It's not just about benchmaxxing. Sincerely targeting those long-autonomy benchmarks is questionable in the first place, because naturally it drives the model to assume more and more about what you want.

The target market for frontier models is CEOs who want to lay off entire departments of their company. So the long autonomy benchmarks would seem to be sending exactly the right signal.

They're still going to have to communicate with the bots replacing those departments they lay off, or they're going to have a bad time.

We know how useful classification models are, but massive pretraining is even more useful. So useful that it's tempting to just use a generic autoregressive model for everything, rather than trying to train a specialized classifier.

I think specialized models of all sorts, if it were possible to train them on equally much data, would outperform general models. In cases where we can have the data, like in self-play in games, they clearly do and have done so for a long time.

But it's a lot of effort to train specialized models. I think it's too early to tell if Jev delivers on its promises to give the best of both worlds, but if it doesn't, we'll have to make it work. It's just the right thing to try.

> but the fascinating part is its use of partial evaluation (comptime) as a single coherent mechanism that replaces a myriad of other partial-evaluation mechanisms (macros, templates/generics, constexprs). That one mechanism is the core of the language, like macros are in lisps, and that design - whether you like it or not - is revolutionary.

That IS cool. I loved partial evaluation the first time I heard of it - Futamura projections, here we come! And if it can be used to obsolete language features I never was happy with (macros), so much better!

But until now, I never heard of it as a selling point for Zig. I assumed Zig was just yet another "C replacement but we don't want to deal with Rust's borrow checker". I will certainly have to take a closer look on Zig now. But this should be up and front in their self-promotion!

> But this should be up and front in their self-promotion!

It is! "A fresh approach to metaprogramming based on compile-time code execution and lazy evaluation" is the second selling point after simplicity: https://ziglang.org

Obviously, they can't call it partial evaluation because not many people know what that is. Zig's approach was eye opening to me. I'm very familiar with how macros are used in Scheme, but comptime is intentionally weaker (unlike macros, it's referentially transparent, so strictly weaker) and I was surprised by just how far it can go. It's not everyday that you see a new kind of a partial evaluation construct, let alone a language that's almost entirely based on it (like Lisp only for comptime).

Yeah, and this suggests to me that the link isn't causal... If you succeed at learning a language late in life, you clearly have an unusually healthy brain and your act together.

This is one of the first games which changed so much that I picked it up again and realized it's not the game I liked anymore. It may still be a fine game, but it has little association to Linley Henzell's Crawl left. These days, as more non-open games go into "updates forever" mode, it's probably going to happen more. I can see it happening real time with Minecraft.

I love that you can play every previous version for free too! I THINK I like the current version the best, but there are some other good versions in the past, like just before food disappeared (I actually thought food was a great mechanic).

Now that I'm getting back into it, I am tempted to stick with the current version for a couple of years so I can really learn it.

I'm giving ToME a go on steam deck now, I like the idea of having more of a narrative and open world. I feel like the dungeon in DCSS is more like a series of individual encounters.

I was pretty shocked at the rate of change in Crawl. I played it for a couple of years and during that time entire major game mechanics changed, like moving from many kinds of food, to only one kind of food, to completely eliminating food and hunger (while adding a completely new form of clock to kill players who didn't keep moving).

Why was the rate of change so huge? And the changes so deep? These roguelikes particularly tend to be stable, almost frozen - at least in terms of mechanics.-

There were a couple of people who were very active in development and were very opinionated about good game design. Then they slowed down a bit, and some other people came in with more ideas, they may not have been quite as radical but by then it was established that upending designs was something DCSS did. (A bit of like the problem the Esperanto offshoots had).

It was at all time driven by very heavy players, and IMO the theme especially suffered from it (by the time you're a heavy player, you have completely abstracted it into a bunch of systems, and it's very hard to see it like a new player would see it).

I started playing (from a thread on HN) ~7 years ago. The changes are both broad and deep - devs are not afraid to rework something three times until happy with the outcome. I missed food/hunger entirely but very few players speak highly of it.

There is no difficulty playing previous versions, changing and compiling your own etc etc. Devs are changing the game to be better (in their own view), more fun, more engaging. They should do that! It is a living piece of art imo

Well, players who speak highly of it probably wouldn't play the modern crawl :)

Old Crawl had a lot of systems around food. You had races that were mostly or entirely vegetarian, which made them significantly harder to play in that respect. Then there were the kobolds, which could eat almost anything, the ghouls which didn't even mind rotten chunks - but they had other problems to balance it a bit. You could opt out of the entire hunger system by playing mummies - those went all the way back to Linley Henzell, and he had the philosophy that if you wanted a bit tedious play style, you could have it (mummies were really hard to play, though!). The Hive was a dungeon side branch with a lot of bees and honeycombs, a big breakpoint on the food clock if you weren't one of the carnivorous/saprovorous races, but nasty if you didn't have poison resistance. I think they removed it entirely in modern crawl? They might as well, without food its whole purpose is gone.

Of course the game works fine without food. ToME has no food, and it's fine. But it says something about how much of the original game they ditched.

Most people I've spoken to, even players who long since ditched the game, describe food and hunger as tedium. All sorts of tedious gameplay has been targeted for a variety of reasons - but i freely concede it cost something in terms of story/worldbuilding/etc.

Mummies are still miserable to play, only now they have no upsides at all... xD

Something similar happened to Angband, when I went back to it recently it had been given an overhaul that included making the game shorter. So I went back to v3.4.1.

It's always like that - no matter what they do, you refuse to believe that "the boys in uniform" aren't basically still good and decent people.

How do you explain the double tap?

The version of "meat proxy" I've heard is the person who forwards a machine-answer to someone else who didn't ask for a machine answer, and could just as easily made one themselves.

Calling someone a "meat proxy" for being the human in the loop is wild.

I don't think that's quite the problem the OP is pointing out. A corollary of "meat proxy" is "cognitive surrender." [1] In both cases, blindly pasting AI output and being an "Enter key presser", you have surrendered your cognition. In the latter case, you're in the loop but there's not a lot of reason for you to be.

Your underlying supposition is right, though. If you are truly thinking and in the loop, you are by definition not a meat proxy. I think those are the people who are successful with AI coding tools, but it takes discipline and it's still hard to find the joy and reward when you're not the one solving the problems.

[1] https://swiftrocks.com/using-ai-without-losing-critical-thin...

I think it might be appropriate if what we're talking about is a human connecting an LLM to something it can't natively access. Like something in the physical world - a piece of hardware, a different human, software the LLM can't access. As the human in the loop your role might be to mostly just shuffle information back and forth from this external resource. But depending on the situation there could still be quite a lot of interpretation and analysis performed by the human in the loop.

But it won't tell you if you don't ask. It won't tell you, "This approach is stupid, Ada SPARK exists".

Why would any LLM 'think' in terms of trying to cite prior work?

It itself is prior work. It's asking a fish to show where the water is. The fish can't imagine that absence, and the LLM can't imagine anything not being prior work.

Nowadays AI chat often will do web searches and include links. It will do that more if you ask.

"Know what to ask for" is what will keep me with a job for a while longer, I guess.

It doesn't anchor the scale to water, but it still scales the scale to water since it's still exactly 100 degrees between its boiling point and freezing point.

At sea level (or more precisely, at an atmospheric pressure of 101325Pa).