News for hackers. Stuff that matters.
Re:Dreams of new physics fade with latest muon magnetism result by sesm on Friday August 11, 2023 @10:40AM (#37089197)
It feels dissatisfying because most of those theories are overfitting of models with tons of free parameters to all available experimental data, and when new experimental data appears a new free parameter is introduced to the model. The actually useful theories, that reduce the number of free parameters are yet to be discovered.
Re:Dreams of new physics fade with latest muon magnetism result by spacebanana7 on Friday August 11, 2023 @10:53AM (#37089340)
Parameter estimation on its own isn't real science. Hypothesis should be statements about reality, like that a given material is a superconductor.

With enough lawyering, one can argue that statements around whether model #53sa7r describes a phenomenon accurately qualify as hypothesis. But the success or failure of such hypothesis rarely make for interesting papers to anyone but statisticians.

And they don't often contribute to the human ability to reason about reality, build new things or predict the future. Like econometric models that never forecast with accuracy >10 years in advance.

Re:Dreams of new physics fade with latest muon magnetism result by Nifty3929 on Friday August 11, 2023 @10:53AM (#37089357)
This exactly. Or at least it seems this way to me, as a non expert with some knowledge. Add another particle, or term, or whatever to make it match. Feels very inelegant. I like your description of overfit, which hadn’t occurred to me before, ironically though I spend a lot of time with ML.

When teaching ML I talk about adding loops to the Ptolemaic planets-orbit-the-Earth-model - and how it actually worked pretty well because they spent a lot of time adding loops to the orbits to make it fit. It was overfit.

Re:Dreams of new physics fade with latest muon magnetism result by Timon3 on Friday August 11, 2023 @11:32AM (#37089836)
As far as I understand the Standard Model is not just corrected for inaccuracies, it has also been used to make a number of predictions. This means there is some value in it and it's not pure overfitting.
Re:Dreams of new physics fade with latest muon magnetism result by occamrazor on Friday August 11, 2023 @11:31AM (#37089826)
Physicists haven’t added new particles in a long time.
Re:Dreams of new physics fade with latest muon magnetism result by judofyr on Friday August 11, 2023 @11:45AM (#37090014)
Googling "overfitting" gives this definition:

> Overfitting is an undesirable machine learning behavior that occurs when the machine learning model gives accurate predictions for training data but not for new data.

However, most of the particles introduced in physics have been done with concrete predictions. For instance, when the top/bottom quarks were first proposed (1973-1975) the theoretical paper also included a bunch of predictions. In the following years all of the predictions turns out to be true.

Trust me, everyone wishes the Standard Model was simpler and was more "fundamental", but no one has figured out how to make it work. I'm not quite sure how "overfitting" fits here.

Re:Dreams of new physics fade with latest muon magnetism result by cyanydeez on Friday August 11, 2023 @11:00AM (#37089450)
So, basically, AI
Re:Dreams of new physics fade with latest muon magnetism result by nologic01 on Friday August 11, 2023 @11:06AM (#37089529)
Not the case. We are talking physical theory not LLM's. It is only in cosmology where there is arguably quite a bit of parametric freedom.
Re:Dreams of new physics fade with latest muon magnetism result by PartiallyTyped on Friday August 11, 2023 @11:12AM (#37089595)
Adding, physical theories are predictive. The theory is constructed first and is more constrained than what it aims to predict. It’s not machine learning where you learn a model from data, we build the model and then verify its predictions.
Re:Dreams of new physics fade with latest muon magnetism result by empath-nirvana on Friday August 11, 2023 @11:29AM (#37089806)
I think it's sort of remarkable how _few_ free parameters there are to describe, like, _all of existence_. How many are there, a few dozen? It's pretty incredible if you think about it.