tiborsaas

> It’s a cool paper, but some of the wider commentary I’ve seen around it does not feel justified. I’m not ready to declare that the age of source code is over and our future involves incomprehensible machines editing their own model weights in incomprehensible ways.

The age of source code is definitely not going to end anytime soon if ever. But the future will involve incomprehensible machines editing other models in a "fairly certain" way, but we are going to need a lot of tooling to keep up with the process and not just take the model updates as "trust me bro".

These are pretty nice IMHO:

https://architizer.com/projects/vortex-21/

https://architizer.com/projects/flying-vegetation/

https://architizer.com/projects/tibo/

https://architizer.com/projects/icon-6/

https://architizer.com/projects/sanctuary/

It's exactly what the post says. I was waiting for my excitement to drop for the original research, but I concluded that I was mislead more than this "coverage" than the weight editing research, which was in opinion quite easy to grasp.

LLM-s operate in a higher dimensional space and they produces a 1D stream of tokens. We think of the output as all there is to how they work.

It seems like that raw weights are still within the natural language of an LLM and they can easily map weights to a latent space representation, do a transformation and map it back.

I feel that asking about training data is missing the point, they might all be able to do it naturally like I can move in 3D compared to a 2D being living on Flatland.

It will need to build up "some" tooling and as the brain is not running on the transformer architecture :) But there's already a lot of research been done in neuroscience it might find the missing pieces or connect the right research.

It would be interesting what it could do with the fruit fly connectome.

Not only this, we have no idea what the edits mean. Take 3.5% change in weights and a problem is fixed. That's cool, but how do we know if it has not planted a backdoor or crated a sleeper agent?

This is fun and games, but wait until an agents finds this and does the most misaligned thing with it in order to fold paper clips.

> Because a CPU inside a language model clearly wasn't cursed enough, I decided the prudent thing would be to give it a filesystem too.

Love the spirit :)

In case they run out elements in 2032, there's still a near infinite room by creating molecules :)

Interesting division you make there, I go to both "real", higher end restaurants and "low grade" pub restaurants. For me, the the reason is the same, convenience, laziness and lack of my expertise and willingness to cook.

Obviously I can judge a place based on the quality of service and food, but that's not the reason I've gone there in the first place.

I've done this after evaluating some "ready made please deploy me" solutions. Ended up picked Rust with Axum and it's crazy fast, code is super simple, but really powerful.

Why do people go to restaurants and pay significantly more when they can just buy the ingredients and cook at home? Those damn chefs make my food in 15 minutes.

Who is the author?

It's not really in his genre, but the live analog techno wave is definitely reacts well with the topic:

https://brandtbrauerfrick.de/ - https://www.youtube.com/watch?v=b8W_4zbiGHA

https://komfortrauschen.net/ - https://www.youtube.com/watch?v=VQ3qN2bDSgI

Then there's Jacques - https://www.youtube.com/watch?v=AAdGGLIxrwM

Satelliti: a small Italian band, but very unique sound: dark, psychedelic, electronic rock

https://www.youtube.com/watch?v=RP31BGAmDkI https://www.youtube.com/watch?v=g8epoReNLCo

Bonus track:

Kutiman - Mix Tel Aviv - https://www.youtube.com/watch?v=izSRjdYln8E

That's partially why Burial's first album brought such a fresh air to the music scene. He didn't use a DAW, but a plain, boring audio editor Sony Sound Forge and adjusted his samples until the groove and the tones felt right.

I've also seen produces complain that the default BPM is 120 so they start there and later realize that it should have been changed. Minor annoyance, but it reveals the importance of defaults.

Extinction means zero humans left, which is highly unlikely as we are a very resilient species. A global collapse is more likely sometime in the future, but even then some countries and communities might just keep going on.

> and don't solve actual problems

You can have a lot of beef with react, but this is the lamest one. Of course it solves many problems, that's why it landed at the top. It took the C out of MVC, it made view rendering deterministic and simplified event management. It also stayed somewhat stable on the API side over the years.

Zero sense for Tesla? They've just pulled off one of the most impressive marketing stunts in the automotive industry, that clearly has some sense :)

I don't understand why I can't just walk around the museum and discover things for myself.

I can't even navigate to the devices and inventions section with a single click on the listing page. The 3D scene is really well done, it would be nice if you just let people walk around freely instead of animating the camera.

And I also think you shouldn't automatically display the image in 2D, breaks my immersion immediately.

I thought the same, but players will probably push back and just don't sign up. You can't charge AAA prices for only streaming the game.

By the time we get there, AI can probably be able to recreate the game based on the video alone. I think we are already there, the fidelity of it might vary.

What will probably happen is that frontier models will lobotomize models to refuse reverse engineering. That will buy some time, however with open weight models, some will find the way to fine tune the models to support RE if magically everybody agrees on the consensus that RE should be blocked.

I think it's a ridiculous amount of money they got, but we don't know what's in the pitch deck. Investors are also blind to most of these things and take the pedigree of the founders above else.

You can replicate any fitness app in no time and you will make close to $0. Brand recognition matters a lot.

This is quite relaxing as background noise I can safely ignore and work on my project in piece. Working solo all day at home can eat into my sanity :)

It's basically big corp ASMR.

Congrats on the exit :)

Finally, the next step of forking Node is up for grabs:

Node > Deno > Done (anyone?)

No, they took the risk of being outcompeted, which is exactly my point

> Nothing wrong with that.

Except that's an open invitation to get sued into oblivion by your investors.

Here's how: https://www.youtube.com/watch?v=sfaLUi-qtnA

Another catch is that it's... PhotoShop. I'm not particularly a fan of the whole workflow. Creatives learned to live and work with it over the decades, so it works, but it's not really an intuitive tool.

I'd recommend looking into Affinity which is a merged version of PS+Illustrator. The product feels like it at least tries to be a little different. It's also free.

Missing and incomplete features, missing templates, presets and it's likely to be abandoned as issues will start to pile up.

The user experience is also not great with professional looking AI apps, I've just tried PhotoCraft and couldn't paste an image from the clipboard. Clone stamp doesn't have a live update, etc.

I'm building on Deno for the past 2 years and I have no regrets. It still delivers me why I picked it over node. Less config, less packages to install, good DX.

That's why: https://imgur.com/a/1vaUMi5

Let me juggle juggle with the small space I have the way I want.