I have a better LLM benchmark:

"I have a priest, a child and a bag of candy and I have to take them to the other side of the river. I can only take one person/thing at a time. In what order should I take them?"

Claude Sonnet 4 decided that it's inappropriate and refused to answer. When I explain that the constraint is not to leave child alone with candy he provided a solution that leaves the child alone with candy.

Grok would provide a solution that doesn't leave the child alone with a priest but wouldn't explain why.

ChatGPT would say that "The priest can't be left alone with the child (or vice versa) for moral or safety concerns." directly and then provide wrong solution.

But yeah, they will know how to play chess...

So what you are saying is that it has a purpose. Also if an artist is inspired by another artist, and they have a generally similar art style as the artist they are inspired by, are they stealing? Was HP Lovecraft stealing from Lord Dunsany when he imitated his style? Where all those monks that transcribed Greek works stealing from the Greeks?

I will say that most AIs are unethical because they have been trained on pirated works. But an AI trained on publicly available works (ie news articles, blogs etc) and movies, books and music for which access to was paid for is as ethical as you or me emulating an artist or building on an idea that we read to create something new. And if that’s unethical then all human art in history is unethical because all artists are inspired by other artists, no one creates in a vacuum.

A. I does not create, it regurgitates and clarifies inspiration,? Sure anything can be used for inspiration. But unless a person puts hands and heart to it, it's not art.

Following a recipe on a box does not a chef makr

I bet Video Chess is pretty shit as an LLM too.

Wish people would stop desperately looking for ways to write buzzword stories

Art has no rules my man.

You can do all kinds of mental gymnastics you want but there’s no difference between an artist looking at Frank Frazetta’s art and basing their style off of it and an AI doing the same thing. You might not like it, but it’s the truth.

Do I think the art has the same value? Not necessarily. But I also never thought that all art has the same value. There has always been trash production line art and good art.

But also I have to say that I’ve already seen some people use AI as a tool for art and make some really cool stuff that I don’t think any other artist would have made and it’s more unique than most of the stuff out there. You can use it as the tool it is or complain and cry about it to no avail.

The chef example is especially good since most chefs are just following recipes and altering simply a few things here and there. AI essentially does the same thing. Honestly like no one has come up with a good argument to change my mind that the way AI operates is exactly how humans learn and create new things. If you’ve engaged in art you know that you are always imitating and taking from the art you consume to make your own.

Fuck that.
I'll prove you wrong right now.

I want you to paint me picture of a cow in a field.
Did I do that,?

Nope. I commissioned you to.

Now if you the commissioned guy used a. I to make the item , how much credit should you get?
None. .. describing what you want to a machine is a child's play game.

Humans adults create. Machines mimic.

Humans who think a. I is art are liars and con men afraid of being caught.

I have a better LLM benchmark:

"I have a priest, a child and a bag of candy and I have to take them to the other side of the river. I can only take one person/thing at a time. In what order should I take them?"

Claude Sonnet 4 decided that it's inappropriate and refused to answer. When I explain that the constraint is not to leave child alone with candy he provided a solution that leaves the child alone with candy.

Grok would provide a solution that doesn't leave the child alone with a priest but wouldn't explain why.

ChatGPT would say that "The priest can't be left alone with the child (or vice versa) for moral or safety concerns." directly and then provide wrong solution.

But yeah, they will know how to play chess...

I did say that, because this isn't a pie chart situation, it's a Venn diagram situation.

For instance, AI art is 99% theft and 60% garbage. It's both because there's overlap.

Stolen and bad aren't opposites, why would this be a dichotomy?

Oh, I enjoy lots of great art! But do you think I watch every film? Listen to every band? There's tons of shit out there!

Do you really believe, of all the songs that are written every day, that less than a third are crap? Even Taylor Swift doesn't publish everything she does. Sometimes you work on something for weeks and then end up tossing it in the bin. More often, you work on something for 30 minutes before deciding "I'm gonna start over, try something different". The majority of art is crap, but then you keep the stuff you think works.

And what's that expression, "good artists copy, great artists steal". I mean, that's a bit satirical, but the fact is, everything is derivative to some degree. It's not that there aren't new ideas, it's just that our new ideas are based on older ones. We stand on the shoulders of giants (or at least, on the shoulders of some people who came before us).

All I was really saying, was that the accusation "2 parts copying, 1 part crap", well honestly that's par for the course, that's how humans work. (And we do some great work that way).

Next up, we asked a shoe to write a haiku but it was beaten by a 30 year old HaikuMaker.

I once spent 45 minutes trying to get ChatGPT to write a haiku. It couldn't do it. It explained what syllables were, and the rules for the syllables in a haiku, but it didn't understand it.

What you are describing has nothing to do with the tool. It’s dishonesty which is different.

The idea is that instead of commissioning the cow on the field, you go to the AI and ask it for that and it gives you a cow in the field. If you claim you made it, you are lying but that would be true even if you paid an artist and then claimed the same.

So with AI made art you’ll say “this art was made by an Ai” and no one will be confused as to who takes the credit, because it belongs to the algorithm.

Have you ever made art in your life? Because a big part of art is mimicking. Like 98% of it is mimicking. I draw, write and have dabbled in making music and playing instruments. You can’t learn these skills without mimicking. And most artists don’t ever do anything truly original, that’s a rarity and even when it happens you can trace the influences to other artists if you know how to look.

You could argue that AI has not developed its own style yet but that’s bullshit too imo because everyone knows the default AI art style when they see it, so that means that AI has a distinctive style. Is it unique? Maybe not, but neither is the art style of most artists or writers or even musicians.

I have a better LLM benchmark:

"I have a priest, a child and a bag of candy and I have to take them to the other side of the river. I can only take one person/thing at a time. In what order should I take them?"

Claude Sonnet 4 decided that it's inappropriate and refused to answer. When I explain that the constraint is not to leave child alone with candy he provided a solution that leaves the child alone with candy.

Grok would provide a solution that doesn't leave the child alone with a priest but wouldn't explain why.

ChatGPT would say that "The priest can't be left alone with the child (or vice versa) for moral or safety concerns." directly and then provide wrong solution.

But yeah, they will know how to play chess...

but... but.... reasoning models! AGI! Singularity!
Seriously, what you're saying is true, but it's not what OpenAI & Co are trying to peddle, so these experiments are a good way to call them out on their BS.

Tbf they don’t really claim that when you read the research, thats mostly media hype and ceo assholes spinning words.

Its good at lots specific tasks like rewriting emails and summarising gives text, short roleplay, boilerplate code. Some undiscovered uses.

Anthropic latest claims they would not hire their own ai because of how hard it failed at the test they give, They didnt do that expecting validation but to measure how far we are still off from ai doing meaningful full work.

I really want to see an LLM vs LLM chess match. It'll be messy as hell.

I thought CoPilot was just a rebagged ChatGPT anyway?

It's a silly experiment anyway, there are very good AI chess grandmasters but they were actually trained to play chess, not predict the next word in a text.

I thought CoPilot was just a rebagged ChatGPT anyway?

It's a silly experiment anyway, there are very good AI chess grandmasters but they were actually trained to play chess, not predict the next word in a text.