Lightnews — Scholar-powered news

Ben Recht @beenwrekt.bsky.social · 13h

but there's (3) predictions of bad practice based on the iid model seldom pan out.

this is a lesson that I'd like social scientists to pay more attention to.

1

Ben Recht @beenwrekt.bsky.social · 13h

OK, this is conflating two things (my fault).

(1) the iid assumption is a stupid crutch we use to apply the law of large numbers willy nilly
(2) in machine learning, all we know is that more data is better.

(1) applies to social science, (2) doesn't.

1 1

Ben Recht @beenwrekt.bsky.social · 14h

I don't think I understand. Who is collecting small data from a non-adaptive system?

1 2

Ben Recht @beenwrekt.bsky.social · 14h

Are they real problems, though? The evidence is far less clear than the replication police want you to believe.

1 3

Ben Recht @beenwrekt.bsky.social · 14h

Almost a decade ago, I coauthored a paper asking us to rethink our theory of generalization in machine learning. Today, I’m fine putting the theory back on the shelf.

Reshelving generalization

You don't need a theorem to argue more data is better than less data

www.argmin.net

1 4 16

Ben Recht @beenwrekt.bsky.social · 1d

Astute. How do we break that cycle?

2 1

Ben Recht @beenwrekt.bsky.social · 1d

Would be happy to discuss. Though I find myself confused more often than not every time I try to read a paper from either camp. They are all convinced that explanation lies in the models rather than in the training data.

1 5

Ben Recht @beenwrekt.bsky.social · 1d

Absolutely. Yes.

2

Ben Recht @beenwrekt.bsky.social · 1d

LOL, agree.

2

Ben Recht @beenwrekt.bsky.social · 1d

There's been a bifurcation into two camps, one in the FACCT orbit and one in the Less Wrong orbit, and they want different things.

Which camp were you interested in?

3 6

Ben Recht @beenwrekt.bsky.social · 1d

I should give Brave a spin. I was using Arc, but the company pivoted to AI and abandoned their browser.

1

Ben Recht @beenwrekt.bsky.social · 1d

Keeping it real.

1 2

Ben Recht @beenwrekt.bsky.social · 1d

Kevin successfully shamed me into closing all of my tabs. And for that I am thankful.

ktb ⧉ @kevinbaker.bsky.social · 1d

I have an extension-enforced hard limit of five tabs.

YouGov @yougov.co.uk · 1d

How many browsing tabs do you typically have open?*

1: 6%
2-5: 54%
6-10: 14%
11-20: 8%
21-30: 2%
More than 30: 3%

*across all windows, on desktop/laptop

yougov.co.uk/topics/techn...

1 10

Ben Recht @beenwrekt.bsky.social · 1d

Yes. What is the diagnosis if I create dysfunctional to-do lists in multiple modalities?

I'm aware I need help.

2 3

Ben Recht @beenwrekt.bsky.social · 1d

I'd die. Currently rocking 32.

Also, I am aware this is real Sicko shit...

2

Ben Recht @beenwrekt.bsky.social · 1d

I'm always haunted by the splash image for your Boston Review essay.

A conversation. Or, Operation Head. By Niki Sublime. https://www.flickr.com/photos/nikisublime/221259238/

1 1

Ben Recht @beenwrekt.bsky.social · 1d

I couldn't figure out how to elegantly work a citation into the main text, so I went with an oblique reference in the title. I'm going to add a couple of pointers at the end to your essays.

1 1

Ben Recht @beenwrekt.bsky.social · 1d

Possible addendum: The bots replying to me on Twitter suggest our world is a pastiche of Philip K Dick and Paul Verhoeven.

Ben Recht @beenwrekt.bsky.social · 1d

OpenAI’s new recursive nostalgia ads corroborate @himself.bsky.social's theory of Philip K Dick and AI.

Henry Was Right

OpenAI selling the world as a mumblecore adaptation of Philip K Dick

www.argmin.net

2 5 15

Ben Recht @beenwrekt.bsky.social · 2d

How do you feel about "incentives?"

1 4

Ben Recht @beenwrekt.bsky.social · 2d

Optimization theory has a few important lessons for machine learning, but the list of actionable lessons is surprisingly short.

Minimal Theory

What are the most important lessons from optimization theory for machine learning?

www.argmin.net

13

Ben Recht @beenwrekt.bsky.social · 3d

Likely title for my next post: "Momentum Is Real." IYKYK.

1 1

Ben Recht @beenwrekt.bsky.social · 3d

Wrote about football again, amused by cases where even the data nerds concede you might have to consider the messiness of human psychology.

Sunday Never Knows

When do you give up on your model and trust your gut?

www.argmin.net

1 8

Ben Recht @beenwrekt.bsky.social · 3d

A pox on early meetings messing up my ritualized posting schedule.

4

Ben Recht @beenwrekt.bsky.social · 4d

Hmm, what is the gap in the go for 2 down 8? This old article by Walder says it's 59% win percentage if going for 2/47% for the XP. Using Walder's analysis, it's 50%/50% for this other case. These numbers are not precise enough to be different.

Going for 2 down 8 points: Packers pull off come-from-behind win thanks to this strategy

The Packers profited from being the first team this season to try this unconventional 2-point conversion attempt. Here's why others should follow their lead.

www.espn.com