It's Friday, so here's a huge @AskMarvinAI 1.1 release!

🤝 Support for @AnthropicAI and @Azure
🗺️ Map over data in parallel
📝 Guide behavior with instructions

factsmachine.ai/p/marvin-11-mo… In addition to OpenAI, Marvin now supports Anthropic and the Azure OpenAI Service.

Set MARVIN_LLM_MODEL or change the global setting at runtime:

🥳 Marvin's turning 1.0!

@AskMarvinAI is our lightweight AI engineering framework for building software that's reliable, scalable, and easy to trust.

If you know Python, you know Marvin: its LLM abstractions were designed to feel as natural as possible.
factsmachine.ai/p/marvins-turn… @AskMarvinAI 🦾 Marvin has 4 main components:

🧩 AI Models for structuring text into type-safe schemas

🏷️ AI Classifiers for bulletproof classification and routing

🪄 AI Functions for complex business logic and transformations

🤝 AI Applications for interactive use and persistent state

🤖 EVERY real-world AI application needs to convert unstructured text into structured data.

🧩 Today @AskMarvinAI is introducing AI Models to make that process easy.

🧰 AI Models combine the power of LLMs with the type-safe schemas of @pydantic.

github.com/PrefectHQ/marv… @AskMarvinAI @pydantic 🤝 AI Models are drop-in replacements for regular Pydantic models. Their superpower: optionally using AI to process their inputs.

That means they automatically have:
🔒 schemas
🦺 type-safety
✅ and validation

But also gain:
💡 inference
🔎 parsing
📏 and standardization

🚨 DON'T PANIC.

👋 We're open-sourcing @AskMarvinAI to make it easy to build AI-powered software!

🪄 Marvin introduces AI Functions: minimalist functions with no source code that can generate typed outputs with AI.

😛 It's like no-code... in your code.

askmarvin.ai @AskMarvinAI The way we build software is evolving quickly:

✍️ Artisanal: hand-written source code
🧑‍💻 Copilot: AI-assisted source code
🤖 GPT-4: AI-written source code
🪄 AI functions: No source code

AI functions aren't constrained by the limits of what can be expressed in code.

Two years ago, I joined my friend and part-time nemesis @jposhaughnessy on @InfiniteL88ps to discuss frameworks and techniques for navigating an uncertain world and thinking about the future. But do they actually work?

Today, I'm back with receipts.

https://twitter.com/jposhaughnessy/status/1585642205976604672

@jposhaughnessy @InfiniteL88ps First and most important, this is the perfect @PrefectIO cake referenced at

Like beer?

Then you're going to LOVE statistics!

One of the most influential statisticians of all time was, secretly, the Head Brewer of Guinness.

(a lot of beer fans are nodding now)

He published under a pseudonym: "Student."

(a lot of statistics nerds are nodding now) William Gosset joined Guinness in 1899 as its Head Experimental Brewer.

He immediately began one of the first data science programs.

Given limited resources, he was especially interested in ways of extracting confidence from small samples.

1/ On Sunday September 14, 2008, the largest players in the credit derivatives market convened a special trading session in a last-minute effort to offset risk from the seemingly-certain bankruptcy of Lehman Brothers.

I was there, and it was the most memorable day of my career. 2/ The session itself was extraordinarily unusual: trading would last from 2pm-4pm and all trades were contingent on Lehman filing before midnight. Otherwise, the trades evaporated. It was like a very grim version of Cinderella.

1/ The idea that successful investors "have to be right while everyone else is wrong" is a fallacy.

What they really need is to have an opinion different from the market-clearing median opinion -- which may or may not be different from everyone else. 2/ Given: a market is a machine that turns opinions into numbers.

Those opinions are usually distributed in something like a bell curve, with the median of the distribution roughly corresponding to the stock's current price.

1/ Recent market activity -- and the short-vol trade in particular -- provides a fascinating, close-up look at "wrong-way risk."

Wrong-way risks are positive feedback loops that can have tragic effects on a portfolio.

So tonight: wrong-way risk with 🏠 & ⛈️! 2/ Say you own a 🏠 in Florida, and you buy ⛈️ insurance from a company that specializes in natural disasters.

You're all set, right?

Now consider: what if that company, despite -- or possibly because of -- their expertise, became overexposed to Florida properties?

1/ I'm so glad people enjoyed the boat and the bathtub metaphor. I got a few requests to tackle option gamma next, so... let's play ⚾️! 2/ Gamma is a tricky concept that many can recite (2nd derivative of price!) but is tough to illustrate, at least in an intuitive way.