"Well, well, well if it isn't the consequences of my own actions. "

• Linus Probably

stop trying to force a reality that fits your narrative

AI is the car of modern times. It's convenient, ruinous for the environment, and a danger to anyone in proximity of one. Hell, we even wasted tons of land on highways kinda like what we're doing with datacenters. But boy, does it get you from point A to point B quick.

Then just let AI use AI. I'll collect my Nobel prize now thx.

"Knives cause lots of stabbings, so let's ban knives"

Some humans cannot own knives responsibility. It doesn't mean I have to cut the pineapple with a spoon.

There's an example used in regulatory circles: electricity would never be permitted in homes under today's regulatory standards - we only accept it and its hazards because it is so incredibly common and useful in daily life and we've been accepting it since before we cared so deeply about safety.

So, have today's goalposts moved so much that "AI" no longer passes modern society's definitions of acceptable? I feel like such a technology would have been cheerfully embraced in the 1960s.

When reality is shit one must rage at it.

Oh no, you are one of those "if you don't adapt you will be left behind" people, aren't you?

Why is it bad if people get angry at things that are happening in the world? Would you rather we get angry at fiction?

Yea I'm sure when Google came out "Real Programmers" were angry at all these people just looking up solutions to their coding issues, and using Stack Overflow instead of studying the API or books.

Only to be replaced by people that embraced the new technology.

Warning! very unpopular opinion ::: spoiler Title I hate STEAM, preferred when it was just STEM. :::

I've said it before, and I'll say it again : eventually you just fucking end up with "be a complete person; here's the entire dewey decimal system."

STEM is meant to deal with logic based thinking and the problems unique to that industry, like for instance, a huge gender gap. Shoving arts and humanities in doesn't fix the issue. Cross disciple collaboration is great, expanding your horizons and putting ethics into practices is great, but that's not what STEM is.

There are already plans for these things in practice (hippocratic oath, iron^[https://en.wikipedia.org/... ring^[https://en.wikipedia.org/wiki/Iron_Ring] ), which need to become more widespread. We don't have to have STEAM toys and SHMEAT courses.

Have it do the tedious work

Have it create a better testing environment

Have it do bunch of edge case things or automate ui testing.

Use it as a tool. You can still be the brains.

attack it with ethics!

How do? Does this apply even if it's a locally run LLM? Even if it's done using green energy like Mistral? Even if all of the AI written code is carefully reviewed by professionals?

Is it a problem of making it too easy for people? If so, what is the correct amount of easiness for people? Shall we start taking clothing down to the river to scrub instead of using washing machines too?

You are claiming it's irresponsible, why do you say so?

AI is just a technology, how you use it makes it bad or not.

Nuclear is the cleanest safest energy we have, but you can also make bombs with it.

This is exactly how I use it too. Google and other search engines have gone to complete shit in order to cater to LLMs. So that's how I use the LLM. You are my search bitch, Gemini. No, you're not allowed to actually create anything.

That's the neat part, you can.

If I had to pick one, it'd be Olmo 3 from Ai2: Non-profit. Ai2 is a research institute, not a company that needs to sell you a subscription or harvest your prompts. Fully open, not just "open weights" Ai2 publishes the training data, the training code, every checkpoint and the logs. Anyone can check exactly what went in, and in principle rebuild it from scratch. You don't have to take their word for anything.

Runs locally. The 7B version runs on an ordinary gaming laptop or a recent MacBook. Your code never leaves your machine, there's no data centre involved when you use it, and the energy cost is your laptop being on. Apache 2.0 licence. No usage restrictions and no strings attached.

Good enough for the tutor use case. Explaining errors, walking through why code misbehaves, and pointing out likely bugs are all well within what it can do.

To be fair about it: the training data is largely scraped from the web, like nearly every model. The difference is that Olmo is honest and auditable about it.

If strict consent matters most to you, StarCoder2 is cleaner (only permissively licensed code, with an opt-out)

I used a local model today to write a long form regex for me today. it would have taken me 30 minutes. instead it took 3 seconds and worked out of the box.

I think this is a perfect example of what LLMs should be used for.

Most tokens sold come from China today and open weight models, and most (EDIT) 42% of energy produced in China is from clean sources like nuclear or renewables. Don't blame AI for ethics or lack thereof in neoliberal capitalism and a lack of planning or regulations. China also has the most and smartest regulations on AI so far.

Nah im still firmly anti AI regardless of what daddy says. Still based, but hypnotized by the tokens.

It's a thing, good for some things, not so great for others.

It's a new thing, people are just starting to figure out what it's really good for. It's also improving rather dramatically over the past year, hard to say if that's the start of an exponential curve, or if we're hitting an s-curve rolloff already...

If it continues to grow its energy / environmental impact at the current growth rate, that's a bad thing, but it's not "in a bad place" yet... if you're outraged about the last 2 years of AI environmental impact, you should have already been apopleptic over the past 15 years of Bitcoin's environmental impacts.

rationing:

Limit lines of code or diff size. Limit scope. Limit PR per user per month. No bots, obviously. New users don't get to send anything, they can comment and vote, at best. And bring in negative ratings for contributors: low quality, suspicious code, bad PRs.

This is another way of saying deceleration, a word that accelerationists hate.

Because the pace of code production has officially overwhelmed the human capacity to review it. As it was discussed recently. Linus answer was to use more AI, with the implementation of code review tools that gobble up code and spew out summaries.

For the time being, humans reading the summaries count as human reviewed. But at some point in time this system will crack. The current zeitgeist is to restrict who can contribute. Which leads to the obvious follow up question of who decides based on what, and how to decide who deserves to contribute.

You just witnessed the first steps into the philosophical and political erosion of the open source movement. Not just on the kernel but it is also affecting almost all popular FOSS projects.

But of course, that doesnt matter because Linux is not and has never been political. /s

Stop complaining about vibecode, just review it

Hahaha. Not a developer, huh?

There is no way I could possibly review my coworkers' 1000s of lines of code he posts daily. I just blindly approve it because fuggggit we don't care about quality only velocity.

This is different from before because before he would normally post 2 or 3 small code changes per week. Also, he would have tested it himself so starting off it would already be in a good state. Also, since he tested himself, we could talk about the change since he still had the context.

Now it's just ship, ship, ship. We literally can't "review it" like we did before. That world is gone.

so what happens when i start barfing up gold coins and never can't afford food again, what then?

i mean they've already shown that you can just straight up have an image model generate the game in real-time without even writing code

the future is the reverse matrix: all-but-sapient computers forced to dream our games
I miss when computers were fancy rocks we tricked into doing maths..

There are things that we don't know about yet that are too complicated to code by hand, and whey the models are good we will discover their limits then and start working on those

Rather than writing a program, it just codes and runs the code at the same time

Yeah, so there's been a lot of improvement very quickly over the last year in LLM agent's coding abilities.

If they continue to improve at an exponential improvement rate like they have been... maybe 5-10 years from now you might see such a "code and run" as you make the descriptions LLM performance, for simpler things.

I'm curious, with "unlimited access" to the inference engine hardware capabilities, just how much faster the LLM agents can run when they're not in timeshare over the internet mode?

Often enough to give his statements serious consideration.