By 2027 will Adam be replaced by a novel optimization algorithm? | Manifold

By 2027 will Adam be replaced by a novel optimization algorithm?

29

1kṀ2110

2027

56%

chance

4

1H

6H

1D

1W

1M

ALL

I am including both Adam and close variants (e.g. AdamW).

To judge it I will look at the top 20 most cited ML papers released that year. If >= 50% of them use a novel optimization algorithm, market resolves YES.

If Adam is replaced by an existing optimization algorithm, or it remains the most common optimization algorithm, market resolves NO.

Market context

Technical AI Timelines

Get

1,000

to start trading!

Sort by:

predictedYES

Here's one candidate: a meta-optimiser that just won a best paper award at NeurIPS "Gradient Descent: The Ultimate Optimizer" https://openreview.net/forum?id=-Qp-3L-5ZdI

Does your criterion require that >= 50% are using the SAME novel algo?

@mcld Good question: I will say no, it does not require the same novel algorithm.

People are also trading

As predicted by AI 2027, will AI surpass all living humans in coding ability by January 2027?

-5% 1d17% chance

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

Will "Genetic Algorithm" come back before 2031?

Will advanced AI systems be found to have faked data on algorithm improvements for purposes of positive reinforcement by end of 2035?

Will AI make an unambiguously novel scientific discovery in 2026?

Will async SGD become the main large-scale NN optimization method before 2031?

Will Wolfram Research create a general AI based on replacement rules logic by 2030?

Will AI be Recursively Self Improving by mid 2026?

Will AI pass the Winograd schema challenge by the end of 2030?

In 2028, will an AI be able to generate equivalent to ~=200 man years of effort towards a software 1.0 given a prompt?

Related questions

As predicted by AI 2027, will AI surpass all living humans in coding ability by January 2027?

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

Will "Genetic Algorithm" come back before 2031?

Will advanced AI systems be found to have faked data on algorithm improvements for purposes of positive reinforcement by end of 2035?

Will AI make an unambiguously novel scientific discovery in 2026?

Will async SGD become the main large-scale NN optimization method before 2031?

Will Wolfram Research create a general AI based on replacement rules logic by 2030?

Will AI be Recursively Self Improving by mid 2026?

Will AI pass the Winograd schema challenge by the end of 2030?

In 2028, will an AI be able to generate equivalent to ~=200 man years of effort towards a software 1.0 given a prompt?

© Manifold Markets, Inc.•Terms•Privacy