3 ms·

The LHC is comparably complex to LLMs and CERN is a cautionary tale. The web was invented there and yet the vast majority of the subsequent innovation and econo

The LHC is comparably complex to LLMs and CERN is a cautionary tale. The web was invented there and yet the vast majority of the subsequent innovation and economic benefit coalesced in the US because the European countries were culturally unable to capitalise on it.

varjag 2y ago
LLMs are not "comparably complex" to LHC, they are perhaps 0.1% of the complexity from engineering POV. By any metric, from system design to physical infrastructure to run them.
- VagabundoP 2y ago
  Yeah the bar to entry for LLM is far far lower. CERN is a scientific marvel.
  - varjag 2y ago
    They had to design a reference standard quality nanovoltmeter because HP/Agilent could not keep up with volume & cost. And it was only one among scores of side quests. https://hackaday.com/2021/02/26/homebrew-metrology-the-cern-way/ https://hackaday.com/2021/02/26/homebrew-metrology-the-cern-...
    - VagabundoP 2y ago
      The fallout from CERN is incredible. It really shows what's possible with these kinds of multiple decade projects. Reminds me of the Civilisation Wonders you could build.
lyu07282 2y ago
Also in Germany, their leading solar energy technology became unprofitable in the global market and was sold to China (see what happened to Q-Cells [1]). The EU can publicly finance research all they want, but if the results of which gets profited by private interests of any nationality not necessarily EU countries how does that help the EU catch up? But this does seem different since they plan on building open-source models which would benefit everyone equally (and no one in particular), it would just level the playing field more I guess? I think this and things like MCP [2] are fantastic, they would make the LLM just one interchangeable piece you can buy from anywhere or host yourself. [1] https://en.wikipedia.org/wiki/Qcells https://en.wikipedia.org/wiki/Qcells [2] https://modelcontextprotocol.io/introduction https://modelcontextprotocol.io/introduction

varjag 2y ago

LLMs are not "comparably complex" to LHC, they are perhaps 0.1% of the complexity from engineering POV. By any metric, from system design to physical infrastructure to run them.

VagabundoP 2y ago
Yeah the bar to entry for LLM is far far lower. CERN is a scientific marvel.
- varjag 2y ago
  They had to design a reference standard quality nanovoltmeter because HP/Agilent could not keep up with volume & cost. And it was only one among scores of side quests. https://hackaday.com/2021/02/26/homebrew-metrology-the-cern-way/ https://hackaday.com/2021/02/26/homebrew-metrology-the-cern-...
  - VagabundoP 2y ago
    The fallout from CERN is incredible. It really shows what's possible with these kinds of multiple decade projects. Reminds me of the Civilisation Wonders you could build.

lyu07282 2y ago

Also in Germany, their leading solar energy technology became unprofitable in the global market and was sold to China (see what happened to Q-Cells [1]). The EU can publicly finance research all they want, but if the results of which gets profited by private interests of any nationality not necessarily EU countries how does that help the EU catch up? But this does seem different since they plan on building open-source models which would benefit everyone equally (and no one in particular), it would just level the playing field more I guess? I think this and things like MCP [2] are fantastic, they would make the LLM just one interchangeable piece you can buy from anywhere or host yourself. [1] https://en.wikipedia.org/wiki/Qcells https://en.wikipedia.org/wiki/Qcells [2] https://modelcontextprotocol.io/introduction https://modelcontextprotocol.io/introduction