This is very misleading, just look at the data: https://en.wikipedia.org/wiki/List_of_unicorn_startup_compan... Yes, it's different for different countries, and some EU countries are less startup friently. But countries like Norway or Estonia have ~1 unicorn per 1M people, just 50% of US (~2)

Might it be because of more relaxed culture, and less pressure on researchers and engineers? Innovation likes freedom and exploration, not deadlines.

This stopped being a territorial dispute long ago.

Russia is only interested in a ceasefire as a pause before occupying most of Ukraine.

Not a single time Russia has even hinted to accepting presence of any sort of third party force in Ukraine for peacekeeping.

This is rich. Russia occupied and annexed Crimea, and a part of eastern Ukraine. Of course Ukraine was seeking NATO membership.

I'm not talking about the fact that Ukraine is seeking NATO membership now, I'm talking about the fact that NATO agreed to not expand to Russian borders and did anyway, putting pressure on Russia.

NATO never agreed not to epxand. That's a lie Russia is spreading, unfortunately with some success, as we can see.

Is it better to invest into TSMC or NVIDIA stocks, in context of AI growth?

nice! did you use cloud GPUs or built your own machine?

PX4 with Pixhawk FC would be much better in terms of working out of the box. But of you use it with a random "supported" FC then still lots of debugging is needed

how does it compare to 8-bit/4-bit quantization in terms of speed/accuracy?

hard to say for now, I’m curious as well, but I used simpler tests so far because of the implementation issues - most test suites are geared towards testing models and not model implementation.

I didn’t want to wait any longer with the release, but better tests will be coming soon I hope. Anecdotally, I think 30% effort should be comparable to Q8z

More importantly, this algorithm should work on top of Q8. The quality is not yet certain though - I could use help with the implementation.

I've worked a bit on inductive syntax-guided synthesis, but did it at the worst time -- right before it was made obsolete by LLMs: https://grgv.xyz/inductive_program_synthesis/

I am confused. Isnt the whole point of this that these programs provably do what they do? Isnt that the opposite of LLMs?

Seems more like LLMs made something more immediately useful.

LLMs today still struggle to met specifications exactly, which is what synthesis is good at.

There is interesting research in having LLMs cull the search space (for instance, asking the llm to provide which instructions are needed to complete the spec and provide an implementation) and then using more traditional techniques like ilp (or other symbolic means) to find the final program that corresponds to the spec. That way the result meets the spec but converges much (much!) faster than the traditional way for far complexer cases. Sometimes the llm gets it right in one shot (and the engine just verifies it is correct), sometimes it will never get it correct but ilp will finish the job just fine.

I agree this is an interesting direction. IIRC the top (open) bitvector synthesiser works like this (albeit not with ILP).

Trouble with ILP is that it doesn't work on GPUs (as of April 2024) and on CPUs synthesis and other forms of learning are hopeless. That may change in the future ...

If this is what programmers believe, I pity software users of the future.

I recall working on this for synthesizing UEFI and Linux drivers https://trustworthy.systems/publications/nicta_full_text/769... back in the day. Luckily some of the artifacts made it to open source. Unluckily the generated source code was tough for humans to read and maintain. The readability of LLM-based source code solutions from tools like Github CoPilot is amazing in comparison.

I tried to make a robot simulator with rapier, with a mixed sucess: https://grgv.xyz/blog/simulator4/ The performance is great but realism of joints and motors is lacking, and not suitable realistic simulation.