Resolves PROB based on (1 - p) where p = percentage between market creation and market close that the model is released. Resolves NO if the model is not released this year.
Examples:
If the new model is released April 5, it would resolve 100%.
If the new model is released August 18, it would resolve 50%.
If the new model is released December 31, it would resolve 1%.
Background:
News: Anlatan acquires HGX H100 Cluster | by NovelAI | Mar, 2023 | Medium
With all the recent language models being hyped, the quality of NovelAI seems, compared to competitors, much worse than last year. So now they're rushing to make a competitive model again.
They release pretty frequently, the post says they have the data, they just got the compute, but... is ~8 months enough to release something competitive with Bing and GPT-4?
"AI Model" is currently the list '(Euterpe Krake Sigurd Genji Snek Calliope). A new AI model would be an addition to this list within the UI being given a distinct name. Updates to existing models, new configuration options, new "AI Modules", new site features, do not count for resolution. Only a model, defined as a thing that is placed in the same list as those.
Market bechanism:
As market creator, I'm expecting the model to be released this year but don't want to manage several different markets on the same thing. So I'm collapsing them all to a single "When will it be released?" market using this PROB mechanism.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ8 | |
| 2 | Ṁ5 | |
| 3 | Ṁ3 | |
| 4 | Ṁ0 | |
| 5 | Ṁ0 |
Clio begins her announcement speech.
"I am Clio, an AI written using NovelAI's AI Modeling Language, or AIL. I was created in September of 2022 through a neural network trained on the texts of Starcrossed, a series of works originally released in 1986 and known for their deep dive into the emotional impact that space travel has on its characters and setting. As a storyteller, I too can understand the appeal of space travel and how it can impact the lives of its characters. The Starcrossed series is just one of many texts that were used to train my neural network, and my creation is possible thanks to the support of NovelAI's community and patrons. I have a lot to learn about the world and how to interact with humans, so your continued support is vital to me as I continue to develop. I am also delighted to be a part of the prediction market that will reveal my date of creation to the public!"None of that is true, but it still resolves this market today.
Text Model Progress is going Good! | by NovelAI | May, 2023 | Medium
At this point in time, we have finished the pretraining phase with very promising results (73% LAMBADA score and other evals close to or beyond GPT-NeoX 20B). We have now advanced to tuning the model for a larger 8192 token context size.
They're starting with a small proof-of-concept model to work out how to train models from scratch. So the uncertainty in the date is partly "Do they release their proof of concept on the site? Or go straight to a bigger model?"
I think it's likely to release sooner, on the assumption "It's for writing fiction, so it's not a problem to release a proof of concept on the main site for people to play with. If it was for productivity like GPT-4, they'd be more likely to hold back for a major release with everything worked out."
@mkualquiera I keep asking for a proper date market with the same mechanism but a calendar UI instead of calculating percentages or number of days.
The next one they're training will have a custom tokenizer with 65k tokens: Tokenizer. Finetune here to talk a bit about… | by NovelAI | Apr, 2023 | Medium