Bruno Blumenschein

Dad, husband, sports freak, entrepreneur, in love with life, @founding grad, always looking ahead

[Bruno Blumenschein @Brunoblumen]

I just wrote my first note on Substack
https://t.co/8CWKyOaOvK

Twitter is becoming more & more blaaaah, with so much spam, bots, BS, so I'm giving it a go - Substack

[Bruno Blumenschein @Brunoblumen]

.@todoist I’ve opened a tech issue ticket yesterday. When can I expect solution or at least a feedback? Thanks

[Swish 🍒 Tennis @Zwxsh]

🇮🇹 Jannik Sinner since the start of 2024:

RECORD vs OTHER OPP.: 91-3
RECORD vs ALCARAZ: 0-5

The reincarnation of the Federer - Nadal rivalry.

[docyogi @docyogi1]

I am going to print this, put it up on my wall, and
look at it when any situation looks utterly hopeless.

[ArtButMakeItSports @ArtButSports]

Maria und ein Uhrmachermönch, Bibliothèque Royale de Belgique, ~1450, 📸 by @ThibaultCamus

[Football Meets Data @fmeetsdata]

🌍 Southern Europe rules European national team football in 21st century.

▪️ EURO winners:
2004 🏆 Greece 🇬🇷
2008 🏆 Spain 🇪🇦
2012 🏆 Spain 🇪🇦
2016 🏆 Portugal 🇵🇹
2020 🏆 Italy 🇮🇹
2024 🏆 Spain 🇪🇦

▪️ Nations League winners:
2019 🏆 Portugal 🇵🇹
2023 🏆 Spain 🇪🇦
2025 🏆 Portugal 🇵🇹

[Real Madrid Info ³⁶ @RMadridInfo]

The MOST decorated player in Real Madrid 123-year history.

👑 Luka Modric — 28 trophies

[Real Madrid C.F. 🇬🇧🇺🇸 @realmadriden]

🤍 THANK YOU.
#GraciasLuka

[Gautam Kedia @thegautam]

TL;DR: We built a transformer-based payments foundation model. It works.

For years, Stripe has been using machine learning models trained on discrete features (BIN, zip, payment method, etc.) to improve our products for users. And these feature-by-feature efforts have worked well: +15% conversion, -30% fraud.

But these models have limitations. We have to select (and therefore constrain) the features considered by the model. And each model requires task-specific training: for authorization, for fraud, for disputes, and so on.
Given the learning power of generalized transformer architectures, we wondered whether an LLM-style approach could work here. It wasn’t obvious that it would—payments is like language in some ways (structural patterns similar to syntax and semantics, temporally sequential) and extremely unlike language in others (fewer distinct ‘tokens’, contextual sparsity, fewer organizing principles akin to grammatical rules).

So we built a payments foundation model—a self-supervised network that learns dense, general-purpose vectors for every transaction, much like a language model embeds words. Trained on tens of billions of transactions, it distills each charge’s key signals into a single, versatile embedding.

You can think of the result as a vast distribution of payments in a high-dimensional vector space. The location of each embedding captures rich data, including how different elements relate to each other. Payments that share similarities naturally cluster together: transactions from the same card issuer are positioned closer together, those from the same bank even closer, and those sharing the same email address are nearly identical.

These rich embeddings make it significantly easier to spot nuanced, adversarial patterns of transactions; and to build more accurate classifiers based on both the features of an individual payment and its relationship to other payments in the sequence.

Take card-testing. Over the past couple of years traditional ML approaches (engineering new features, labeling emerging attack patterns, rapidly retraining our models) have reduced card testing for users on Stripe by 80%. But the most sophisticated card testers hide novel attack patterns in the volumes of the largest companies, so they’re hard to spot with these methods.

We built a classifier that ingests sequences of embeddings from the foundation model, and predicts if the traffic slice is under an attack. It leverages transformer architecture to detect subtle patterns across transaction sequences. And it does this all in real time so we can block attacks before they hit businesses.

