Yeah, I’m quickly seeing a future where stuff like this AI rewrite of the Adobe suite may need to be distributed as the model that makes the software and not the software itself. If that becomes the case then we really will have entered into a world where tech bros control everything as you will quickly find yourself in a world where you can’t use the software if you don’t have the compute to make it.
I can see a completely different kind of future where local models that run on commodity software get good enough for what most people need to do. At that point we’d be in a world that Stallman envisioned with every user having full control over what’s running on their machine, and being able to build whatever features they want. We already see how it’s becoming trivial to reverse engineer complex apps like Photoshop, or to crack PS5 games using LLMs. We are likely end up in a world where closed source is dead because code itself no longer has any value because anybody can copy and reproduce any application trivially.
And we are already starting to get there. Qwen 3.8 27b is comparable in terms of capability to Opus 4.6 which was a frontier model just half a year ago. If Qwen 4 gets to about Opus 5.5 level sometime next year, that’s going to be good enough for making vast majority of software that people need. After that it’s just a matter of further optimization or hardware getting better so it can run on most computers. But just to reiterate, it’s already running on consumer grade hardware, and it can do real work today.
I’m not sure I agree with the idea that the necessary level of compute will be made available to the average person. We are already well into a paradigm where the average “PC” is an iPad or similar, with Google locking down Android and Microsoft… being Microsoft, I just don’t have any optimism that the future will have the same kind of access to personal computing power or even software access. The “tin-foil hat” part of my brain says that the purchasing years worth of chips output is part of a conspiracy to keep consumers and potential competitors at bay more than it’s a requirement for new models.
I hope I am wrong, but I am not holding my breath.
And this tech is now being developed in China as well, so you might not have to rely on western hardware either going forward. Chinese companies are building stuff on top of RSIC-V like the Alibaba chip I linked above, and I fully expect there will be a lot of focus on Linux as China continues to decouple from the US tech stack.
It looks like many AI skeptics have been having a change of heart and maybe some growing dread. You spend the time and it streamlines through the tough problem and it cracks firmware, builds valid specs, reverse engineer enterprise software, it’s been an interesting week.
In some ways I think it might be good, perhaps a significant decrease in proprietary prison cage ecosystems. But at the same time, I have an ever growing dread that we are jumping off the knowledge cliff. The most experienced people are seeing revolutionary productivity, while the least experienced are gaining more capability without the foundational knowledge.
The future may have students who learn nothing, know nothing, and worse they will be happy with nothing.
I’m ready to unplug and live a near Amish lifestyle. Fuck modernity.
How the education system adjusts to this tech will be interesting to watch. I expect that humans who understand how stuff works is going to be a requirement for the foreseeable future, and societies will inevitably adjust. There’s always been panic and chaos every time new a disruptive technology has been invented, but humanity inevitably adjusts to it. I don’t see any reason to think this will be any different.
The tech bros have insane GPU lease agreements. More than they can reasonably afford/use. When “amateurs” can recreate photoshop, 1000x “free” (would be wasted) compute at a better photoshop they control is something they can do too, and beat OSS and adobe. There are capable decompiler to readable/organized source code agents now too. At one point, the LLM labs can stop releasing new models, and just steal everything themselves.
Yeah, I’m quickly seeing a future where stuff like this AI rewrite of the Adobe suite may need to be distributed as the model that makes the software and not the software itself. If that becomes the case then we really will have entered into a world where tech bros control everything as you will quickly find yourself in a world where you can’t use the software if you don’t have the compute to make it.
I can see a completely different kind of future where local models that run on commodity software get good enough for what most people need to do. At that point we’d be in a world that Stallman envisioned with every user having full control over what’s running on their machine, and being able to build whatever features they want. We already see how it’s becoming trivial to reverse engineer complex apps like Photoshop, or to crack PS5 games using LLMs. We are likely end up in a world where closed source is dead because code itself no longer has any value because anybody can copy and reproduce any application trivially.
And we are already starting to get there. Qwen 3.8 27b is comparable in terms of capability to Opus 4.6 which was a frontier model just half a year ago. If Qwen 4 gets to about Opus 5.5 level sometime next year, that’s going to be good enough for making vast majority of software that people need. After that it’s just a matter of further optimization or hardware getting better so it can run on most computers. But just to reiterate, it’s already running on consumer grade hardware, and it can do real work today.
I’m not sure I agree with the idea that the necessary level of compute will be made available to the average person. We are already well into a paradigm where the average “PC” is an iPad or similar, with Google locking down Android and Microsoft… being Microsoft, I just don’t have any optimism that the future will have the same kind of access to personal computing power or even software access. The “tin-foil hat” part of my brain says that the purchasing years worth of chips output is part of a conspiracy to keep consumers and potential competitors at bay more than it’s a requirement for new models.
I hope I am wrong, but I am not holding my breath.
There are already chips being developed specifically for running LLMs, so it’s almost certain this will become akin to having a GPU in the future.
And this tech is now being developed in China as well, so you might not have to rely on western hardware either going forward. Chinese companies are building stuff on top of RSIC-V like the Alibaba chip I linked above, and I fully expect there will be a lot of focus on Linux as China continues to decouple from the US tech stack.
It looks like many AI skeptics have been having a change of heart and maybe some growing dread. You spend the time and it streamlines through the tough problem and it cracks firmware, builds valid specs, reverse engineer enterprise software, it’s been an interesting week.
In some ways I think it might be good, perhaps a significant decrease in proprietary prison cage ecosystems. But at the same time, I have an ever growing dread that we are jumping off the knowledge cliff. The most experienced people are seeing revolutionary productivity, while the least experienced are gaining more capability without the foundational knowledge.
The future may have students who learn nothing, know nothing, and worse they will be happy with nothing.
I’m ready to unplug and live a near Amish lifestyle. Fuck modernity.
How the education system adjusts to this tech will be interesting to watch. I expect that humans who understand how stuff works is going to be a requirement for the foreseeable future, and societies will inevitably adjust. There’s always been panic and chaos every time new a disruptive technology has been invented, but humanity inevitably adjusts to it. I don’t see any reason to think this will be any different.
The tech bros have insane GPU lease agreements. More than they can reasonably afford/use. When “amateurs” can recreate photoshop, 1000x “free” (would be wasted) compute at a better photoshop they control is something they can do too, and beat OSS and adobe. There are capable decompiler to readable/organized source code agents now too. At one point, the LLM labs can stop releasing new models, and just steal everything themselves.