At the end of 2024, I will take all major AI language models released in the year, and take the product of their version numbers. Resolves to YES if that product is greater than 1000.
"Major" = released to the public by one of OpenAI, Anthropic, Google, Inflection, or Meta, with the best available performance of any model released by that company.
So far, we are at 33.75: Claude (3), Gemini (1.0) Ultra, Inflection (2.5), Gemini (1.5) Pro, LLaMa (3)
Multiple models with the same version number count as long as they are not released simultaneously. So Gemini 1.5 Pro and Gemini 1.5 Ultra would be distinct if they are released apart from each other, but if they are released to the public at the same time, then they would not count. They do have to be different models, however - the various versions of GPT-4 would not count, for example.
"Best performance" is based on some sort of arguable best performance, basically just meant to exclude small models. Gemini 1.5 Pro would probably count, but 1.5 Nano would not.
In 2023, the following models were released:
GPT-4
Claude 1.3
Claude 2
Claude 2.1
LLaMa 2
Inflection 2
So the product would have been 87.36.
I won't bet on this question (except for maybe the last bet before resolving, assuming the resolution is clear).
Related questions
I'm quite confident (~90%) that all these happen this year: Gemini 1.5 Pro, Inflection 2.5, Claude 3, Llama 3, GPT-5 (or 4.5)
that's 150-168, or ~1/6 of 1000
possible additions:
Gemini Ultra 1.5
Gemini 2
Claude 3.1 (or something similar)
Claude 4
Llama 3 (non-simultaneous release)
Quite likely Gemini 2 happens, and fairly likely there's a 1.5 Ultra.
According to precedent Anthropic is relatively willing to do 0.1 increments. Fair chance but <50%(?) of Claude 4.
Very unlikely GPT-4.5 and GPT-5 both happen this year.
@MaxGhenis No, I want to keep it to the companies listed. It's a little arbitrary, but I'd rather people not have to think about all the potential companies that could release models.