mempko

If this was true, the big tech companies would be the most optimized. Maybe efficiency isn't the biggest factor.

I believe agents are the wrong abstraction. I came up with a superset called an Abject. See https://abject.world

Seems safebox went after a subset.

Writing code is not the hard part of software development. This is coming from someone who has programmed for 30 years, writing an average of 100k+ lines a year.

The sooner programmers start thinking about modeling the domain, user mental models, architecture and data structures and less focus on the mechanics of writing code, the better.

Writing code is the EASY part. LLMs have basically solved the easiest part of software development. They however are bad at all the stuff I mentioned. LLMs don't have a point of view, you do as a software developer.

I've been building an object oriented system re-imagined in a world with LLMs called Abject (https://abject.world) and one thought I had was to build an OS that boots into my project. One way to do it would be a minimal linux distro (think firefox os or similar). Has anyone done something like this with their projects?

The author makes a great point about learning. Learning is what increases your intelligence and if we substitute learning for AI lookup we will literally get dumber. That said, AI models have a lot of information and can assist in learning. It's a tool, how will people use it? My fear is they won't use it to help learn.

I love blender. They should get all the money they need.

Most of the emissions are done by the 100 biggest corporations. That's where I'd start fixing things.

It's called evolution. Unfortunately the rate at which we are changing the climate is causing a mass extinction and die-off event.

In the article it says "does not mean that polar bears are at any less risk of extinction". Doesn't sound like 'adapting nicely' to me.

This is a harness problem just as much as it is a model problem. I've been working on Abject (https://abject.world) and the project has agents. I took a different approach than most agent frameworks via the goal system, but still I was surprised with some of the stuff the agents generated even with guardrails. It actually helped harden the system!

Back in the 90s I made a 3d engine (software renderer) and used frustum culling. But computing the frustum intersection every time was too slow. So one technique I did was add a count to each polygon. If the polygon was outside the view frustum, i added a count of N frames. Each frame if the count for a polygon was 0 it would check against the frustum, otherwise it would reduce the count and skip the polygon rendering entirely.

This worked very well but of course if the camera turned quickly, you would see pop-in. Not a modern technique, but an oldschool one.

Iran has showed it's neighbors something powerful which is US military can not protect you. The damage Iran did to us military bases is under reported.

The Dawn of Everything is not a throw away source. It's one of the most well researched anthropological and sociological works.

Thanks! Can you elaborate on what you mean by the lost art of Documentation?

Actually a lot of the enlightenment ideas (which our government is based on) came from native American critiques of European societies. Read The Dawn of Everything for the details.

Why limit to a browser? Why not the whole system. Check out this horrible thing I'm building. https://abject.world

If you don't want to hack on it, you can just download the desktop apps for Mac, Linux, or Windows. Note the system is alpha, but I believe usable enough to do some fun stuff!

pnpm awake starts the backend pnpm scry starts the client.

That's really all you need. If you want to run your own signalling server you can use whisper.

I'll update the docs to make it clear, thanks.

Their objectives keep shifting and starship is far behind schedule. Sure, it's a success if you keep objectives small. They could have tried for LEO ages ago but didn't. Each launch should maximize learning and having small objectives is anathema to that. And very wasteful.

It's almost like if you don't have dependencies and coupling, you really just end up with a toolbox with widgets that don't do anything special when put together.

This article really misses the point of Collaboration. In biology there is the concept of symbiogenesis (https://en.wikipedia.org/wiki/Symbiogenesis). We are writing and speaking on the web because of collaboration.

The point of collaboration is to put people together that when combined are greater than the sum of their parts. You take person A individually and they may to X, you take person B individually and they may do Y. But individually X and Y might not be that valuable, it's their combination and the glue that combines them that is valuable.

What the article misses is that yes, 20% do 80% of the work. However, you can't predict which 20% will do 80% of the work. Not only that but it's not always the same 20% that do 80% of the work for all tasks and projects.

Collaboration is the 'glue', that little bit of added information, that combines the work of individuals into truly something great.

The challenge is how do you combine the work of individuals and I can tell you what doesn't work, rigidly planning and executing.

You are right. We don't need more EVs. Lets get rid of cars completely and built cheap electrified public transport. Make ICE cars illiegal. Going all EV won't help the environment. Going all public transport would.

John Carmack already had his bag and reputation when he did the code dumps.

I think the magic sauce in this project is the fact that they convert diffs in spec to diffs in code, which is likely more stable than just regenerating the whole thing.

Senior engineers have been 'vibe coding' for over a decade before AI. Think what they do, they look at PRs all day and comment. Magically the code change reflecting their comments. It's the same thing now but machines are doing it, not humans. The issue is that junior engineers have no experience working like senior engineers. The reality is that it's not that hard to work in this way. There is no excuse for software companies not being able to re-train their more junior engineers to work this way.

Literally everyone working for a wage is creating more value for others than they consume. The problem is the rent seekers are capturing that value. This is a crucial part geohot is missing. What people often don't think about and they should is that when you are employed for a wage by someone, you are the creditor and the employer is the debtor. Why? Because you work for say 2 weeks, and THEN they pay you. That whole time working you are crediting the employer and they are in debt to YOU. Think about how crazy this is. You the creditor must do what your debtor tells you. Maybe we don't think about it because it's too painful.

That's it, we have reached the Million Dollar Homepage part of this ride (https://milliondollarhomepage.com/). Make sure to take off your seatbelts and exit to your right.

I used to use a Sony phone with Sailfish but stopped when US shifted to voice over LTE and phones I used were not supported by the networks. If this phone works on US networks, I can't wait to get rid of my Android phone for sailfish. I vibed with Sailfish so hard.

The fact that you are thinking at the company level instead of an international world wide Manhattan project level should suggest how much you might not grasp the scale of the problem.

You forget the UI killed a million Iraqis and also had a torture prison. I don't think the US has every had the high ground.

Since the industrial revolution we've emitted about 1.5 trillion tonnes of CO2. Direct air capture requires roughly 1,500 kWh per tonne, so recapturing all of it would take around 2,250,000 TWh. Current global electricity production is about 30,000 TWh/year. That's 75 years of the entire world's electricity output just for capture, before you even convert it back to fuel, which costs even more energy. And thermodynamically you can never break even: we only extracted maybe 30-40% of fossil fuel energy as useful work, but reversing the dispersal of CO2 from 420ppm in the atmosphere fights entropy all the way back. It will always cost more energy to put back than we got taking it out. As for the nuclear numbers: Vogtle, the only recent US build, came in at ~$16B/GW, not $3-4B. The world started construction on 9 reactors total in 2024. The all time peak was ~30/year in the 1980s. 300/year has never been close to reality. Average build time is about 9 years per reactor. I'm not anti-nuclear but you can't hand-wave your way past thermodynamics and industrial scaling with "it's just a policy choice."

Regarding carbon capture, it will take more energy to capture the carbon than we burned putting it up there in the first place. Alan Kay, who actually did some systems work on the environment, explained it to me that the climate system is like an upside down coke bottle. It doesn't take much energy to tip it over, but it takes a lot more to put it back up.

In other words, we shouldn't have tipped it over in the first place. We may not have the energy to put things back to a habitable place.