For doomscrolling or watching videos it makes little difference. For an iPad that advertises drawing with the pencil, the latency of the brushstroke is a massive difference between the two. The more immediate feedback unlocks more confidence and reduces a lot of friction.

Oh right, peaceful protest for one's right totally deserves to be punished by government authorities with this kind of violence. When did the script flip?

I switched back to Arc a few weeks into Dia because there are so many powerful features they refused to port to Dia. I submitted them as tickets, answered their forms, replied to their "what Arc features should we bring to Dia" twitter posts and still they never made it.

Dia is a joke.

People were extremely dismissive of chinese models until recently. They went from 1 year behind frontier to 6 months behind frontier to 3 months behind frontier extremely fast.

To be clear I'm not trying to dismiss European AI. I am a proponent of it.

But Chinese models have very much earned their place. The same cannot be said of Europe, so far.

Most comments you've made here sound like you're dismissing European AI though

I think one can be realistic about where things stand now, while also hoping for more in the future.

Serious people haven't been very dimissive of Chinese models since at least DeepSeek-R1 in January 2025. Sadly, we in Europe are much behind.

Important to point out that these 'X months behind frontier' really refer to the public frontier, and not the actual frontier, which private companies are free to protect indefinitely. Perhaps open models are in actuality 18 months behind the actual frontier - how would any of us know?

Because we have access to these models and use them maybe? Because of benchmarks? It does not feel like there is much moat, look you can even use small models these days and do productive work. It is rather that private ones starting to get into diminishing return area which without moat is quite risky.

It's a bit below DeepSeek 4.1 Flash at about twice the size. For a model that was supposed to come out a few months ago this is pretty good. Mistral catching up to the chinese open-weight models is great news. Excited to see how they will build on that!

I noticed they compared with the previous best deepseek model, 4 pro 0813 for the two coding benchmarks (DeepSWE and Terminal Bench) and not Deepseek 4.1 flash. A glaring omission really.

They will RL it to be better.

What I'm looking for with this is of it hallucinates less that DeepSeek with its sparse attention, and can be used in better summarization and content generation in multiple languages.

DeepSeek sucks with languages other than English and China.

The price is really good with ML4.

But the capture is conditional so it makes sense.

if (evaluation) result = evaluation

if (evaluation) |result|

Although assignments as expressions in C are convenient, I don't think they make a ton of sense and find Zig's capture easier to read (easier to immediately see what's being evaluated)

I think you could argue for it both ways. Then you could claim ternaries are inconsistent too, shouldn't it be:

resultA : resultB ? (evaluation)

if we always put result first?

I think the most important piece of if statements and ternaries is the condition being evaluated and it's worth putting that first?

but zig has no ternaries, so it does not apply to a discussion about zig’s syntax.

and ternary is a branch with no variable introduction, so it doesn’t apply again.

(it’s basically a weak switch expression, and the value moves leftwards towards its invocation.)

22k characters of system prompt is crazy, and that is without the tool definitions.

What makes it crazy?

The fact it needs to implicitly be stated.

Every prompt will get this info, probably not only initially, since it could run out of context window. But every re-prompt in existing chats, even, gets one system prompt per model used per chat dialogue done.

Adds a fraction of cent of electric power just to every usage.

At home, I work with 65k context window, and if my system prompt and agents.md were both this length, I would spend two-thirds of the input window, before compacting which perhaps alleviates the issue for my use case, on re-feeding what mostly the reinforcement-learning should have implicitly baked in.

But what makes it crazy? If nobody told you it was 22k chars, and it gets the provider the results they want for their benchmark goals, why is it crazy?

The only observable side effect, as a user, is that the system does what you want more often than one without this prompt. Or, it stays more aligned with the provider’s guidelines. Or some combination of both.

If it was a 65k context window, then it might be a bigger deal. But it isn’t, so the comparison is moot.

That's the thing I don't think it gets the provider the results they want for their benchmarks.

It gets them what they want for their legal safety, but it actively harms the performance.

Pi with its 300 words system prompt outperforms Claude Code and Codex both in token usage and passing rate, when using the same model + effort configuration [1].

So yeah not only does it bloat context, but it runs worse too.

[1]: https://www.databricks.com/blog/benchmarking-coding-agents-d...

Because everybody has the same system prompt, the KV caching will make this a non-issue. The only cost is the reduced max context length.

Why would anyone rely on the world knowledge built into a model when the harness can just let it search for current information?

Intent understanding is a big point for sure, but world knowledge I'm not sure I see a use case for it.

I read "coding" as literally the act of being in the text editor and writing the code, not the encapsulating job of defining requirements, desigining a solution, etc. which is definitely not a solo activity.

But I agree with OP that when comes the time of sitting down and writing the code, it is a solo activity. I want to lock in and not see multiplayer cursors or avatars popping up around the editor. The part before and after that are for collaboration, but writing your piece of the code is (or at least I like it to be) solo.

Deepseek V4 Flash 0731 was such a massive jump in capability for such a small model (and price), that I'm a bit disappointed by this release.

I keep my agents on tight leashes, using them very interactively for bouncing off ideas, architecture, and then writing code (especially prototyping) and Flash has been crushing everything I ever needed it to do.

Maybe my ambitions are too tame compared to people needing Fable / Sol grade models, but I'm probably staying on Flash and not moving on to Pro for the foreseeable future.

People when OW LLM looks good in benchmarks: benchmaxxxed

People when OW LLM looks mediocre in benchmarks: disappointed