This approach improved our detection rate for card-testing attacks on large users from 59% to 97% overnight.

This has an instant impact for our large users. But the real power of the foundation model is that these same embeddings can be applied across other tasks, like disputes or authorizations.

Perhaps even more fundamentally, it suggests that payments have semantic meaning. Just like words in a sentence, transactions possess complex sequential dependencies and latent feature interactions that simply can’t be captured by manual feature engineering.

Turns out attention was all payments needed!

[Ben Heisler @bennyheis]

A new Pope from Chicago needs a true Chicago introduction... 🔊

[Bruno Blumenschein @Brunoblumen]

Kad ti dragi i dobri ljudi pomognu ❤️🙏🙌

[PDC Darts @OfficialPDC]

PETER WRIGHT DETHRONES LUKE HUMPHRIES 🤯

An incredible performance from Peter Wright as he beats the now former World Champion Luke Humphries 4-1.

Simply breathtaking from Wright, averaging nearly 101 and 70% on the doubles.

📺 https://t.co/ItCofNEHJs
#WCDarts

[Josh @joshpearson180]

Clearly just rewatched Luke Littler’s record breaking 140.91 average winning set from last night

Magic 🪄

[PDC Darts @OfficialPDC]

LITTLER WINS WITH A RECORD BREAKING SET 🤯

An eleven darter, a ten darter and an eleven darter from Luke Littler as he averages 140.91 in the final set to beat Ryan Meikle.

Simply ridiculous from this special talent 🙌

He's into the Third Round!

[Juha Pekka Salo @Revontulet69]

Ova kratka priča me totalno razjebala.

[Josh @joshpearson180]

Closest we’ve ever got 👀

[Historic Vids @historyinmemes]

Mr. Bean doing a Martin Brundle impression to Martin Brundle, who had absolutely no idea

[Joe Pompliano @JoePompliano]

Phoenix Suns owner Mat Ishbia has announced new concession prices:

• $2 hot dogs
• $2 waters
• $2 sodas
• $2 chips
• $2 popcorn

A family of four used to spend $98 on hot dogs, water, and popcorn but will now pay $24 instead.

That's awesome 👏

[Joel Gascoigne @joelgascoigne]

Today is a special day at Buffer - we just crossed $20M in ARR 🎉🚀

What makes this extra special for all of us, is that this is the second time we've reached this milestone. This has been a years-long whole-team effort to turn a decline back around to growth and thriving. It has been hard-earned and we've been through a lot to get back to this place.

34 people in our 73 person team were around when we first hit $20M in ARR back in early 2019. It means the world to me that so many people stuck with Buffer through some of the toughest years we've experienced. I'm proud of the fact that we stayed true to our values, and had optimism and conviction that we'd figure it out.

And we've achieved $20M more efficiently this time around - last time we were 86 people which translated to $234K revenue per employee. This time we're 73 people and are achieving $274K revenue per employee.

Almost everyone on the team today has been part of the Buffer journey of decline and back to growth. I think this is going to serve us very well going forward. We know the pain of losing momentum, how deviating from our original purpose of serving entrepreneurs and small businesses led us astray, and we know how much we can achieve when we make bold thoughtful steps alongside achieving strong consistent output.

A fun fact about this milestone:

- It took us 7 months to go from $18M to $19M
- It took us just 4 months to go from $19M to $20M

I'm delighted that not only are we back to growth and profitability, but our growth rate is currently accelerating. We're committed to keeping up the diligence and the pace we've established, as it is going to give us more resources to invest in making Buffer even more useful.

[PDC Darts @OfficialPDC]

First 180 on the World Champs stage followed by the Cold Palmer celebration 🧊😂

Who cares if it leaves double seven... An Ally Pally legend is born in Rashad Sweeting 🇧🇸👏

[Adam Mares @Adam_Mares]

I can't believe they made grown men dress like elves for a basketball game 🤣🤦‍♂️