Comment Re:Maybe Apple Is waiting for AI stability (Score 1) 24
What is the first moved advantage for AI companies to train their own models?
If I accept your argument that making work indexable makes that work fair game for including in LLMs... then the LLMs also fall into the same category. It's relatively easy to index an existing LLM and use that to train a "new" extremely similar LLM. The hundreds of billions spent curating these LLMs seem like a boondoggle for the investors. I think a couple more DeepSeek type events will see the massive money pulled out of training models.
While I see little value in companies training models, I do see opportunity in creating systems to run models. All the current server-based LLMs will be able to run locally in the next couple years and the first companies to enable local low-latency hundred-billion-parameter LLMs will have a true first mover advantage.
If I accept your argument that making work indexable makes that work fair game for including in LLMs... then the LLMs also fall into the same category. It's relatively easy to index an existing LLM and use that to train a "new" extremely similar LLM. The hundreds of billions spent curating these LLMs seem like a boondoggle for the investors. I think a couple more DeepSeek type events will see the massive money pulled out of training models.
While I see little value in companies training models, I do see opportunity in creating systems to run models. All the current server-based LLMs will be able to run locally in the next couple years and the first companies to enable local low-latency hundred-billion-parameter LLMs will have a true first mover advantage.