OpenAI announced these API updates 3 days ago:
- new function calling capability in the Chat Completions API
- updated and more steerable versions of
gpt-4andgpt-3.5-turbo - new 16k context version of
gpt-3.5-turbo(vs the standard 4k version) - 75% cost reduction on our state-of-the-art embeddings model
- 25% cost reduction on input tokens for
gpt-3.5-turbo - announcing the deprecation timeline for the
gpt-3.5-turbo-0301andgpt-4-0314models
There are open source LLMs. I am not saying it is wrong to consume LLM as a service, the issue is openAI seems intent on intent on not being very open.
Ah, i think there is a miss understanding of their name, it’s not open as in open source, it’s open as in open research. They publish all their research for others to duplicate. And yes, there is other models out there, but on one as good as GPT4. Unless you have a computer with 640 ram, you can’t run it. So yeah, compared to fetching data from a database that could be done on a Raspberry and generating data that requires a monster computer, i understand that they wanna put a price on the API.
Well yeah, I do get that they conduct open research, but I still think it is disingenuous for the company to not release their source code. Or at least their LLM, if we are going to be somewhat charitable and allow that their specific tooling and API infrastructure should be proprietary so that they can maintain a business. There is no guaranteed that the code running on the other end of their API adheres to any of the research that they have revealed!
I’m not too worried though, because other LLMs and parameter sets have gone open source, so the cat is already out of the bag. I also don’t really believe in the commercial viability of LLMs either, because there is no way to automate the verification that that they are generating correct content so whatever.