Latin American government includes any country's government institution or a University from any country, without outside help from a non Latin American country. E.g., "Mexico partners with the United Kingdom on LengMex," or, "Colombia partners with AWS on CoLLLM," does not count. It needs to be, "The Autonomous Univeristy of Mexico City," or whatever.
Latin American countries will indicate ANY American country except the US or Canada. Iceland, or any Atlantic or Pacific Island State also not included in case there is any confusion on that. Territories or protectorates not included, must be completely soverign nations.
20230607 Update: Attempting to refine the bet futher:
Peg performance to be equivalent to GTP-3
Must be a fully generative LLM, equivalent to GPT-3, not a text classifier, not a fine-tuned preexisting LLM, not something that just looks for an element such as text or sentance similarity, etc.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ202 | |
| 2 | Ṁ70 | |
| 3 | Ṁ21 | |
| 4 | Ṁ8 | |
| 5 | Ṁ5 |
I have done a bunch of Google searching over time and have not found any evidence. I wanted to at least wait a while to see if news came out later. I may not have structured this market properly, but I am erroring on the side of being up-to-date rather than being, "super ground truth," finding to the fullest extent.
While there may have been some small foundational LLM trained by some group and not publicized widely that I may have missed, there seems to be nothing, "major," meaning covered widely. There may have been a military or espionage project of some kind, but obviously this has not yet been disclosed and would not have fit the criteria of the market above unless it was disclosed.
Here is some research I did with PerplexityAI which seems to agree with me on the topic. Overall it seems that private sector adoption of ChatGPT and other major services is high, but as far as foundational model training there is a huge gap, at least from what we know publicly. The largest countries in the space seem to be Brazil, Mexico and Colombia.
Have not finished doing research on this. I feel like this should close at some point soon so we're not locking up people's mana for too long, but I also think there should probably be at least a month past 2023 to let reports come out. Sorry for the arbitrary cut-off, I just don't want markets hanging out there unresolved forever.
@RobertCousineau Good question. I would say most people are expecting higher performance these days, so let's peg it to GPT-3.
@RobertCousineau Here is some mana to help offset any potential losses and as a thank you for helping to refine the market. https://manifold.markets/link/SxlVzGFU
@PatrickDelaney Feel like this is significantly moving the goal posts. That said I only have 10 invested so 🤷♂️
@BTE This was not an LLM for text generation, it was for sentence textual similarity, recognizing textual entailment, and named entity recognition.
@EdwardOrsi no, needs to be non trivial, I am thinking more statecraft level, actual LLM, otherwise it's just an LM. No fine tuning.