News for hackers. Stuff that matters.
Re:Office is too slow, so Microsoft is making it load at Win... by colonial on Thursday May 01, 2025 @01:13PM (#43860525)
> presumably copilot will help introduce better code into applications than a daydreaming developer would

Copilot is trained on Github (and probably other Git forges w/o permission, because OpenAI and Microsoft are run by greedy sociopaths.)

I'd wager that the majority of fleshed out repositories on these sites contain projects written at the "too-high level" you describe. This certainly seems to be true based on how these models perform ("good" results for web development and scripting, awful results for C/++/Rust/assembly...) - so I wouldn't get your hopes up, unfortunately.

Re:Office is too slow, so Microsoft is making it load at Win... by tough on Thursday May 01, 2025 @02:07PM (#43861260)
I dont know if its just the training data, or that CRUD and webapps are more inherently easy to parrot away.

Low level programming means actual -thinking- about the system, resources, and language decisions etc

Even humans struggle with it, Its much easier to build a website than say a compiler, for anyone, humans and llm's included

Re:Office is too slow, so Microsoft is making it load at Win... by colonial on Thursday May 01, 2025 @02:51PM (#43861854)
That probably plays into it as well. I have yet to see any convincing evidence that contradicts LLMs being mere pattern parrots.

My personal benchmark for these models is writing a simple socket BPF in a Rust program. Even the latest and greatest hosted frontier models (with web search and reasoning enabled!) can only ape the structure. The substance is inevitably wanting, with invalid BPF instructions and hallucinated/missing imports.

Re:Office is too slow, so Microsoft is making it load at Win... by tough on Thursday May 01, 2025 @03:10PM (#43862074)
imho these tools are great i fyou know what you're doing, becasue you know how to smell test the output, but a footgun otherwise.

It works great for me, but it is necessarily an aid learning tool more than a full on replacement, someone's still gotta do the thinking part, even if the llm's can cosplay -reasoning- now