Please change title, we don’t want clickbait here

What’s in this article is true, but to be honest I’ve never seen anyone using lines of code as an optimization metric. Even among the most AI enthusiastic people. I mean: the author of the article seem to be fighting non-existing problem.

Memory requirements does not mean that something is bloated.

Windows XP required 64 MB of RAM. Does it mean Debian is bloated then?

AFAIK perfomance and low resource usage has never been a main objective of this distro

but these do not work like “trending”, right? They are chronologic?

It’s global trending. Which I find unfortunate, because it basically defeats a purpose of thematic servers. For example I’m on Polish server, full of people taking in Polish, but I see only global trends which are not at all in Polish 🫤

true, I don’t like the curl [something] | sh pattern for installation. Calling it is just like letting random guy from the internet control of your PC to download some binaries. I’m seeing this trend more and more in Github repos

Yeah, that’s the idea, but they are still not there

Who exactly is using OpenRouter? Is this used for coding? Bots? Casual conversations? Because that could tell what exactly those top models are good at.

I wish GitHub Copilot shared such data. To see what models do programmers use for work.

Let this sink in down

You have 128GB of RAM

Who do you think I am, a multimilionaire?

The main advantage of fediverse over, let’s say X, is that you can change server if you find owner not trustworthy. So just do it, it’s exactly why it was designed in this way, to let you do it easily.

But talking about funding… I might indeed reconsider doing this…

Hopium administered

Cloud. Windows is going to be sold as remotely accessible virtual machine hosted on Azure. The change will first take place in government offices, then in companies, and finally (after people get used to it at work) among consumers.

Why would gov and enterprise like it? Because of:

• safety - all enterprise data will be stored on Azure servers and won’t ever leave it. It makes preventing data leakage so much easier
• maintenance - software updates can be applied even outside of working hours, Microsoft could launch VMs and update at any time
• ease of upgrade - need better specs? you don’t need to buy better hardware anymore, you just buy better subscription. Hardware won’t become obsolete anymore that quickly

Consumers will also like it. No need to pay hundreds of dollars for new GPU when you just want to play newly released game. Also, all your data accessible from anywhere in the world.

And why Microsoft would like it? Kinda obvious, it would be even harder for users to quit a subscription, they will be tied to ecosystem even more

Yes, sounds ridiculous, but how will this ratio change if we take into account the cost of hiring a programmer and the costs of implementation of a niche feature that this experiment provides at a cost of LLM inference?

Also: we can cache and reuse enpoint implementation.

I believe the same thing was said about the Internet in the ’90s: “It speeds up communication, but how would anyone earn money from it?”

Although I don’t think we’re anywhere close to AGI or anything like that, current AI development fundamentally changes a few things in our lives: how we find and process information (information retrieval works very well), how we interact with computers (using natural language instead of clicking through interfaces), and how productive we are.

Video generation models are going to bring entertainment to a whole new level. A single person can now create an entire movie without even buying a camera. Entire game development studios can build worlds larger than ever before. Text generation makes disinformation and propaganda insanely cheap and effective. Surveillance will be much easier now, as owning a communication platform not only allows you to search for messages by phrase but also by meaning. Ads will be far more personalized, as AI chat platforms now know us much better than Google — the current leader in this field.

So:

there isn’t anything real there?

I really don’t think so.

So how dangerous is that really? I assume one day we’ll finally see investors saying, “Nah, that’s a bubble. I’m not gonna see any returns from those companies - I’m selling.” Then stock prices will fall, and some investors will lose money by selling for less than they bought. After that, AI unicorns will start to lose funding and close their businesses, laying off people.

But will I - a person who does not work in the AI industry and has not invested in AI companies - be affected by this?

I wish there was something like this for tags. So: display only posts that contain tag X, but sort them with some algorithm. Ideally: steered by number of likes and date of adding. May be even the same that is used by Lemmy.

Usefulness really comes down to which model is being used. I’ve noticed most developers choose GPT for Copilot because that’s what they are familiar with (or they often don’t have a choice due to company policy). I recommend to try Claude Sonnet. How it works is true magic.

But I agree, repetitive tasks is what it should be used for. Planning is still programmer’s job

Please add some context for someone not up to date with all that jargon. There is a trend here and on hacker news to just post a model name that says basically nothing and I often just don’t even know why I should care. Or maybe I shouldn’t?