United States | News & Politics

Investing in 'ai' companies who are effectively researching tech that requires the kind of cards Nvidia make is giving Nvidia a future. Companies like OpenAI think they will have massive future earnings if they manage to get society dependent upon their data centre hosted LLM's, while staying afloat in the meantime.

The reality is that these data centre LLM's aren't accurate enough for mass deployment, and I haven't seen anything that suggests they are going to be in the foreseeable future.

If your llm is 98% accurate at each step, after five steps it is 90% accurate, after ten steps 80% accurate, and 20 steps only 66% accurate. This is not good enough for plenty of day to day work activities that, when broken down into the kind of discrete steps a computer takes, easily hits these numbers.

In order to be able to fine tune an llm to make it accurate enough for mass deployment requires an organisation to host the llm itself. Which removes OpenAI or Microsoft's data centre from the equation. I wouldn't be surprised if it was already cost effective for most organisations that wanted to utilise retrieval augmented generation to improve accuracy to host their own llm in order to do so.

The size of companies large corporations or government might want to procure deployment from - Microsoft, OpenAI - are not focused on these tools and techniques, for obvious reasons (it renders their data centres redundant in many cases). Instead they are (most likely) bullshitting about their data centres being able to do everything FIRE economy workers now do, presumably with their fingers crossed hoping for huge breakthroughs in the accuracy of their llm's.

The technology is developing so rapidly that pretty much any outcome seems possible, so fuck knows. Maybe they will pull it off and 'developed' economies will become entirely dependent upon the likes of Bill Gates, Peter Thiel, Sam Altman, and Elon Musk lol.