Will a SOTA open-sourced LLM forecasting system make major use of quasilinguistic neural reps (QNRs) before 2027?
AIOne-Off2mnd
Manifold MarketsGeen KYCGeverifieerde afwikkelingsgegevensGoed gekalibreerdHuidige gemeenschapsvoorspellingNee 83%
Voorspellers
2
Vraagtype
binary
Methodologie
Play-money forecasting platform
Brontype
Voorspelling
Marktdata
Bijgewerkt 10 uur geleden
Verouderd25 apr 24, 08:15 UTC2 jan 27, 07:59 UTC
Trends
Regels
My understanding is that the talented Owain Evans (https://owainevans.github.io/) is very interested in AI judgemental forecasting nowadays.
- He links to a document by Eric Drexler on his site about quasilinguistic neural representations (QNRs) (https://owainevans.github.io/pdfs/QNRs_FHI-TR-2021-3.0.pdf).
- This market resolves YES if a commenter links to an open-sourced repository for AI judgemental forecasting (utilizing LLMs) which:
- (a) has a reasonable claim to being SOTA among AI forecasting systems within a category of judgemental forecasting (the category can potentially be niche, but it must be an area which is well-traded by humans)
- (b) outperforms the 80th percentile of Metaculus forecasters in the category, or appears at least that effective according to other measurements, according to sound testing which has taken place for more than 2 months or across >3000 reasonably distinct markets.
- (c) Makes use of QNRs in a substantial manner, such that the system would perform markedly worse if replaced with more common text and knowledge representations, such as human-readable text or vector embeddings without a graph structure.
