UBI will likely come, but with many strings like a surveillance state. But the only reason UBI is necessary is because the money is broken. For a debt based system to keep going it has to keep expanding the money supply which hurts everyone that doesn’t own assets. As the free market creates more deflation the system has to expand the money supply more to keep up with it because a debt based system cannot allow prices to drop or the system collapses.
If the money was finite and totally open to everyone to use, everything would be getting cheaper in that money as the free market created more value. The system we live in steals those gains by the central planners and those closest to the money printer.
As Jeff Booth has stated before (can’t find exact quote but something like this):
“in a debt based fiat system, AI will be used to enslave you. In a Bitcoin standard, AI will be used to free you. The AI control system in the Fiat system will be created to maintain the system”
This is why centralization in powerful systems is always bad, especially money, because as the ponzi system breaks down they have to centralize and control more to keep it going. AI, CBDCs, digital IDs are all in service of that system.
Agree, the real finite currency is time. Which shows the money we use should be finite and incorruptible. Luckily we already have that money. Hopefully everyone realises it before they get ushered into another control system which steals our time and energy.
Did a discovery today that surprised me. I tried codestral 22b, qwen coder 2.5 32b but GPT OSS 20B were levels above those two. GPT OSS 20B never seems to be in the top in all tests I see but in the real world it is levels above those 2 I tried.
I don’t have the hardware to run anything beyond 8b models myself. For basic specific use-cases the tiny models can work fine, but for anything general, I immediately go to Kimi 2.6 or Deepseek v4 - not free, but not too expensive.
So I wonder what are your use-cases for these low-mid models? Have you tried Gemma 4 - I have heard it’s pretty good for mid-range things, but I don’t know the specifics as I don’t really know what the actual use-cases are for these models.
Coding python apps since both chatgpt and Anthropic have small limitis. I gave a simple litte prompt just to test the three different models. The prompt was to make a wall street exchange type digital board with scrolling text, codestral was the worst an gave a full screen window with only words in the middle that changed, qwen gave random white scrolling text using pyside2, gpt oss gave pyside6 gui with scrolling green made up up tickers with numbers with corresponding up/down arrows. The output of gpt oss was also levels above, it showed the complex reasoning it made and was easy to understand and to use the code, qwen just output code.
I used LM studio together with quite good hardware, 64GB ram, a recent 16GB Vram gpu, around 20b models work very well, around 30b struggle and slow but works. I might look into those you mentioned.
I have done some thinking and as I don’t do professional work on Ai models then I think I stay on gpt OSS 20B, it does like 95% of what I need and the only thing that costs is the electricity. gpt OSS 20b does some more small mistakes than the cloud models but it tends to fix them quite easily and the token cap is unlimited.
