The key thing is that you teach multiplication tables in a structured, incremental manner. Yes, it's just rote memorization, but the structure makes it way easier. You don't just dump all tables on the student at once and start quizzing them until they get it.

Imo not being able to select a subset of intervals to train heavily limits how useful this is.

How is the raw Gemini 3 CoT accessed? Isn't it hidden?

There are tricks on the API to get access to the raw Gemini 3 CoT, it's extremely easy compared to getting CoT of GPT-5 (very, very hard).

What are you referring to? I see the 'reasoning' in OpenRouter for GPT-5.2, I was under the impression that is the CoT.

Yes, that's exactly what I'm referring to. When you're using the direct Gemini API (AI Studio/Vertex), with specific tricks you can get the raw reasoning/CoT output of the model, not the summary.

in antigravity gemini sometimes inserts its CoT directly into code comments lol

They're not making money on inference alone because they blow ungodly amounts on R&D. Otherwise it'd be a very profitable business.

Private equity will swoop in, bankrupt the company to shirk the debt of training / R&D, and hold on to the models in a restructuring. +Enshittification to squeeze maximum profit. This is why they're referred to as vulture capitalists.

> These games are the starting point, but the bulk of the game is new puzzles combining mechanics from different games together

Seems like the puzzles are novel, but the mechanics are not?

With Blow the devil is simultaneously in the details and at the meta level.

For example in the Witness, which I consider one of the best puzzle games ever made, you get a fairly simple core mechanic, but the game builds upon it in very interesting ways. It feels like a journey of learning and always challenges you in some novel way at each step. There are also several revelations along the way, where you discover new layers on top of the core puzzles.

I would expect that this new game will feature similarly careful design.

To each their own. I found the Witness to be excruciatingly monotonous, forced and, ultimately, boring.

I enjoyed the Witness for a while but I bounced off it pretty hard in the Mountain. It wasn’t until I watched a let’s play on YouTube that I learned there was a film room, a hidden cave complex under the mountain, a time trial, and other optional secrets. I can absolutely understand a certain type of gamer liking this but for me Talos Principle (both 1 and 2) is peak puzzle genre.

That said I’ll probably buy this game if it comes out next year.

What did you think of the puzzles?

I found them quite boring since they are all repetitions on the same theme - just drawing lines on a square. It could have been a mobile game. The world doesn't feel connected to the puzzles, and the exploration aspect of it could have been a completely separate game. It feels like two games glued together, which is IMO not a good design.

It's also not a game that's very demanding from a technical performance perspective, and really has very limited numbers of active entities / animations, so why should I care about his opinions on game architecture or anything else?

> I found them quite boring since they are all repetitions on the same theme - just drawing lines on a square.

And programming is just pressing buttons on a keyboard.

> It could have been a mobile game.

I tried to play it on iOS and found the controls clunky. Interacting with some of the puzzles was difficult with my thumbs in the way.

Monotonous. More of the same. I mean, I can appreciate the creativity behind squeezing every drop from the concept, but I saw no fun in solving them.

I like puzzle games (Baba is You is fantastic) but I also didn't get far into The Witness. Braid was fantastic though.

I think 3D FPS is generally a terrible interface to puzzles. This is 2D though so maybe it will be better.

The basic mechanics look like very standard type of puzzle mechanics (e.g. Sokoban) that have been in many games over decades.

He hired a level designer who also wrote a Sokoban game. (Can’t remember the name, but it was free and web-based, IIRC.) That game had some really great, unique ideas in it, and I’d be shocked if the new Blow game was bog standard.

It was Jack Lance, who wrote Enigmash. Tragically, he died in 2023 at the age of 25. Jack Lance superlatively creative. I cannot find the words to express how much the world lost. I do not know of a finer puzzle designer.

https://jacklance.github.io/games.html

Oh, man. Yes, that’s the one. I had no idea he’d died. :/

interactions between the various mechanics in the games likely yield countless surprises, and let you build something considerably more elaborate than thesum of its parts..

The Puzzle Boy / Kwirk series of games is Sokoban-based, but has 3 different mechanics on top of that: turnstiles, pits (that can be filled by blocks), and blocks larger than 1x1. One of the things I love about it is that, each mechanic is interesting on its own, and each combination of mechanics results in levels with very different feels. Lots of puzzles with a bunch of mechanics try to throw tons of them into each level, and each level ends up feeling very samey. But judicious use of combinations can lead to a lot of interesting variety.

(I'm a fan of Kwirk. I had it as a kid on Gameboy, and thought it would have aged badly, but no it's still good!)

An almost 50% price increase. Benchmarks look nice, but 50% more nice...?

#1 models are usually priced at 2x more than the competition, and they often decrease the price right when they lose the crown.

There are too few examples to say this is a trend. There have been counterexamples of top models actually lowering the pricing bar (gpt-5, gpt-3.5-turbo, some gemini releases were even totally free [at first]).

Unfortunately not, this model is noticeably worse. I imagine horizon is either gpt 5 nano/mini.

Even near-perfect LLMs would benefit from the compiler optimizations that types allow.

However perfect LLMs would just replace compilers and programming languages above assembly completely.

It's interesting to think about what is 'optimal' when discussing LLMs; considering that the cost is per-token. So assembly would be far from optimal as it is not exactly a succinct language... A lot of common operations are repetitive and require many operations; a more abstract, higher level language might actually be inherently more succinct.

It's not just that humans aren't good at thinking in assembly language or binary, but the operations are much more granular and so it requires a lot of operations to do express something as simple as a for-loop or a function call.

I think the perfect AI might actually come up with a language closer to Python or JavaScript.

This is the fault of sloppy language. In Lean, _proofs_ (equivalent to functions) and _proof objects/certificates_ (values) need to be distinguished. You can't compute proofs, only proof objects. In the above quote, replace "proof" with "certificate" and you'll see that it's a perfectly valid (if trivial - it essentially just applies a lemma) proof.

a distinction without a difference wrt what i'm pointing out: this proof uses exactly zero mathematics just effectively checks all the values of maxDollars_spec.

Caveat: Coercions exist in Lean, so subtypes actually can be used like the supertype, similar to other languages. This is done via essentially adding an implicit casting operation when such a usage is encountered.

I think the concept of a game DSL is cool, but it just feels so undercooked to me.

Like, I'm a huge fan of gradual typing, especially TypeScript's, but gdscript's is just so primitive. Not even to speak of something like intersection or union types, even something basic like an interfaces mechanism is missing. has_method is an awful substitute - in general way too much relies on strings, making even simple refactoring a headache and breaks autocompletion. Lots of things also just aren't typable e.g. because generics are missing, pushing one to Variant. These aren't deal breakers, especially for the small-ish projects I've done, but it just feels bad.

A 'fully realized' version of gdscript would probably be great, but as is I'm just really not very fond of it and progress currently isn't exactly happening at a rapid pace (which is of course understandable).

Also - and this is definitely a lot more subjective - but I find its C++ FFI pretty ugly, even for basic stuff like working with structs. In theory using gsdcript as glue and C++ for the more core things would be a great approach (like unreal with its blueprints), but in practice I just want to avoid it as much as possible.