mattxxx

typst is so good! the syntax is cleaner, and it's much easier to write re-usable and clever formatting.

LaTeX will have a special spot in my heart, but it's pretty bloated (even minimalist distributions) and suffers from being an early pass at a problem.

It's pretty daring to do something like this. Something so brazen has a 100% chance of getting caught given enough time...

That said, installing any package is a liability, whether it's a library or an mcp server.

well, manipulation is more about intent, rather than technique... I think it doesn't matter if you chose the correct "door" if you're trying to get someone to operate against their own interests.

> unless you're a mathematician

As a young math researcher, my mentor definitely did not believe that Math was the absolute descriptor of the universe.

You can definitely imagine a scenario where the world does not operate perfectly mathematically correct though Math still exists - as an abstract separate entity.

You can do this such that everytime you recognize a new quirk in the world, then you can invent some new math/logical framework to match/approximate the current understanding. I don't know if this is the reality of this world, but when you look at things like complexity theory you have to wonder "okay... maybe we designed a useful system rather than discovering a true law of reality"

honestly, that's pretty good

I guess it's kinda nicely genuine that the "four panel comic strip" has some errors in it (misunderstanding caption + cat high-fiving itself in the bonus fifth panel)

great article - it's very true that:

1. it's very difficult to verify how a llm will behave without running it 2. there is an intentional ignorance around the security issues of running models

I think this research makes the speculative concrete

also love when a fansubber will just outright give you an asterisk explaining a joke that relies on nuance or wordplay

all code is legacy code; git repos are mausoleums

Weird thing: the timestamp on the upper right part of the camera is "01-02-1970" :thinking_face:

The article is good, but maybe a bit negative on the postgres feature. I think the article reads much better with the slant:

  "LISTEN/NOTIFY got us to this level of concurrency; here's how we diagnosed the performance cliff, and here's what we're doing now."

Which is like... cool, you were able to scale pretty far and create a lot of value before you needed to find a new solution.

agreed - it helps transpose skills.

That said, this comes up often in my office. It's just not giving really good advice in many situations - especially novel ones.

AI is super good at coming up with things that have been written ad nauseam for coding-interview-prep website

Well, firing someone for this is super weird. It seems like an attempt to censor an interpretation of the law that:

1. Criticizes a highly useful technology 2. Matches a potentially-outdated, strict interpretation of copyright law

My opinion: I think using copyrighted data to train models for sure seems classically illegal. Despite that, Humans can read a book, get inspiration, and write a new book and not be litigated against. When I look at the litany of derivative fantasy novels, it's obvious they're not all fully independent works.

Since AI is and will continue to be so useful and transformative, I think we just need to acknowledge that our laws did not accomodate this use-case, then we should change them.

Humankind cannot gain anything without first giving something in return. To obtain, something of equal value must be lost. That is Alchemy's First Law of Equivalent Exchange.

agreeing with the point the question makes here; the game theory of global politics does not work with the same morals that we prescribe individual people

Personally, I think everything should be hackable, however...

Limiting the ability to _easily_ modify what's running on a system is more about public cyber-health than the individual's freedom. Viruses + malware much more easily infect systems when they are running outside of a sandbox.

"We got GTA III on dreamcast before we got GTA VI"

^ This and we need to be continually learning on an energy budget similar to how much a human spends per hour.

Can we extract the latent understanding it has?

Would be really cool to convert it's predictive model into a computer program that predicts written in like python/C/rust/whatever, and I think that would better serve our ability to understand the world.

Agree with this choice. DNT didn't work, and now it's just a signal that could _help_ organizations track you.

I think this is funny, but reading the code also makes my head hurt.

A+

From the title, I wanted to dislike this, but Dan McKinley drops another banger slide.

Giving people the keys to the car is both 1. how you make a happy person and 2. build systems that understand and operate with the bigger picture

I think they spit out human-readable code, because they've been tried on human authors.

But you make an interesting point: eventually AI will be making for other AI's + machines, and human verification can be an after thought.

This reads solely as a sales pitch, which quickly cuts to the "we're selling this product so you don't have to think about it."

...when you actually do want to think about it (in 2024).

Right now, we're collectively still figuring out:

  1. Best chunking strategies for documents
  2. Best ways to add context around chunks of documents
  3. How to mix and match similarity search with hybrid search
  4. Best way to version and update your embeddings

Here's a few cursor perks:

  1. Auto-complete makes me type ~20% faster (I type 100+ WPM)
  2. Composer can work across a few files simultaneously to update something (e.g. updating a chrome extension's manifest while proposing a code change)
  3. Write something that you know _exactly_ how it should work but are too lazy to author it yourself (e.g. Write a function that takes 2 lists of string and pair-wise matches the most similar. Allow me to pass the similarity function as a parameter. Use openai embedding distance to find most similar pairings between these two results)

A+ article. This was excellently worded, and provides a cogent breakdown of both how people misinterpret these topics and why people need some education on their differentiation.

As someone who has been a high-performing IC, a low-performing IC, and a manager, I think so much about performance comes down to whether the organization understands that many things come down to bets. So some amount of failure has to happen, if you're going to be making truly useful things.

There's a great corollary here that bad code sticks around, because it's much harder to remove

JOINs are just hard period. When you're operating at a large scale, you need to be thinking about exactly how to partition + index your data for the types of queries that you want to write with JOINs.

Streaming joins are so hard, that they're an anti pattern. If you're using external storage to make it work, then your architecture has probably gone really wrong or you're using streams for something that you shouldn't.

I've spend a lot of time writing things using async in rust, python, and typescript, and I still find it un-intuitive / conceptually incorrect.

Once you're in an async function the rules then fundamentally change for how everything operates; it's almost like programming within a dialect of the same language. In particular, I'm referring to everything from function calling, managing concurrency, waiting on results, sleeping threads.

Comparatively, when you're in a go-block in go, you're still writing within the same dialect as outside of it.

This seems wrong, since you can't push transactionally + consistently in S3.

They address this directly in their section on concurrent writes: https://github.com/awslabs/git-remote-s3?tab=readme-ov-file#...

And in their design: https://github.com/awslabs/git-remote-s3?tab=readme-ov-file#...

But it seems like this is just the wrong tool for the job (hosting git repos).