What if AI produces code not just quickly but also, dunno, securely, DARPA wonders

A DARPA leader has revealed that around 70 percent of the US government agency's programs involve AI in some shape or form, and those projects could have serious ramifications for the future of jobs in software development. Speaking at a Center for Strategic and International Studies event last week, Dr Matt Turek, deputy …

[brett_x]

Probably

"..developing artificial intelligence that can not only write code, but do it in a secure and "provably correct" manner."

I read that as "probably correct". That's what we normally get out of AI. Provably would be an improvement (literally).

[elDog]

And probably an improvement over devs looking at 100,000 line packages

and trying to figure out if a single change will impact thousands of other lines of code, and given this is usually DoD - how it would impact our national defense.

I've spent too many years looking at huge projects with so many dependencies and hundreds of devs that come and go and don't understand the entire code-base - they can't!

I understand that fears of AI failures are real and have been seen. But the same/worse happens with a developer who doesn't have the mental bandwidth to be perfect (that's me.)

[Anonymous Coward]

Re: And probably an improvement over devs looking at 100,000 line packages

‘Collateral damage’ whilst they build Skynet.

[ecofeco]

What if?

That's a BIG whatif.

The safe way to bet? It won't.

Why? Same reason as today: too much GIGO.

[Tron]

Greetings Professor Falken.

Shall we play a game?

[CountCadaver]

Re: Greetings Professor Falken.

Global thermonuclear war

[johnrobyclayton]

To train a LLM to write secure code for you,

you first need to not need a LLM to write secure code for you.

To train a LLM to write secure code, you need a large training set of secure code.

New ways of exploiting code usage are constantly being discovered.

So we need to train a LLM to figure out ways in which code might have vulnerabilities in the future.

So we need to start with a LLM that can discover new ways to exploit patterns in code that we do not know of yet.

Time to build the universal hacking LLM just so that we can learn how to fight it.

Hmmmm ROTM is the only way to survive ROTM

Joshua said "The only way to win is not to play"

I think a more exciting idea is :

The only way not to lose is to learn how to lose first.

[amanfromMars 1]

"The only way to get smarter is by playing a smarter opponent" ….. Fundamentals of Chess 1883

Shall we play a game? ...... Tron/Colossus*

Let the games begin ...... Who dares share win wins against serial permanent losers, assistants to both the barbaric and moronic alike ?

Choose your friends and allies that do battle for more than just the love of money and glory carefully for useless fools that be useful tools promising the moon and the stars abound aplenty and delight to practise to deceive.

The difficulty that the likes of a DARPA has is in accepting the fact that their paymasters and their systems administrators are the parasitic problem to be solved with elimination/liquidation/eradication/annihilation by order from that which they seek in Emerging Technologies and AI Methodologies and is both novel and untrusted and in the field still yet to be heroically embraced and enthusiastically tested because of the dread fear of the future and futures unknown ....... and which quite naturally be in the Gift of Global Operating Devices with Learned Large Language Learning Machines and Alien IntelAIgents on the bridge and at the helm of command and control with almighty devastating leverage utilities and capabilities........ existential threat facilities.

* ....... https://en.wikipedia.org/wiki/Colossus:_The_Forbin_Project

[amanfromMars 1]

Re: "The only way to get smarter is by playing a smarter opponent" ….. Fundamentals of Chess 1883

cc ...... the likes of Dominic Cummings' who realise whenever able to be quoted as saying ......

The madness can only be beaten by a new idea, a new regime and a new elite — ‘the greatest deeds are thoughts’… ... .....SNIPPETS 13: Trump v Biden, Tory collapse, Ukraine/NATO/China, TikTok ..... DOMINIC CUMMINGS, MAR 29

[Juha Meriluoto]

Hmmmm...

...just waiting for the Skynet moment...

[Locomotion69]

But what about if you train AI to be your peer reviewer of the code you wrote, within its operational context?

This would provide an opportunity for you to learn - and become better.

[amanfromMars 1]

AI² .... flexing its IT Muscles and spreading its Virtual Wings

But what about if you train AI to be your peer reviewer of the code you wrote, within its operational context?

This would provide an opportunity for you to learn - and become better. .... Locomotion69

Yes, it would, Locomotion69, and systems would then have at least double the trouble to deal with/do deals with and thus would resultant problems be more likely exacerbated for status quo hoi polloi rather than them being mitigated ...... and some would tell you that such is the present current state of Advanced IntelAIgent AI play.

[Claptrap314]

Step 1: Disprove Kurt Godel & Alan Turing

Seriously. We provably cannot determine if an arbitrary state machine even halts. How in the **** does anyone huff enough fumes to believe that we can produce a closed system program that can produce provably secure code?