Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Curiosity Code Podcast

Curiosity Code Podcast artwork

Jev vs LLMs: What Should Decide Fraud and Credit?

Nov 10, 202610 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Lovely listener, share with us your thoughts! When you tap your card, a model decides whether the payment is fraud in less than a tenth of a second. This autumn a startup raised $870 million for a model built for exactly these yes-or-no calls, one that cannot write a single word. Six days after it launched, independent researchers found a problem with the one thing it was sold on. A solo episode of Curiosity Code. Should the systems that decide fraud, credit and payment routing run on small specialized models or on large language models? My answer: small against large is the wrong line. Draw it at numbers against text. Transactions go to specialized models you own and can explain. Text judgment calls go to a cheap classifier first, with an LLM for what it isn't sure about. And a probability you haven't checked on your own data is only a number. Three things to do: list the decisions your systems make and mark each one as numbers or text, and keep the numbers ones on models you own, can explain and can retrain. For the text ones, put a small, cheap model in front, set the threshold by what a mistake costs, and send what it is unsure about to a large model or a person. Before you trust any threshold, take a few hundred of your own cases where you know the answer and check that 90% means 90%, for the small model and the large one. The question I can't settle: US bank regulators left generative AI out of their new model-risk guidance. A model like Jev generates no text, but a security firm says it appears to be a language model underneath. So which rulebook is it under? Tell me in the comments, or message me on LinkedIn. Chapters 0:00 A model that can't write a word 0:36 Where I stand, and the model that started this 1:37 Numbers vs text: where I draw the line 2:17 What a past guest already told me 3:50 Numbers: what fraud models run on today 5:25 Text: cheaper, and the calibration test 7:59 What the rules already say 8:47 Three things to do, and what I can't settle Episodes mentioned Abhishek Taneja, "How Do We Make Financial Advice Independent of the Advisor?": https://youtu.be/ZzGozVZEVY4 Srishti Jain Andreasen, "AI Fraud Detection in Payments": https://youtu.be/VytBROWzqDA Sources TypeSafe AI, "Introducing System One models and Jev" (15 September 2026): https://typesafe.ai/blog/introducing-system-one-models-and-jev TypeSafe AI, "TypeSafe AI raises Series AI", $870M led by Andreessen Horowitz (9 October 2026): https://typesafe.ai/blog/series-ai Sardine, "Testing TypeSafe AI's Jev for real-time fraud decisions" (2 November 2026): https://sardine.ai/blog/typesafe-ai-jev-fraud-detection Stripe, "How we built it: Stripe Radar": https://stripe.com/blog/how-we-built-it-stripe-radar Stripe Sessions 2025, payments foundation model (7 May 2025): https://stripe.com/newsroom/news/sessions-2025 Mastercard, "Mastercard supercharges consumer protection with gen AI" (1 February 2024): https://newsroom.mastercard.com/news/press/2024/february/mastercard-supercharges-consumer-protection-with-gen-ai Adyen, "The AI behind Uplift" (17 January 2025): https://www.adyen.com/knowledge-hub/the-ai-behind-uplift Visa, Large Transaction Models (2026): https://usa.visa.com/content/dam/VCOM/regional/na/us/Solutions/documents/large-transaction-models.pdf Revolut, PRAGMA (arXiv 2604.08649): https://arxiv.org/abs/2604.08649 Nubank, nuFormer (arXiv 2507.23267): https://arxiv.org/abs/2507.23267 Valdes Gonzalez, small models vs LLMs on cost, latency and accuracy (arXiv 2602.06370, February 2026): https://arxiv.org/abs/2602.06370 Ibrahim & Zaki, "Evaluating Decision Models for Text Annotation in Computational Social Science" (arXiv 2609.24574, September 2026): https://arxiv.org/abs/2609.24574 TransUnion, score migration study (12 July 2023): https://newsroom.transunion.com/score-migration-study/ EU AI Act, Annex III: https://artificialintelligenceact.eu/annex/3/ US Regulation B, 12 CFR 1002.9 (adverse action notices): https://www.ecfr.gov/current/title-12/chapter-X/part-1002/section-1002.9 Federal Reserve, SR 26-2 attachment, footnote 3 (17 April 2026): https://www.federalreserve.gov/supervisionreg/srletters/SR2602a1.pdf 🎙️ Subscribe for weekly episodes on fintech, AI, and innovation 🌐 Website: https://curiositycodepodcast.com/ 📱 Follow on LinkedIn: https://www.linkedin.com/in/alexkhomyakov/ 🎧 Listen on Spotify: https://open.spotify.com/show/4ePKf2qCLzcKATS7o6w2U5 🍎 Apple Podcasts: https://podcasts.apple.com/podcast/id1797349967 Music: "Soundbed" by Blue Saga, via Epidemic Sound. 🎙️ Subscribe for weekly episodes on fintech, AI, and innovation 🌐 Website: https://curiositycodepodcast.com/ 📱 Follow on LinkedIn: https://www.linkedin.com/in/akhomyakov/ 🎧 Listen on Spotify: https://open.spotify.com/show/curiosity-code 🍎 Apple Podcasts: https://podcasts.apple.com/podcast/the-curiosity-code-podcast/id1748596092

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Fraud Runs as a Network. Why Don't the Defenders? | Dave Bergeron

    Nov 17, 202628 minAudio

    Lovely listener, share with us your thoughts! Fraud stopped being one criminal and one victim. It runs as a supply chain: one group runs the scam, another rents out the accounts, a third launders the money and wires it...

    Credit the people on this episode

  2. How Do We Make Financial Advice Independent of the Advisor? | Abhishek Taneja

    Nov 3, 202645 minAudio

    Lovely listener, share with us your thoughts! The most important data in wealth management is not in a database. It lives in advisors' heads and in their notebooks. Abhishek Taneja, an AI and machine learning...

    Credit the people on this episode

  3. AI Coding Agents in Finance: What to Hand Over, What to Keep

    Oct 23, 20269 minAudio

    Lovely listener, share with us your thoughts! 9 out of 10 professional developers now use AI coding agents every week. This year, the best agent in a benchmark built to look like real work passed fewer than 15% of its...

    Credit the people on this episode

  4. Who's Liable When AI Agents Make Payments? | Abril Benhumea

    Oct 14, 202633 minAudio

    Lovely listener, share with us your thoughts! An AI agent books your flight, stays inside the budget you gave it, and pays for the wrong date. Who is responsible? Abril Benhumea, fintech and payments executive and...

    Credit the people on this episode

  5. Your Team Doesn't Need to Understand RAG. It Needs These 3 Things

    Oct 9, 20269 minAudio

    Lovely listener, share with us your thoughts! Prudential in Singapore built an AI assistant for its insurance representatives. Six of its own experts graded the answers, and 59% failed. Four months later 99% passed....

    Credit the people on this episode