On January 1, 2027, a Transformer-like model will continue to hold the state-of-the-art position in most benchmark
41
1kṀ5710
2026
84%
chance

Tracking external bet: https://www.isattentionallyouneed.com/

This will resolve according to that site, whatever the conditions and caveats they have agreed on.

AI
Get
Ṁ1,000
to start trading!

People are also trading

Will any model get above human level on the Simple Bench benchmark before September 1st, 2025.
4% chance
Google DeepMind announces a model that outperforms humans on the ARC-AGI-2 benchmark before January 15, 2026
21% chance
Will an AI achieve >85% performance on the FrontierMath benchmark before 2027?
45% chance
Will transformers still be the dominant DL architecture in 2026?
81% chance
On LMSys, what will be the difference between the top model and GPT-4-1106 on Jan 1 2026?
Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?
63% chance
Will Transformer based architectures still be SOTA for language modelling by 2026?
80% chance
By EOY 2025, will the model with the lowest perplexity on Common Crawl will not be based on transformers?
10% chance
Will superposition in transformers be mostly solved by 2026?
73% chance
When will a non-Transformer model become the top open source LLM?
Sort by:

Is mamba-trwnsformer hybrid transformer like? Falcon H1 is an example of such architecture.

@ProjectVictory does it use some kind of attention mechanism?

What constitutes Transformer-like?

Is it all, or some, of these?

  • Token inputs, token outputs

  • Positional encoding

  • (MH QKV SDPA -> MLP) layers

  • Residual connections

bought Ṁ50 NO

https://arxiv.org/abs/2501.00663

Titans: Learning to Memorize at Test Time
Abstract page for arXiv paper 2501.00663: Titans: Learning to Memorize at Test Time

@JaundicedBaboon Let's see if it can leave the academia limbo where the Mamba's got stuck...

https://x.com/_philschmid/status/1846452029582959012

@MalachiteEagle Transformers, yes. Attention, not really:

followed by two shared attention blocks.

https://blog.rwkv.com/p/eagle-7b-soaring-past-transformers

It even links the wager page in its bragging points:

All while being an “Attention-Free Transformer”

I thought this wager was made today haha

Does this strictly resolve to the outcome of the bet?

@FranekZak Yep. Whatever details they come up with are in.

People are also trading

Will any model get above human level on the Simple Bench benchmark before September 1st, 2025.
-4% 1d4% chance
Google DeepMind announces a model that outperforms humans on the ARC-AGI-2 benchmark before January 15, 2026
-29% 1d21% chance
Will an AI achieve >85% performance on the FrontierMath benchmark before 2027?
45% chance
Will transformers still be the dominant DL architecture in 2026?
81% chance
On LMSys, what will be the difference between the top model and GPT-4-1106 on Jan 1 2026?
Will the most capable, public multimodal model at the end of 2027 in my judgement use a transformer-like architecture?
63% chance
Will Transformer based architectures still be SOTA for language modelling by 2026?
80% chance
By EOY 2025, will the model with the lowest perplexity on Common Crawl will not be based on transformers?
10% chance
Will superposition in transformers be mostly solved by 2026?
73% chance
When will a non-Transformer model become the top open source LLM?
© Manifold Markets, Inc.•Terms•Privacy