hrjn LLM Many people say that it’s always darkest at the tip of the iceberg, but I often feel that I don’t know if I’ve ever been able to see five years ahead of my time. Well, there is no doubt that natural language processing tasks have changed in dimension and are more difficult (there are some reports that using LLM for various tasks seems to exceed the performance of existing methods).
-
hrjn I have the impression that the fact that LLM is being developed in a relatively open way is more revolutionary than LLM itself.
-
For example, Google was already doing things like using AI to automatically make phone calls and gather information years ago, and the basic technology is now available in an easy-to-use form for the general public.
-
hrjn It used to be a virtual monopoly market where people without machine power and data like Google could do nothing, but before you knew it, the base model was open to the public, the API was open to the public, and even few shot learning was possible. But before you know it, the base model is open to the public, the API is open to the public, and if you want, you can even do “few shot learning”.
-
hrjn This has led to the commoditization of LLM use at a stroke. It opened up new avenues for research and business. I think it’s true that the LLM is a new way of research and business.
-
hrjn In terms of atmosphere, it may be like linux to unix, or IBM and AT compatible machines, or something similar to the pattern where what was originally a pro-platform machine becomes freely available for some reason, and then ecosystem is created around it. I wonder if it is similar to the pattern of ecosystem that is created around the originally pro-platform machines that became freely available as a result.
-
hrjn In that sense, what kind of path this LLM will take in the future may depend on whether the LLM community can be fostered, whether an open culture can be maintained, etc. I wonder if the future of this LLM depends on whether the LLM community can be fostered and whether an open culture can be maintained.
@nishio: in Unics (later Unix) against Multics (heavy box) ( Multics was proprietary until 2007)
from Diary 2023-05-01 [Ecosystem formation through open LLM]
This page is auto-translated from /nishio/オープンなLLMによるエコシステム形成 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.