Yeah these newer systems are crazy. The agent spawns a dozen subagents that all do some figuring out on the code base and the user request. Then those results get collated, then passed along to a new set of subagents that make the actual changes. Then there are agents that check stuff and tell the subagents to redo stuff or make changes. And then it gets a final check like unit tests, compilation etc. And then it's marked as done for the user. The amount of tokens this burns is crazy, but it gets them better results in the benchmarks, so it gets marketed as an improvement. In reality it's still fucking up all the damned time.
Coding with AI is like coding with a junior dev, who didn't pay attention in school, is high right now, doesn't learn and only listens half of the time. It fools people into thinking it's better, because it shits out code super fast. But the cognitive load is actually higher, because checking the code is much harder than coming up with it yourself. It's slower by far. If you are actually going faster, the quality is lacking.
The book was one of the first pieces of media I actually found to be scary in something like two decades. The idea of someone or something being out there or even in the same room. You look at it, see the threat and then disregard it completely. That's some straight up nightmare fuel.
The premise falls apart if you think about it for even a second, but still an excellent base for the story.
I have been working as a professional in the dev space for over 25 years. I have contributed to developing the same software over a dozen times, some even at the same company. Over time the tools have changed, the people have changed and my role has changed. But the overall specs have remained very similar if not the exact same.
Recently I did a job for a company where I was a consultant on a large project, they specifically sought me out for my expertise in the subject matter (having been involved in many similar projects). One of the C suites left the company near the end of the project, no reason given. This week he contacted me, he is now a C suite at another big company. They want to do the exact same project and want to hire me because of my expertise. I laughed my ass off, but am very happy with the work and he is a good guy (as far as C suites can be at least).
Yeah it looks like someone else did the headings as compared to the text itself.
There is a section called "This pattern has a name", only for it not to tell us what that name actually is. It may have a name, but they aren't inclined to share it with us apparently.
It isn't really an OS, it doesn't have the layers or the infrastructure of an OS. It's just a shitty app that can do a few pre-programmed things and those don't even work. They call it an OS because it runs without needing an OS. But that doesn't make it an OS itself. Before EFI you could just write machine code into the boot sector of any disk and the BIOS would boot from it. This used to be totally normal. My first machine was an MSX computer and it didn't really have an OS. You'd have Basic for doing stuff, or you'd plug in a cartridge and it would run the code on the ROM inside that.
No joke, when that picture went viral of a couple of idiots transporting a couple of giraffes on the freeway with a too low overpass in sight, I felt sick for a week. I still think about how horrible that must have been.
I bet I can make a bot that does that, I even have some tricks up my sleeve to make it smarter than anyone else's AI. Oh noes we have fallen victim to the paperclip problem and are wiped out by a meme machine
Riker is a bitch and it's no wonder he was stuck as a first officer all that time. You get a new captain who brings in a ton of experience and likes his ship to be run a certain way. Riker could have objected using actual arguments instead of being a dick about it. But when your captain gives you an order, you follow that damned order. Jellico demonstrated a huge amount of patience and gave Riker a lot of room. Obviously he understood Riker was having a hard time adjusting. On the other hand Riker wasn't living up to the standards of a Starfleet officer, let alone a first officer. In the words of Picard: "You don't deserve to wear that uniform"
No such thing as a free lunch tho. It's like saying solar energy on Earth is free, it's obviously not. Sure, once the panels are produced and installed, the running costs are minimal. But that doesn't mean that energy is now suddenly free. When I did the calculation on my solar installation, I took the costs of buying the panels, installing them, maintaining them and in the end tearing them down and properly recycling them. Then we calculated the estimated total energy produced during the lifetime of the system and thus arrived at a cost per unit of energy. Then we can compare that to what the cost would be as compared to other energy sources. At the time it didn't make financial sense, as over the lifetime other energy sources (which might have been solar as well, just out of large scale installations) would be cheaper. But some government subsidies, plus a feeling the cost of energy in the future is unsure and wanting to contribute to sustainable energy production made me pay for them anyways.
It's the exact same with launching a datacenter into space. Once it's up there the energy might not cost anything and running costs per satellite might be relatively low (although there still are running costs for sure, often just spread out over the entire constellation), but that doesn't mean the thing is free. Investers would want to see a return. So that means a lot of the costs are upfront, developing the system, paying the launch provider, getting the right licenses, etc, etc. Then during the lifetime of the system, it needs to sell the compute in order to make a profit. When directly competing with newer ground based systems that run cutting edge technology, it doesn't really matter where or how the compute is done. It's simply a unit of work being sold at market rate. Newer technology will push the price per unit down, as the new tech is more efficient. And it might make your compute less attractive as it's lacking in newer capabilities, so it can only be sold at a lower price.
So even if the system would be designed for a lifespan of 10 years and put into an orbit that can last 10 years, the compute would be very hard to be sold for any reasonable price after 5 years. And as mentioned, operating a satellite is far from free, there are many running costs associated.
Plus when you build a datacenter on Earth you can use it for decades. You can swap out small parts (like the servers and networking hardware), which keeps it useful. Cooling and power setups are often good for a very long time and those can also be upgraded if needed. The building itself and all of the supporting infrastructure is good for at least 50 years. And a lot of the building is dedicated to easy access for humans to do stuff like maintenance. This is a design requirement for any datacenter.
When shooting shit into space, that's it, you can't access it for upgrades or maintenance. And we've seen these past years cutting edge AI hardware is good for maybe 3 years at best. After that it's basically worthless, maybe useful for some niche uses, but mostly useless and definitely not profitable. Not that this matters much, as to keep latency down the orbits would be so low they deorbit within 3-5 years anyways, like with the current Starlink constellation.
But this is of course very useful for a cheap launch provider, as it keeps them yeeting shit into space non-stop. And what a surprise, Elon Musk is one of the people pushing this concept hard. No alternate motives there for sure.
Yeah these newer systems are crazy. The agent spawns a dozen subagents that all do some figuring out on the code base and the user request. Then those results get collated, then passed along to a new set of subagents that make the actual changes. Then there are agents that check stuff and tell the subagents to redo stuff or make changes. And then it gets a final check like unit tests, compilation etc. And then it's marked as done for the user. The amount of tokens this burns is crazy, but it gets them better results in the benchmarks, so it gets marketed as an improvement. In reality it's still fucking up all the damned time.
Coding with AI is like coding with a junior dev, who didn't pay attention in school, is high right now, doesn't learn and only listens half of the time. It fools people into thinking it's better, because it shits out code super fast. But the cognitive load is actually higher, because checking the code is much harder than coming up with it yourself. It's slower by far. If you are actually going faster, the quality is lacking.