I thought one core result that led to LLMs was the realization that a specialized model is not necessarily better at a task than a general one.

That goes all the way back to at least to Stein's Paradox in 1955, sadly too few people get educated about Statistics and keep thinking specialized models will necessarily be better. If you want to estimate the batting averages of 3 MLB baseball players from samples, you are better off building a model to predict all of their batting averages than computing the mean from a sample of each one separately.

https://en.wikipedia.org/wiki/Stein%27s_example

When I setup a global agents.md/claude.md for its communication style, I got the following response during the next session:

> The function you added is load-bear-very important [...]

I never felt this mocked by a computer.

The resulting value can be anything you want. You can turn a list into a tree, or another list.

Bad example:

    reduce(lambda x,y: x+x.extend([y+2,y*2,y**2]), [1,2,3,4], [])

Reduces the list to another list three times as long.

It's a reduction in the sense of a transformation (also often seen in complexity theory), not in the "this makes this smaller" everyday usage that I think about first.

I wonder if someone just got super annoyed at having to ask legal to share their documentation with a cooperation partner for export control clearance, that they decided getting it open sourced would be less of a hassle.

Now they can freely link anyone to the document.

This is 100% a valid workflow in DoD/ITAR fields. Suffer once through review, benefit eternally. It's especially important to get one good reviewed spec / presentation out, then all later reviews can be staged w.r.t. that one and are much easier to baseline and pass.

That'll be $1000 consulting fee on ITAR compliance.

Isn't $1000 a bit low for compliance consulting fees?

Well given it took me 1 minute to type that, I consider it a loss leader.

I 100% support this "loophole"

Based on the battery of 55kWh this is about efficiency, not packing more batteries:

> 6.48 kWh/100 km

At a constant 68km/h in perfect conditions (no up/down) and no additional consumers of energy (climate control).

Even on a real route they apparently got some useful stats with 7.51 kWh/100km including charging. Cool.

My car is around 17-20 kWh/100km at a 64kWh battery for a medium cost car in germany. I wish it was even close to that efficiency.

The lower is your state of charge, the more you are willing to reach the destination with your remanining capacity :) The time spend riding slower is usually not so different than when you drive normally, and it reduces the inconvenience of looking for a charger almost at the goal.

seems you compare apples and oranges, I very much doubt your stated consumption is under same conditions you describe for their consumption

ordinary cars can get to 13kWh/100km which ain't that far from 7.5 you mention

I mean I never drove Wolfsburg -> Poznań -> Vienna, and I would probably get a bit better milage if I drove an average of 70. However, even when my usage gets close to that, I still don't get down to the optimistic 17kWh/100km i put down.

> ordinary cars can get to 13kWh/100km

Unfortunately I don't have such a car. :)

> which ain't that far from 7.5 you mention

I'd say that needing a 30kWh battery for 400km range vs 52kWh is a significant difference, but that'd be opinionated hair-splitting. 13 is much better and much closer to 7.5 than what my car can do.

Based on their faq: https://www.tiktok.com/support/faq_detail?category=web_accou...

> You shouldn't see mature or complex themes, such as:

> • Profanity

> • Sexually suggestive content

> • Realistic violence or threatening imagery

> • Firearms or weapons in an environment that isn't appropriate

> • Illegal or controlled substances/drugs

> • Explicit references to mature or complex themes that may reflect personal experiences or real-world events that are intended for older audiences

It is clearly described as a content filter. (It also deactivates gifting on live and "following" feed.)

I wonder what counts as a complex theme? Surely that is a euphemism for something, but I can't figure out what it would be and certainly don't trust TikTok to do a good job of deciding what is too complex for the youth's minds.

I'd guess not-suggestive sexuality-related content, such as sex-ed. Possibly political stuff too?

Whenever I see a list like this, I wonder to myself: "Who gets to decide the list?" Why are we letting companies just decide on their own what's appropriate for "minors?"

Because 99% of adult people have other things to do than manually build content filters.

You could have, at least back when Trust and Safety roles weren't getting crushed by the AI stick.

Theres tons of research which can be brought to bear, genuine problems and challenges that require deep, thoughtful work, and some of the greatest people you will work with.

It is also absolute madness. There are so many insane, broken things about how safety is provided in the world, that you will feel like you are in a Elizabethan slaughter house, just at species scale. This is all excluding the world of child safety. Adding that to the mix gets you to existential nihilism quite rapidly.

Curious what they consider an “appropriate environment” for firearms and weapons

There are multiple Olympic events that use firearms and weapons. Or history tiktok is a thing.

I don't know what TikTok thinks, but I could imagine a target shooting sporting event as one such example.

> Prior to this change most Rust developers would never have cause to ever use `!` for any reason.

As a type.

"!" is in the most basic example nearly every rust developer has seen:

    println!("Hello, World!");

Unless you specifically watched a presentation or read a blog post about the never type, you probably haven't seen it as a type.

If the average (and nearly all new) rust developers encounters "fn bla(blub: i64) -> !" I suspect they will mostly go "huh??" (or wonder what kind of weird macro that is) until it becomes common to encounter early and gets it's own early entry in the rust book.

Compared to reading "fn blub(bla: &str) -> Never", which seems rather straight forward imo. However, "Never" might be the name of some actual Struct or Enum in various codebases.

Well, one thing that every Rust programmer knows is the `panic!(...)` macro, and its return type is "!".

That's why this compiles, even though not even match arm produces a value, which I think many Rust programmers are familiar with:

    let b = match a {
        0 => 123,
        2 => panic!("mustn't be two"),
        _ => todo!(),
    };

Other expression that are evaluated as having the type "!": return, break, continue.

The standard advice I've always read: Get active in the niche for your app/game, be it a subreddit, a discord, tiktok or whatever.

Then you get run out for self promotion.

Until yesterday I used openrouter mostly as a portal to the big providers: google, anthropic, openai, without having to maintain keys and accounts and credits for all of them separately.

Yesterday I wanted to do a bit of benchmarking a prompt across multiple models. Small requests. Outside the big providers, the experience became awful. This explains that experience.

Seems like a very popular use for AI, I work on a version for Korean. (A very diffrent feature set, more exercise generation, maybe someday I'll get to live conversation, which would be a great thing.)

However, for voicing sentences I use murf.ai, which seemed very nice for korean.

I tried to talk to the openai version in (my bad) korean, and it responded in japanese :D