bckr

Very appealing.

What do you think voters should do? How do we increase the number of doctors?

You’re both doing what you want, which I find awesome.

I’m not worried that much about random acts of violence from desperate or misguided strangers.

The crime I want eliminated is that of the elite.

> stochastic terrorism

This is a bugbear for me. The point of terrorism is that it’s a random act of violence.

“Grows” is sometimes used as a synonym for “becomes” or “becomes more”.

“Becomes smaller”

A lot of PubMed links. Awesome.

I’m curious what the research says about interacting online in e.g. Discord servers and other forums for niche communities. Also, is calling your childhood friend on the regular more or less potent than meeting IRL but recently-made friends?

It’s perhaps even more maddening than that. Even if all these factors are at play, it doesn’t mean they actually matter all that much to anyone involved. These two coworkers might otherwise really get along and respect each other, but this is one of the games that they are playing with each other.

On the surface, implicitly negotiating over who is more important sounds horribly dramatic, but it’s a game that’s happening constantly among everyone. Usually folks push and pull over some equilibrium point, one person making concessions, then the other, in turns, with the actual hierarchy determining roughly how many turns each person should concede before making a demand of the other. This is where you get dynamics like “he’s a very demanding boss but he cares a lot about his employees” (high amplitude of switching between demand and concession) or “she’s very sharp but also hard to get along with” (doesn’t concede enough to make others feel important).

Concession in this game can be anything, small to large, from being the one who opens the door to let the other through, to offering help during personal problems, to letting someone take more credit on a collaboration.

But, again, these are all played in the implicit layer. They can be raised to the explicit layer by having a “heart to heart”, like “you’re always so kind. I appreciated when you did XYZ”, or “I’d really like if sometimes you did ABC”.

For now. Open Source AI continues to make progress

He’s starting with Java? I wonder if that’s the right language to start with. What is he most interested in doing? Anyway thanks for nurturing the next generation.

Dude, I didn’t really think about using GitHub pages to make games like this! Thanks for inspiring me today.

This is instead named for an animal with a shell

We’ve only had these tools for, less than 2 years?

I think those “fall apart in 2 months” kinds of projects will still keep happening, but some of us had that experience and are refining our use of the tools. So I think in the future we will see a broader spread of “percent generated code” and degrees of success

I’m moving this way as well after about 6 months of generating 95% of my code with Cursor/Claude.

My new paradigm is something like:

- write a few paragraphs about what is needed

- have the bot take in the context and produce a prototype solution outside of the main application

- have the bot describe main integration challenges

- do that integration myself — although I’m still somewhat lazy about this and keep trying to have the bot do it after the above steps; it seems to only have maybe 50% success rate

- obviously test thoroughly

The LLM is not part of the application.

The LLM expands the text of your design into a full application.

The commenter you’re responding to is clear that they are checking the outputs.

What’s the relationship between what you’re saying and the concepts of “temperature” and “stochasticity”? The model won’t give me the same answer every time.

Right.

> The LSP could be faster, and it sometimes crashes due to syntax errors or produces zombie processes.

Yep, pure joy to read.

What’s the environment management story like?

Does it have an official language server?

I think the implication is that they were arguing poorly and wasting time

> $20 million

That’s a pittance for such a project. I wish we could see what he’d have come up with.

I’m having a hard time not being sarcastic here.

The most recent news about chatbots is that ChatGPT coached a kid on how to commit suicide.

Two arguments come to mind. 1) it’s the sycophancy! Nous and its ilk should be considered safer. 2) it’s the poor alignment. A better trained model like Claude wouldn’t have done that.

I lean #2

It’s a classic case of overfitting.

$800

> Most of his time is spent mindlessly coding

Okay. So he’s working. What do you want exactly?

It’s the last sentence that makes it sound like Claude, to me

Thanks. I was talking about the confidence measure.

I’m not being sarcastic when I say that I think supervisor agents and agent swarms in general are the way forward here

I think that article is talking about finding a previously unknown exploit. A known and well documented vulnerability should be much easier to identify