> I had thought Xilinx ISE and Altera Quartus had the market cornered on the most difficult environments to get running, …
These tools were the reason why I gave up on FPGAs years ago and chose to spend my time on GPUs instead. And to this day, I still don't understand why they worked so differently from what I had expected in so many ways.
I highly recommend getting a small, cheap FPGA development board with something supported by Yosys, like a Latice ICE40UP5k, Efinix T8F81C2, or GOWIN GW1NR-9.
Yosys is a super lightweight open-source FPGA toolchain that really easy to get up and running. The FPGA models I mentioned have ~5k to ~10k LUTs, can all support simple RISC-V processors, and have built-in memory and DSPs.
From the README.md: If "fancy new AI tech in an old-school minimalist package" sounds like your vibe, you might like this.
What is the question about:
A: Why start a 'new' software project? (instead of old, none, multiple, ...)
B: Why in 'C'? (instead of Mojo, Java, D, ...)
C: Why 'today'? (instead of Yesterday, Tomorrow, never, ...)
For me the beauty of this project is:
Someone tried to figure out for themselves what an end-to-end process would look like for what they wanted to do. What happens to the input? How are files written? What is sent to the LLMs?
And they used the tool they know best. No `import openai`.
The interesting part is what they try to not to say: More indications for K3 is based on distillation from Claude and GPT.
From [1]:
> As you might guess, this suggests that distilling reasoning traces may have been possible for a long time without ever breaking the cryptography.
> An anecdote: we find that prefilling Kimi-K3 reasoning with a few tokens of Opus reasoning measurably shifts its response toward Opus’s
> A small memorization analysis showed that specific Claude and GPT reasoning spans are up to ~6 orders of magnitude easier to extract from Kimi-K3 than from the next-closest model.
