David Feldman

Data science, machine learning, and AI

[David Feldman @DavidFe22353955]

@hot_rails @igypham @jakozloski To not have survival bias you'd need to randomly enforce who gets divorced and who doesn't.

Maybe getting divorced makes you less happy like you're saying. Maybe people who fight to make a marriage work are more likely to be happy in the future. Both are probably true

[David Feldman @DavidFe22353955]

@AldenRiver79959 @NoFilterSkin Why'd you write this? It's vague. I have to assume you're probably just being racist. She has a good list here.

[David Feldman @DavidFe22353955]

@xxlmsxx @JonathanRoss321 Oh nice. Those reasons make sense!

Yeah the app could probably use AI to generate a picture of what the combination of clothes would look like

[David Feldman @DavidFe22353955]

@Spybef0rey0ubuy 2016 was before the AI scaling law paper, before the 2017 crypto boom.

Almost nobody could predict that level of GPU demand.

It's like knowing what $50B market cap company is guaranteed to be $5T in 10 yrs

[David Feldman @DavidFe22353955]

@pcshipp Is the expected value worth the marketing emails

[David Feldman @DavidFe22353955]

@DanielMiessler Good point.

I wonder if long term this forces all AI models to be loss leaders.

Then companies profit on packaging the compute, harness, and security/privacy?

[David Feldman @DavidFe22353955]

@usr_bin_roygbiv Seems like you're angry about unemployed people?

[David Feldman @DavidFe22353955]

@guylander1 @rprovilij @BoringBiz_ @rprovilij what's your response to this one?

[David Feldman @DavidFe22353955]

@TitusFrieling @RyanJDeRose @markessien Supply is less constrained than demand

Demand side constraints:
- Consumer time/attention
- Consumer money
- AIs ability to do the thing instead of software
- Robotics grows software demand

Supply side constraints:
- Legal, regulatory, security needs in software
- Compute

[David Feldman @DavidFe22353955]

@Vishal_anton16 @MichaelArnaldi Find top devs to work for or work with. Just practicing you won't see your biggest gaps

[David Feldman @DavidFe22353955]

@venturetwins Yeah gotta give him bigger projects. And Claude your own questions first

[David Feldman @DavidFe22353955]

@jonallie Yup. I've use Claude Code on low for the same reason.

I've found effort levels need to match the size of the ask, or how precise you're trying to be.

If the AI is good at a thing use high and let it work. If not, use low and keep a tight feedback loop

[David Feldman @DavidFe22353955]

@NataliaHEmanuel This is too blanket statement

Imagine not seeing your kids for an additional 2 to 4 hrs a day, plus the extra toll on your body from sitting while driving.

There's remote pros and cons, they have different impacts based on your life stage and how you live your life

[David Feldman @DavidFe22353955]

@filippkowalski Yep pretty much.

The dopamine used to come at the end

Now it's all at the start, then the rest of the time is grinding through to make the last 10% good

Our brains evolved for the first, struggle then reward

[David Feldman @DavidFe22353955]

@vadi_ms @0rgest I'm sure you've seen this and thought about it but I've seen text prediction from jaw movements as well. Maybe both those combined could use much lower ultrasound power

[David Feldman @DavidFe22353955]

@vadi_ms @0rgest Very cool! If people use this at work every day, they'll be using it 2000 hrs a year. For pregnancy, it's used a total of a couple hours, a few times in a life. Whatever you build hopefully can handle 1,000x to 10,000X the use :)

[David Feldman @DavidFe22353955]

@CharlieWalsham_ @tparsi This sounds like AI

[David Feldman @DavidFe22353955]

@MatthewBerman With 25 years I'm sure all the important stuff got cheaper...😬