Sean MacRae

Staff Data Scientist @ SoFi, Dad to Charlotte & Levi, Husband to Michelle, Coder, Statistician, Bay Area Native

[Sean MacRae @seandmacrae]

@1zzyzyx1 This is AI smgdh

[Sean MacRae @seandmacrae]

@MikeDrucker There's a reason that real science starts with the null hypothesis.

If you skip that and go hunting for confirmation, you’re not doing research—you’re prone to all sorts of issues like: cherry picking, sample bias, p-hackin, etc.

[Sean MacRae @seandmacrae]

@gdb a 15.3k line diff doesn’t raise any alarms for any of you? that’s a reasonable merge request?

[Sean MacRae @seandmacrae]

@iamtrask What about data quality? Information entropy? Not all data is created equal

[Sean MacRae @seandmacrae]

@zekramu time is important

[SoFi @SoFi]

It takes an MVP to know an MVP. That’s why we couldn't be more excited to partner with MVP @joshallenqb! He’s joining #TeamSoFi to give your money an edge with SoFi Plus—the MVP for your finances that can unlock $1,000+ in annual value! Download the app to get SoFi Plus today. https://t.co/L0XeDL6lrm

[Anthony Noto @anthonynoto]

Today we’re welcoming MVP @JoshAllenQB to team @SoFi ‼️
Josh’s grit, determination, and performance at the highest level make him a great match for SoFi Plus - the MVP for your money - inspiring our members to pursue their biggest ambitions.

[Anthony Noto @anthonynoto]

I do‼️👋🔥💥😜

[Anthony Noto @anthonynoto]

Love our company our team and the #SoFiWay #Blessed‼️

[Sean MacRae @seandmacrae]

@xriskology nothing is more incentive for a person to do something than to be told not to do that thing.

[Sean MacRae @seandmacrae]

@JadeCole2112 classic confusion between outputs and outcomes.

who cares if your agent can write 10x more code? and output 10x more docs? unless those outputs affect an outcome like increasing customer satisfaction, or improving unit economics.

[Sean MacRae @seandmacrae]

@JadeCole2112 *For decades* 😆

[Sean MacRae @seandmacrae]

@jeremyphoward I feel like the way researchers evaluate models combined with overfitting on test data to juice their evals is partly to blame.

it’s why Altman claims openai models went from high school to PhD level, because they overfit on the test evals, but nobody feels it in practice.

[Chomba Bupe @ChombaBupe]

Whoever thought statistical compression at scale would result in human level intelligence must have not understood the fundamentals of machine learning that well.

Its okay, we all are incompetent to some degree.

[Anthony Noto @anthonynoto]

So proud of our @SoFi team. Leaving it all on the field every day... head down, no hype, mouth shut just cranking it out and delivering for 8 years for our members.

[SoFi @SoFi]

SoFi was built for one mission: Help people achieve financial independence so they can realize their ambitions.

Thank you to our members and clients for making Q2 another great quarter!

[Anthony Noto @anthonynoto]

The @SoFi Way!!
SoFi was built for one mission: Help people achieve financial independence so they can realize their ambitions.

[Anthony Noto @anthonynoto]

We @Sofi had an exceptional second quarter with record member and product growth and 44% adjusted net revenue growth – our highest in over 2 years. 29% EBITDA margin. The new rule of 70!! We’re building faster than ever to deliver unmatched value to our members and clients. Thank you to our @SoFi members for trusting us to help you achieve your goals.

More here: https://t.co/WUp2mIW9g5

[Sean MacRae @seandmacrae]

@pshichenko @minchoi oh sorry—$100k in VFX. yikes.

for that price I think the expectation is that the items are actual IKEA items you can buy.

I guess for certain ad placements they might not care… but in general, hallucinating products that don’t exist is a problem for commercial uses.

[Sean MacRae @seandmacrae]

@pshichenko @minchoi sure does.

just the lack of fine control in diffusion models seems like an issue for commercial applications to me.

saw one like this for Nike shoes… that don’t exist—and, that seems problematic if your pitch is that this can replace $10k in VFX.