What a car and an LLM have in common

Diallo makes a great point about the developmens of LLMs. He uses cars as an analogy to explain the current race for AGI. Major model providers are trying to get more and more out of these models. Similar to how we optimized a car to go faster.

"More power and bigger models, won't improve the quality of the output. When Yann LeCun says that LLMs are a dead end, this is what he means. Not that the technology doesn't work, but that we have already optimized the metrics, and more horsepower is not going to improve the models significantly. We are seeing this in effect when LLMs are becoming practically interchangeable. For most tasks, it doesn’t matter which model is winning in benchmarks, they will all summarize a document in an almost undiscernible manner."

The model is good enough as it is. The challenge is to improve on the things around the model. Connecting it to the right data. Figuring out how we can use it in the right way.

Find the article here: Not the worst LLM you will ever use