The premise that the long prompt only made the model think 'a second longer' may have more to do with the fact that it knows about the images. So why think harder if you know the answer?

At no point does the author contemplate that.

This is just a really cool way of building.

I'm impressed. I tried Google Stitch but it was slow and useless. Sad, because Gemini has a pretty good creative flair, ironically enough.

I've never had to do 18 joins to extract information in my career. I'm sure these cases do legitimately exist but they are of course rare, even in large enterprises. Most companies are more than capable of distinguishing OLTP from OLAP and real-time from batch and design (or redesign) accordingly.

Databases and their designs shift with the use case.

But there's still value in people exploring new spaces they find interesting, even if they do not meet your personal definition of pareto-optimal.

If you pick Elasticsearch, useful as it is, you now have more than two problems. You have Elastic the company; Elasticsearch the tool; and also the clay-footed colossus, Java, to contend with.

- Chunk properly;

- Elide "obviously useless files" that give mixed signals;

- Re-rank and rechunk the whole files for top scoring matches;

- Throw in a little BM25 but with better stemming;

- Carry around a list of preferred files and ideally also terms to help re-rank;

And so on. Works great when you're an academic benchmaxing your toy Master's project. Try building a scalable vector search that runs on any codebase without knowing anything at all about it and get a decent signal out of it.

Ha.

There was that other site that tried to guess what you were thinking. It'd ask you a series of questions to try and guess --- that one was eerily good at it, too. Feels like a similar sort of thing: how quickly you can converge on a solution.

Having said that, that article seems more interested in talking about things adjacent to the Oracle of Bacon and what its author finds interesting rather than the site itself. Not sure why?

No. The sauce is in KV caching: when to evict, when to keep, how to pre-empt an active agent loop vs someone who are showing signs of inactivity at their pc, etc.

You can dial in on the difficulty: "you must be pedantic and ask that I correct misuse of terminology" vs "autocorrect my mistakes in terminology with brackets".

Super duper useful way to learn things. I wish I had AI as a kid.

Yes, there will always be special exemptions: they suck, and we suffer as developers because we cannot replicate a prod-like environment in our local dev environment.

But I laugh when I join teams and they say that "our CI servers" can run it but our shitty laptops cannot, and I wonder why they can't just... spend more money on dev machines? Or perhaps spend some engineering effort so they work on both?

I start with a Makefile. The Makefile drives everything. Docker (compose), CI build steps, linting, and more. Sometimes a project outgrows it; other times it does not.

But it starts with one unitary tool for triggering work.