$C3.ai, Inc.(AI)$ $Alphabet(GOOG)$ $Microsoft(MSFT)$ $Invesco QQQ Trust(QQQ)$ $Meta Platforms, Inc.(META)$ $Amazon.com(AMZN)$

The paradox of Large Language Models (LLMs) is that while it is commonly believed that simpler models are superior to more complex ones, LLMs with an enormous number of parameters have proven to be highly effective in a variety of natural language processing tasks. This has led to the question of why LLMs work when simpler models are often recommended.

One possible explanation for the effectiveness of LLMs is the recently observed double descent phenomena, which suggests that even for over-parameterized models, increasing the number of parameters can initially improve performance, followed by a decrease in performance, and then eventually an increase again. Furthermore, some researchers have shown that extremely large models may exhibit emergent behaviors that enable complex reasoning?!, and human-level performance.

An older figure, sooner or latter, these companies all have their own version of GPT-like model with pretty similar performance!