> because in many enterprise applications you do not need high performance

Looking at and having looked at some of their cloud spend etc, nah.

To be more specific there's a lot of fat. Whether that's HFT style high performance is a different story.

You can drop cloud spending by 10x factor and still use Spring Boot comfortably. In startup world exist infra costs benchmarks and they are very generous. That fat does not come from frameworks, it comes from not spending an extra hour on solution design.

Ask Codex how to filter it!

> most of which seem tuned to cheating benchmarks

The fix would have been requiring tests to catch the cheating. There were suggestions but it didn't happen.

It was definitely possible to catch not having sent date headers (or caching them) etc.

> Now AI just writes document.getElementById('longVariableName') javascript and straight SQL without complaining at all

I got a newer model that bypasses all that. It takes out Wireshark and send bytes straight.

> devs nowadays aren't willing to pay for development tooling

I can't speak generally because it varies but is this really the case here? Other posters have commented on missing features and issues with their product i.e. Deno Deploy so is it not willing to pay or not worth it?

Devs are notoriously hard to sell. They are hard to impress(I can do that in a weekend, which is also probably true for Deno anyway) and stingy.

> I can do that in a weekend, which is also probably true for Deno anyway

You can build Deno or even Deno deploy in a weekend? Even AI can't (not properly).

1. People say things all the time especially online. They might even tweet about their weekend project YET they might still buy it. Not equivalent.

2. If you have to buy it is usually tied to a business reason e.g. make work better or their business and so there's a lot more to it e.g. safety, regulations, etc.

A lot of the leaders of that century have been going downhill, ever since, e.g. top Japanese manufacturers.

Mistral Small 4 is already announced.

MOE but 120B range. Man I wish it was an 80B. I have 2 GPUs with 62Gib of usable VRAM. A 4bit 80B gives me some context window, but 120B puts me into system RAM

Either some q3 or since it's a MoE, maybe a REAP version of q4 might work (or could be terrible, I'm not sure about REAP'd models).

If you know what you are doing are you still vibe coding?

A lot of these vibe coded projects even have AI respond to you on any questions or issues ie no one knows what’s going on.

> If AI is AI

That "AI" got renamed to "AGI"

> The “I” in “AI” stands for “intelligence”

By that logic the “I” in Siri is 2x more intelligent.