Claude Fable 5: Working At The Frontier

1. 0:00 – 0:00

   Intro

   1. SPSpeaker0:00

      [gentle music]

2. 0:00 – 0:14

   Fable makes previously unthinkable problem classes approachable

   1. SPSpeaker0:00

      There's this whole class of problems that people weren't even necessarily thinking about that are now more approachable with Fable.

   2. SPSpeaker0:08

      [upbeat music]

3. 0:14 – 0:45

   Hebbia: compressing the finance deal lifecycle with higher accuracy

   1. SPSpeaker0:14

      Hebbia is AI for finance. In the world of finance, compressing the deal life cycle has a massive impact to a firm's ability to compete. On our really rigorous dataset, we saw roughly a twenty percent increase. Given how difficult this, this dataset really is, I think it shows where the model really shines. As users throw more and more complex problems at the model, and the model's able to come back with correct answers, that allows them to continue to push

4. 0:45 – 0:47

   Scaling complexity: users push boundaries as reliability improves

   1. SPSpeaker0:45

      the boundaries of what it can do.

5. 0:47 – 1:09

   Cursor: a coding agent that stays on track with minimal nudging

   1. SPSpeaker0:47

      Cursor is a coding agent for building professional software. Previously, I would need to repeatedly nudge the model. With Fable, it really is able to just stay on track. I have not seen a model able to kind of do this sort of really, really deep problem-solving, um, without me really needing to say much besides, like, "Go do it."

   2. SPSpeaker1:06

      [upbeat music]

6. 1:09 – 1:39

   Base44: automating system-prompt rebuilds and expanding product scope

   1. SPSpeaker1:09

      Base44 is a vibe coding platform, meaning it allow any user to build full stack applications. Rebuilding our system prompt would require the top three engineers in the company working on it for a couple of days. With Fable, it ran for four hours and came back with ninety to ninety-five percent of what we needed. It's bringing the product to a whole new areas and a whole new, um, possibilities

7. 1:39 – 1:45

   From fear to exploration: new possibilities enabled by higher leverage

   1. SPSpeaker1:39

      that before that we were, I would say, scared to, to do.

8. 1:45 – 2:16

   Thomson Reuters: using Claude for high-stakes legal drafting

   1. SPSpeaker1:45

      Thomson Reuters has quite an extensive history. Today, we're a technology company that's really focused on professions that require expert information and require to be right all of the time. Motions in the legal context, like, these are some of the hardest to craft, um, documents that a lawyer may file. They, they may spend, you know, days or weeks. This was always a task that required far too much context and precision to be possible. With Claude, we see good potential

9. 2:16 – 2:24

   Compounding quality in legal solutions as models improve

   1. SPSpeaker2:16

      to do this. As these models improve, we see a real compounding effect to the quality of the legal solutions we're able to build.

10. 2:24 – 2:53

    Cognition (Devin): surviving massive enterprise toolchains

    1. SPSpeaker2:24

       At Cognition, we built Devin, the world's first AI software engineer. Some of our biggest customers may, like, bang with, like, thirty thousand engineers. It has decades and decades of internal tools the models will completely struggle with. Fable was a step change above any model we'd seen in terms of the ability to survive in these environments. It just, honestly, like, increased a lot the, the trust that we have to the model to take on these, like, bigger projects. [upbeat music]

Episode duration: 2:55