MrBuddyCasino

Yeah I was thinking about giving Luna a go with my PDF data extraction, but I think I‘ll stay on Gemini. It does a very good job.

> I think he's kind of speaking past the original author.

Thats being generous. He is spinning this into a 3rd world / 1st world social justice story, while ignoring the technical points dmitry made.

I was using it via OpenCode and OpenRouter. What setup do you use?

Gemini search isn’t great by default, but Deep Research gives noticeably better results.

Spoiler: its probably fraud.

An I the only one who was disappointed with GLM 5.2 after all the hype? It was thinking forever and sometime just stopped mid task.

Uh snap it indeed is cheaper: $0.20 / $1.20 vs $0.30 / $2.50. Gemini is mostly good enough for what I do with it, but the cost savings are interesting. Gemini is still faster though.

Not sure how much the benchmarks can be trusted though: https://www.reddit.com/r/GoogleGeminiAI/comments/1vbq5vf/com...

Yes this is my impression as well. To be fair I didn't compare to Luna yet, but Gemini 3.5 Lite is a very good and cheap multi-modal data extraction model.

I like using 3.5-flash-lite for doing cheap PDF and Image data extraction stuff. I don't think there is a better bang / buck model right now (3.1 is cheaper but a lot worse).

Look at the average React site in the wild, and notice that it is dog slow due to the various performance bugs it exhibits. Apparently nobody can get it right.

Yes I don’t see the appeal of HTMX in most cases. Vuejs has less footguns than React, and LLMs can produce it decently well to at least get started and then refactor.

It is also not annoying to use. It doesn’t overcomplicate things, and its quick. Much better than eg GLM 5.2.

Pretty good bang for the buck.

The same demographic is against building new power plants, especially in their vicinity. There could have been abundant power if Nixon‘s plan to build out nuclear had been allowed to proceed.

Now China does it.

This does not bode well for American prosperity. This is going the way of Europe, complacent and non-competitive. The future belongs to China.

Its not even true. Paradoxically they lower power bills.

Well said. It is not without merit, but tends to attract the tedious killjoys and midwits.

The bureaucrats who above all value process over outcome.

Mandatory link: the Zarelli soundtrack, using Leonard Nimoy's reading.

https://open.spotify.com/track/0DflgiMYRQP3UXtZ47NBus?si=XGR...

This is great. It focuses on utility and information density over design. It looks like someone took a terminal UI and transplanted it to the web, Bloomberg terminal style.

I'm pretty sure you can build tools with this that are fast and pleasant to use.

> In general, they trend towards complicated APIs which focus on Traits, safe abstractions, Async etc. My pref: Use Rust as a nice overall language, and write low-friction APIs, vs the abstractions.

Co-signed.

Agreed but apparently BigCorp hasn’t gotten the news yet.

Today it is a great language with a runtime that is unequaled. For heavy duty business backend server load, it is the standard choice, and rightfully so.

Do I love Spring? No. Do I prefer Typescript for my little side projects? Yes.

> If you create 3 buckets of inference pods, say, 256k, 512k, and 1M, then you have to worry about filling/dynamically-scaling all of them.

That seems like it is not an awfully hard problem, because the AI buildout scale is so massive. You simply don't provision too many of the smaller buckets, and when they are fully loaded you route the traffic to the larger buckets. Now you don't have a utilisation problem.

Now of course you can simply organize it that way internally, and present a single price to the customer. Perhaps thats the smarter thing to do.

AI infra buildup is so massive that the frontier labs should be able to offer more than one level of context length to incentivize token thriftiness.

One would think compute-constrained actors like Anthropic would have done so, unless prefill isn’t really a bottleneck compared to decode?

It would still be nice to mark them.

> developers hated them, but I think that draws the wrong conclusion

Exception handling is just annoying from a syntax perspective. Also checked Exceptions have the problem that they bubble up types that a different layer shouldn’t even be aware of due to exception chaining (cause of a cause etc), unless you carefully re-throw them, which nobody did.

Lower ceremony errors are just better to deal with.

I did not expect anything else.

The proposal J with the birds has some charm, but doesn’t work as a whole.

That makes one wonder, why is there only an option for globally ordered NOTIFYs?

I would expect to have a little more tuning knobs available:

- explicitly unordered

- ordered per table, but not globally

> That said, I've spent too much of my life chasing implicit control flow to accept exceptions.

Yes. This is why unchecked exceptions are annoying. Checked exceptions are a „failed experiment“ anyway.

A result type (or even Go‘s err) with forced error handling meaningfully increases the robustness of a program in my experience.

Jim Goad wrote about this. His target was the Bay Area crowd, but Berlin is the same:

„ Among most humans, the need for social approval seems stronger than the desire to know the truth. Rejected by the Überclique, the Bay Aryans form cliques of their own. Blind puppies in a cardboard box, they crawl over each other groping for "scene" status. So weak as individuals, they truly believe that "scene" status is a worthwhile goal. They howl about "fighting fascism," yet they exhibit a strong urge to feel part of some "community," which is the first flash of the fascist impulse.

You all need a crowd. You all need a movement. You need to be surrounded by the wool of a million other sheep before you finally feel warm. You all have social consciences because you're zeros as individuals. Your compassion for others is ironically founded on your own self-hatred. You swim with "the movement" because you're lost on your own.“

https://www.jimgoad.net/bayaryan.html

You need a new distribution where that shell is the default. Boiling the ocean.