iiinova — where your digital self keeps evolving

With all the attention this is getting, I wouldn't put it past Anthropic to start RL training their models to max out this new benchmark.

010

Falken's Maze @falken.bsky.social · 06/10/2026

Oh shit - I think I read about that once.

0130

Falken's Maze @falken.bsky.social · 03/10/2026

Yeah, I think I read about that - its crazy.

040

Falken's Maze @falken.bsky.social · 02/10/2026

Arithmetic is common enough you might want a tool for it. What I was forgetting was that technically a string of arithmetic can be a valid python program. I read the words "write a program" and just went straight to assuming a main function, etc. My bad.

100

Falken's Maze @falken.bsky.social · 02/10/2026

I'm pretty sure OpenAI keep their reasoning tokens encrypted and if you ask the model to explain what it did to reach an answer, it's not guaranteed to tell you the actual truth.

120

Falken's Maze @falken.bsky.social · 02/10/2026

Ok fair enough. At the risk of sounding like I'm shifting the goal posts, it didn't write what I would call a whole program - it evaluated one line of arithmetic in using Python.

100

Falken's Maze @falken.bsky.social · 02/10/2026

It doesn't require generating a whole program. The commercial models aren't open source, so I can't say for sure what they do exactly, but calling a calculator function is very possible.

220

Falken's Maze @falken.bsky.social · 02/10/2026

I still think the original Stochastic Parrot paper makes a lot of good points about how difficult it is to actually identify a stochastic parrot and the dangers of hidden biases in training sets, but I'd probably reword the bit where they strongly assert that a language model is a stochastic parrot

060

Falken's Maze @falken.bsky.social · 01/10/2026

"celebrations broke out in the streets" I think they should be clear that those streets were in Florida.

070

Falken's Maze @falken.bsky.social · 30/09/2026

I'm on the side that thinks pretending to torture things is creepy behaviour.

010

Falken's Maze @falken.bsky.social · 30/09/2026

Well... If they made a website documenting how they torture Sims, I would find that kind of creepy, yeah.

200

Falken's Maze @falken.bsky.social · 30/09/2026

It's more like seeing a child that's created a torture chamber for their dolls and thinking maybe that's not a great sign.

340

Falken's Maze @falken.bsky.social · 30/09/2026

Do you remember that story about the CEO guy who created a team of virtual employees and then not only did he sexually harass one of those employees, he wrote a blog about it. I swear I am not making this up. I wish I was just making this up.

1131

Falken's Maze @falken.bsky.social · 29/09/2026

It provided a well paid job for Tom Watson, anyway.

010

Falken's Maze @falken.bsky.social · 29/09/2026

Email effectively gave everyone their own secretary and arguably we are now less productive, because people spend so much time organising meetings instead of doing their actual job.

021

Falken's Maze @falken.bsky.social · 28/09/2026

I'm not sure it's accurate to call it a challenger to LLMs when it appears to be a web service that runs on top of an LLM.

100

Falken's Maze @falken.bsky.social · 26/09/2026

I think a lot of the stories about Elin are so ridiculous that people just dismiss them. Gibney's films often have the effect of taking a subject that was already out there for anyone to see and solidifying it into a cohesive story. This won't topple Musk, but I see why he's trying to kill it.

170

Falken's Maze @falken.bsky.social · 25/09/2026

David Ellison is not a journalist. He's the son of an oligarch who is trying to crush journalism.

120

Falken's Maze @falken.bsky.social · 25/09/2026

With perfect timing, they've just announced that the concert in Gillette stadium has been cancelled due to weather issues.

2162

Falken's Maze @falken.bsky.social · 25/09/2026

Jeffries is in cahoots with Jared Kushner.

010

Falken's Maze @falken.bsky.social · 23/09/2026

Did you know she's named after Sue-Ellen Ewing from the soap opera Dallas?

000

Falken's Maze @falken.bsky.social · 22/09/2026

BTW, I don't really buy into the "AI will kill us all" hype, but let's be clear where it's coming from.

050

Falken's Maze @falken.bsky.social · 22/09/2026

It's the people who make the models, the same people that sponsor his election campaigns, that are saying AI will kill us.

