By conducting a throughout literature review over possible methods, consulting the experts, and checking if your proposed method is within laws of physics?
I agree. If Albert Einstein had gone with your sensible approach he could have quickly ruled out his silly idea of relativity as impossible and gone back to his more important work of approving patent applications.
In the case of LLMs I would say that VC funding made the engineering possible. The theoretical breakthroughs were largely made in IBM and Google. OpenAI certainly made some improvements to architecture and training, but ultimately they implemented a refined version of a transformer-based LLM.
How do you know if you can or cannot turn the idea into reality without trying?
By conducting a throughout literature review over possible methods, consulting the experts, and checking if your proposed method is within laws of physics?
I agree. If Albert Einstein had gone with your sensible approach he could have quickly ruled out his silly idea of relativity as impossible and gone back to his more important work of approving patent applications.
Venture capital is not a good way to fund theoretical work. And Einstein didn't lie about his achievements.
That there could be useful LLMs was theoretically argued about; essentially VC funding answered what wasn’t happening in academia.
In the case of LLMs I would say that VC funding made the engineering possible. The theoretical breakthroughs were largely made in IBM and Google. OpenAI certainly made some improvements to architecture and training, but ultimately they implemented a refined version of a transformer-based LLM.
But that's what venture capital and research should be about, financing ideas to see if they can be realized.
I mean, the people who put money into Theranos later should have done better due diligence, but I don't fault the initial investment.