed Igor Aleksander's 1983 WISARD had IQ

It was designed, like modern ML systems, to do pattern matching based on training sets. It worked well in the lab when trained on pictures of tanks and cars, but failed miserably on Luneberger Heath. Nobody could tell why. Eventually they figured out the training set was car adverts and tanks in N Europe, and it was discriminating between sunny days and cloudy days. It is rumoured that Aleksander's colleagues wouldn't recognise his existence on sunny days.

Modern ML systems are more sophisticated. They recommend for/against medical treatment based on the font labelling the scan.

You keep quoting that ham fisted botched up case like it represents something significant. It just represents what incompetence can achieve on a good day.

It was an early harbinger of things to come.

The same fundamental problems exist in modern ML systems in spades, except now they can screw up many peoples' lives.

Key concepts:

• you can't test quality into a system (test==training example), quality has to be designed
• "There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies, and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult." Modern ML systems can only be the latter

[coppice]

I don't think anybody claimed Igor Aleksander's 1983 WISARD had IQ

It was designed, like modern ML systems, to do pattern matching based on training sets. It worked well in the lab when trained on pictures of tanks and cars, but failed miserably on Luneberger Heath. Nobody could tell why. Eventually they figured out the training set was car adverts and tanks in N Europe, and it was discriminating between sunny days and cloudy days. It is rumoured that Aleksander's colleagues wouldn't recognise his existence on sunny days.

Modern ML systems are more sophisticated. They recommend for/against medical treatment based on the font labelling the scan.

You keep quoting that ham fisted botched up case like it represents something significant. It just represents what incompetence can achieve on a good day.

It was an early harbinger of things to come.

The same fundamental problems exist in modern ML systems in spades, except now they can screw up many peoples' lives.

Key concepts:

• you can't test quality into a system (test==training example), quality has to be designed
• "There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies, and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult." Modern ML systems can only be the latter

Ah, the same old rubbish you say each time. An ML system doesn't have to be perfect, since we mostly use these systems to replace highly flawed humans. We only need reasonable confidence that they are no worse than a human. An ML system is quite like a human with no common sense. Well, large numbers of humans have no common sense either, which is probably why bad many ML system responses seem a lot like the dumber of the people around us. When you look at applications where things need to be perfect, like the AI chips floor planning tools that are becoming widely used, the tool doesn't have to be perfect. Like a human doing the floor planning, you expect a lot of errors, so you use rigorous tools to see that the final result gets properly cleaned up.

[peter-h]

I think the remaining tricky issue with ML is that you mostly don't know how the machine learnt it.

Hence the funny cases of recognising a husky (dog) with very high reliability, but actually only because all husky pics used to train the software contained a snow background.

I don't know that this has ever been solved - other than by selecting the learning material carefully but then you still don't actually know what the machine is looking at.