The training data here is exaggerated more, actually. This task should take kilobytes, max, and would finish in a fraction of a second. Also, no self-respecting ML engineer would put together an ML system without accounting for every data type.
But a floating point issue is the exact type of issue a LLM would make (it does not understand what a floating point number is and why you should treat them differently). To be fair, a junior developer would make the same type of mistake.
A junior developer is, hopefully, being mentored by more senior coworkers who are extra careful with code reviews and would spot the bug for the dev. Machine generated code needs an even higher level of scrutiny.
It is relatively easy to teach a junior developer to write code that is easy to read and conforms to the teams style guide.
With all the recent hype around AI, I feel that a lot of people don’t understand how it works and how it is useful. AI is useful at solving certain types of problems that are really difficult using traditional programming, like finding patterns that aren’t obvious to us.
For example, object recognition is about finding patterns in images. Our brains are great at this, but writing a computer program capable of taking pixels and figuring out if the pattern is there is very hard.
Even if AI is sometimes going to misclassify objects, it can still be useful. For example, in a factory you can use AI to find defects in the production line. Even if you don’t get it perfect, going from 100 defects per 1M products to 10 per million is a huge difference and saves the factory a lot of money.
Agree, but the joke to me is business folks thinking AI is a miracle and they can just shove it everywhere to print money. Where us devs know what you mean, and would like to add it in where it makes sense. Business thinks it’s ready to replace us.
The key to "AI" is having a human there to take algorithms and apply them to the right problems.
This is what most people don't understand because many of the demos are quite impressive and narrowly tailored to prevent the fact from being obvious unless you know what you're looking for.
Most useful application so far seems to have been to predict protein folding. Have to check up on that, it should allow to cure all sorts of bad things.
It’s saying that instead of spending all the resources needed to gather all of the training data for the LLM, just give a junior dev some coffee as the input instead.
The direct comparison is input and output. Coffee/training data is the input and the code is the output.
Yeah I don’t really get this, you can come to deterministic mathematic conclusions with ML, it just requires different structuring of the problem. While area of a rectangle may not need optimization, there are many such places that do, like file compression, which requires perfectly accurate results.
it’s for programmers, which you’re obviously not if you think this makes no sense…
and guess what, troll, that’s why you have a public comment history…
it’s a primary function… which i used, as intended, to see how much of a troll you are…
I don’t think that not educating people is an option. Even in the highly unlikely case that every job is hypothetically taken over by “AI”: humans like to learnand hone their skills.
Why? Deploying or not deploying so-called AI is a choice we, as a society can make. The only reason why LLMs are pushed so hard by the industry is because some corporate gouls think that they can save money on labour with it.
Sure, but you can still find plenty of info on it by searching for .NET Framework or .NET 4.6. All the documentation is still available. Its just not in the spotlight any more.
Not an intern, but this week I’ve unraveled some mysteries in ASP.NET MVC 5 (framework 4.8). Poked around the internals for a while, figured out how they work, and built some anti-spaghetti helpers to unravel a nested heap of intermingled C#, JavaScript, and handlebars that made my IDE puke. I emulated the Framework’s design to add a Handlebars templating system that meshes with the MVC model binding, e.g.
and some more shit to implement variable-length collection editors. I just wish I could show all this to someone in 2008 who might actually find it useful.
programmer_humor
Newest
This magazine is from a federated server and may be incomplete. Browse more on the original instance.