bckr

It’s called toxoid. It’s here: https://github.com/toxoidengine/toxoid

Start with source separation using demucs

I like this idea a lot! I’ll try it today. I think a version of this is how I, and most professionals(?) already work. But I do believe my process can be sharpened.

This is probably a better idea than mine, to start with at least.

Learn the basics of PyTorch and use a model from the huggingface transformers package.

Collect a small dataset and see if you can get a model to correctly identify the scenarios you give it.

Iterate from there.

Everyone having their own private multimodal assistant on the cheap will be pretty futuristic if we can make it happen

It has been amazing experiencing in real time Google go from hot new thing that works incredibly well and is constantly innovating…

To “the thing I use… because it’s,… infrastructure”

Everywhere from Docs to Gmail to Search they are being out-innovated. It’s sad, to me.

Let’s say I wanted to run a headless Logic Pro for programmatic music production. Would I use this? Or should I containerize the application itself? It’s okay if I have to run it on Apple hardware.

Okay, dialing up good-faith engagement.

How would your interpretation change if the article said this instead?

> The goal for oncall should be to continuously tune the system toward having no outages and no false alarms.

FWIW, I did only attack one sentence. This was not exactly intended to be dismissive. It was my reaction to, in my eyes, the weakest part of your argument.

> Developers don't want to ever be paged because they don't want to be bothered

This is a very reductive statement.

Developers have experienced their best colleagues burning out and leaving jobs because of on-call being completely overwhelming.

Developers want to behave intelligently.

Developers want the system to work.

Developers don’t want to burn their lifespan for false alarms that are being sent because someone didn’t spend 30 seconds thinking about whether a human being needs to be woken up in the middle of the night for whatever widget they’re slapping together.

> Things in the world are a lot more fleeting than I expected

But you can’t kill an idea. The reproduction of the thing is more thingy than the thing, since “things in the world” are expected to be less fleeting.

Electric Giraffe dies, long live Electric Giraffe

The green revolution was not only a result of the Haber-Bosch process, although that's an aspect I wasn't thinking about.

Hybridization of crop plants increased the calories per hectare per year produced. Again, I don't know if the plants were actually producing more edible nutrients per pound of fertilizer.

Consider also the green revolution which enabled more calories to be grown per hectare. I don’t know if it takes less water and nutrients, though.

How can that be possible? Are drivers not binary files?

Thank you for this awesome comment

Of course using a KB+LLM is probably not sufficient for AGI. But I see how the title implies that.

I’m more interested in, what do we know about knowledge representation and reasoning that can act as a substrate for the next generation systems.

Awesome response, thank you! I’ll be learning more about Cyc.

What is the purpose of your report? In other words what interest does ISO have in KBs?

ChatGPT is terrible with music-related prompts. It might be able to get some basic theory, the type you can read about in Wikipedia. But ask it to do anything beyond that and it will fail pretty hard.

Oh wow I got similar results with GPT4o. I’m surprised! It even messed up the second round in a similar way.

Robotic process automation is when you automate a process a business is already doing.

Models like this will be useful for RPA.

And you can give it to your cervix-owning partners.

Another lesson here is to bookmark/ memorize the url of your bank, and don’t trust search engines to take you to your bank

Okay what’s a4d

Article’s last sentence

“And we — the well-meaning engineers — shoulder much of the blame.”

Link it!

Next someone should make a hackernews meta search engine :)

> Personal transformation success is also exceedingly rare, with only 7.5% of smokers successfully quitting despite 55% attempting to quit each year and just 1.8% of obese individuals maintaining a modest 5% weight loss over five years.

I have no idea why they included this. It’s completely disconnected from the paragraph above and below it.

I’m enjoying SvelteKit + Tailwind with shadcn-svelte components.

Picking a framework advice: https://youtu.be/RTLSSGPoASU?si=UeVO_rHtYK_YS48v

Understanding the different types of frontend frameworks: https://youtu.be/CQuTF-bkOgc?si=EMKfE_sNZNqf6c-x

shadcn-svelte: https://www.shadcn-svelte.com/

I love anything that says we need to build better software.

That said, just how crappy are people’s MVPs these days?

The point of an MVP is to make something tiny and less polished, but functional, not actually broken.

> And in the age of AI, there's only going to be more software, along with LLMs that avoid the need to even use software.

This doesn’t make any sense. An LLM is software.

If you’ve got that much money in a project, it’s not a prototype.

This repo should have had all types of static analysis running automatically.

Hell, GitHub has built in secret scanning. Apparently it was only set as the default for all new repos in March 2024[].

[] https://docs.github.com/en/code-security/secret-scanning/con...