By EOY 2025, will the model with the lowest perplexity on Common Crawl will not be based on transformers? | Manifold

By EOY 2025, will the model with the lowest perplexity on Common Crawl will not be based on transformers?

Basic

31

Ṁ3.5k

2025

25%

chance

1D

1W

1M

ALL

If perplexity on Common Crawl is not available for models, I will use other benchmarks as a surrogate. This will inherently be a judgement process. If a model has not been announced by EOY 2025 and no benchmarks have been posted publicly, it will not be counted for the purpose of this market.

"Based on transformers" for the purpose of this question will be anything with multi-headed self-attention that feeds into an MLP.

Get Ṁ600 play money

Related questions

By EOY 2026, will it seem as if deep learning hit a wall by EOY 2025?

On January 1, 2027, a Transformer-like model will continue to hold the state-of-the-art position in most benchmark

Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?

Will openAI have the most accurate LLM across most benchmarks by EOY 2024?

Will a model costing >$30M be intentionally trained to be more mechanistically interpretable by end of 2027? (see desc)

Will Transformer based architectures still be SOTA for language modelling by 2026?

When will a non-Transformer model become the top open source LLM?

Will Adam optimizer no longer be the default optimizer for training the best open source models by the end of 2026?

Will transformers still be the dominant DL architecture in 2026?

Major AI companies continue experiencing problems with public due to AI (based on transformer language model) overcome behavior limitations and acts "strange" by the end of 2027

Sort by:

predicts NO

A MoE of transformer model still counts as "based on transformers" right?

predicts NO

Just to be clear here, if we think that it will be based on transformers, then we should vote No?

predicts NO

@jonsimon correct

@jonsimon wish they didn't not write the title with not so many double negatives

predicts NO

@ConnorMcCormick oh yeah that's definitely confusing people. We'll, better for us who do understand it :)

and thanks for the exit gigacasting! $1 -> $68 in five minutes

predicts NO

there are few situations where you want to buy a new market from 50% to 99.2%, you're just giving free money away, use limit oreders pls

lmao i beat michael's bot in buying shares

predicts NO

giving me a solid +2892.5% profit for m$1

i'mt not entirely sure what NO means on this market though

@jacksonpolack The API only refreshes the data every 15 seconds, so if you're quick on the draw, it's totally doable.

predicts NO

i'mt not entirely sure what NO means on this market though

Lol, me neither.

Related questions

By EOY 2026, will it seem as if deep learning hit a wall by EOY 2025?

On January 1, 2027, a Transformer-like model will continue to hold the state-of-the-art position in most benchmark

Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?

Will openAI have the most accurate LLM across most benchmarks by EOY 2024?

Will a model costing >$30M be intentionally trained to be more mechanistically interpretable by end of 2027? (see desc)

Will Transformer based architectures still be SOTA for language modelling by 2026?

When will a non-Transformer model become the top open source LLM?

Will Adam optimizer no longer be the default optimizer for training the best open source models by the end of 2026?

Will transformers still be the dominant DL architecture in 2026?

Major AI companies continue experiencing problems with public due to AI (based on transformer language model) overcome behavior limitations and acts "strange" by the end of 2027

Related questions

By EOY 2026, will it seem as if deep learning hit a wall by EOY 2025?

Will Transformer based architectures still be SOTA for language modelling by 2026?

On January 1, 2027, a Transformer-like model will continue to hold the state-of-the-art position in most benchmark

When will a non-Transformer model become the top open source LLM?

Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?

Will Adam optimizer no longer be the default optimizer for training the best open source models by the end of 2026?

Will openAI have the most accurate LLM across most benchmarks by EOY 2024?

Will transformers still be the dominant DL architecture in 2026?

Will a model costing >$30M be intentionally trained to be more mechanistically interpretable by end of 2027? (see desc)

Major AI companies continue experiencing problems with public due to AI (based on transformer language model) overcome behavior limitations and acts "strange" by the end of 2027