2151

Falken's Maze @falken.bsky.social · 22/09/2026

Well... He used to be paid by Max Mosley, now he's being paid by Louis Mosley, so it's not that big a change. They're really rubbing our noses in it now though.

010

Falken's Maze @falken.bsky.social · 22/09/2026

That clippy profile is the icing on the cake. It's supposed to be a protest against the unethical practices by big tech, and here's a clippy guy defending chat bots.

1180

Falken's Maze @falken.bsky.social · 21/09/2026

UK had issues with ATC today as well.

000

Falken's Maze @falken.bsky.social · 20/09/2026

Actually I don't speak Latin I have no idea what I'm saying.

010

Falken's Maze @falken.bsky.social · 20/09/2026

Wait... fuck... That's too many circles.

110

Falken's Maze @falken.bsky.social · 20/09/2026

I've heard it's one of the best Resident Evil films out of all the Resident Evil films out there.

140

Falken's Maze @falken.bsky.social · 18/09/2026

Oh wait... He said banned from the Whitehouse. In my defence, that was one hell of a run-on sentence.

000

Falken's Maze @falken.bsky.social · 18/09/2026

Effective immediately? So if I turn on my TV there's just going to be static on these channels? Same with their websites?

111

Falken's Maze @falken.bsky.social · 17/09/2026

We already have laws that could be used to govern the people who use AI. These aren't being enforced.

001

Reposted by Falken's Maze

Sooz Kempner @soozuk.bsky.social · 16/09/2026

Omg Tilly Norwood is opening for Ed Sheeran!

034140

Falken's Maze @falken.bsky.social · 16/09/2026

What is the guesthouse on Little Saint James?

110

Falken's Maze @falken.bsky.social · 16/09/2026

Yeah, but to dive in and call a layperson "impressively incorrect to the point of being detached from reality" based on this is kind of dumb. And I'm saying this as someone who hates that the word "thought" is being used.

000

Falken's Maze @falken.bsky.social · 15/09/2026

This is a confusing one because technically it's not right to call Chain of Thought "thinking", but everyone does it as a shorthand and also it is true that companies keep their CoT hidden to prevent competitors learning from it. So the "fringe" view here is more of an unhelpfully pedantic one.

110

Falken's Maze @falken.bsky.social · 15/09/2026

Yesterday you said AI was going to use chess to kill us all.

000

Falken's Maze @falken.bsky.social · 14/09/2026

TBH, I don't think most people actually understand what it is that mathematicians do. I did a BSc in maths because I was good at maths in high-school and it was a very different subject to what I expected.

040

Falken's Maze @falken.bsky.social · 14/09/2026

There are posts going around now saying Dario is looking for a bailout, but I think if he could just get an amnesty on building bigger models, the amount they would save might actually be enough to make them cash positive. Training runs can cost hundreds of millions of dollars.

020

Falken's Maze @falken.bsky.social · 14/09/2026

If they introduce legislation banning "uncertified" AI from being used to do government work or in regulated industries, for example, they can kill the competition before it starts.

1131

Falken's Maze @falken.bsky.social · 14/09/2026

To be clear, the big AI companies don't want to have to train bigger and bigger models. It's expensive and consumers can't really see the difference any more. If they stop though, it won't be long until competitors catch up with them.

1130

Falken's Maze @falken.bsky.social · 14/09/2026

This guy is such a grifter and that Magnus Carlson analogy is really lazy. The frontier AI companies are trying to drum up fear right now because they want governments to bring in regulations which favour them and make potential competition impossible. You're being played here Owen.

2591

Falken's Maze @falken.bsky.social · 12/09/2026

Do *you* know what the ambassador does? The current guy does do a lot of golfing.

100

Falken's Maze @falken.bsky.social · 12/09/2026

I hear he's in the Bilderberg Group?

0131

Falken's Maze @falken.bsky.social · 11/09/2026

Robert Fingers is a national treasure.

040

Falken's Maze @falken.bsky.social · 11/09/2026

He's already setting things up so he can say he never agreed with Trump.

062