throwaway7783

.. and undo logs make rollbacks and crash recoveries slower. It's a trade off. MySQL storage engine architecture is nice, postgresql extension mechanism is nice.. and so on.

Supposed to be unbiased because "Internal UK government report"?. Both the train and the riots were crimes against humanity. Sorry if Indians don't believe shit coming from the UK.

I am somehow in the "I like Red Delicious" camp. Perhaps because I was used to Kashmiri apples, that apparently share the cultivar roots of red delicious. Never liked the more acidic varieties. Actually like the "mealy" texture - (it feels grainy to me)

Classic SV is about that. Bay Area is 1.5T

Is this like agent substrate?

Do you mean to say training sets are filtered?

I mean, that goes without saying for LLMs. It is an approximation of human knowledge after all.

LLMs are becoming a world expert in everything.

I am exploring this exact area of search and analytics for vanilla postgres as replicas. Guess what, the LLM came up with this exact conclusion of using ctids as docids, all by itself. It was surreal for me to read the blog above , when I hit that paragraph about ctids.

I am no postgres internals expert.

Is the middle option asking LLM to generate a classic ML model? Or generate tons of them and pick the best?

Cap the number, sort by salaries (perhaps by coarse domains of skills and not very narrow . Nurses vs Doctors. But s/w engineers are s/w engineers).

Testing abilities is a gamable proxy. How much a company is willing to spend on the person is the best barometer, if we stay true to the intent of the visa (scarce talent)

They cannot yet outmatch humans in critical thinking. Heck, they cannot outmatch a mediocre person like me yet. How many times have we awen "You are right...." Even on sota models?

They are great tools for research and tasks as of now.

Do you think Flock is in the same category of installing PCs for police departments, when it comes to public impact?

In the short term yes. Long term, no. But I guess you can say the same thing about a team of engineers taking shortcuts

All the whistleblowers have said the alignment projects are underfunded. Surely they don't need external help to fund that? Could've done years ago?

The cynicism is about motivations and not that they are inherently not bad. Perhaps they are as bad as they claim. Or perhaps they're worse. All we have a couple of run of the mill breach examples and some people inside the talking about how dangerous it is. Yes, they are far more qualified than I am (or most people here), and perhaps there is a grain of truth. It is the motivation - and it is always money with corporations.

Yep. And the difference is clear as da for anyone using them both. And in spite of that advantage, OAI is now trying out ads. I can only imagine that even they are getting constrained to compute and are trying to find other ways to plug it

Ban non US models and form a cabal, with the blessings of the government. That's what it is looking like, no?

Same can be said about a hundred other things in the world. All the way from knives to nuclear.

When corporations are involved, it is always a good bet to err towards cynisim.

From my own standpoint, Claude has started sucking really bad (incoherent, uncontrollable verbosity slow and so on) and I stopped using it. OpenAI started experimenting with ads.

So the security issues not withstanding (no different than a human doing it or using it, but at scale), I would put my money on cynisim.

Yes. This one is a major issue. It keeps track of various changes it made in the same session and tries to be backward compatible. Have to repeatedly tell not to be backward compatible.

Things that are automatic for humans and aren't even consciously registered, have to be explicitly stated and even fought for, with LLMs.

Well, I prefer the polars version. And now if I want to reuse the CTEs elsewhere, I have to reach out to another tool like DBT or hand roll something to do string manipulation.

My issue with DBT is it is a mix of SQL, yaml, jinja2 flow controls (and metrics is whole another thing). SQL with jinja2 if/else can get really unmaintainable quickly. It's perhaps better than homegrown sql based transformers.

polars is code and can be version controlled too. Dataframes in my opinion are more elegant, and with the right backends and some lineage enhancements, could serve a much wider set of use cases than what DBT does

we have pre-commit hooks for fast tests (under a minute on a 2024 mac), a playwright script (smoke tests) that can be run on-demand locally, including setup and cleanup (~5 minutes). And on top of this we also have CI tests pre-merge (which we may drop soon)

Maybe using git notes for the conversation or conversation summary is better.

My go-to has been replicas for each use case, with well defined semantics for replication lags. I'm working on something that does most of this seamlessly (transactional,search, columnar & time series, vectors and queues) without having to bother about extension management, replication setup or tuning.

Hopefully there is some value in this - one click multipurpose postgres fleet.

It is not just the software. Let's not forget the existing customer base, an opportunity cost to build something as mature as openrouter (even with LLMs). Perhaps it is worth it for them.

We keep reinventing everything. Eclipse style plugins from a long time back

Also a very nice attack surface.

This is a great project. Thank you!

A question on 20s postgresql time - It does not look like you are accounting for reading data from disk? Wouldn't the aggregation query have to load data from disk first? Or is it somewhat guaranteed that the table is already in memory? The Rust version is clearly in memory (I am no rust expert, so that may not even be actually in memory, if its a generator).

Is this the Galleria test?