Blog

In college I remember being fascinated by the Clay mathematical institute's problems. One of these is the Navier-Stokes problem, which basically means figuring out whether fluid dynamical…

Coffee & Code: Just shipped a feature that automatically improves the data context layer, as I posted about yesterday.

One of the biggest goals in AI is 'self-improvement'. This is true for the underlying models, but is even more critical when LLMs are used for some real world task.

The new GLM 5.3 flash model is performing well on data analysis tasks.

What do we look for in reviewing professional work?

MCP for X expertise ready and run using an agent of my choice. Getting email updates. If everything works well, my MEPS data in a local SQL instance should be all risk scored, reviewed, and…

We have to be careful of our relationship with the "easy button" that is AI. For our own benefit, we need to be clear on the value added by human expertise vs artificial.

I've been building pieces towards a goal: Can we get AI to do complex knowledge work with a high level of determinism and competence.

Coffee & Code 3: The industry is doing all kinds of acrobatics to build in governance, observability, auditability, and determinism into complex AI workflows. Usually these 4 things require…

Coffee & Code 2: one of my favorite publicly available healthcare datasets is MEPS. It doesn't have risk scores however, so I bolted those onto it. Allows me to do some interesting analyses.

Ever since we accelerated the scientific method and an objective understanding of the universe, we have relegated divine interference to the parts of our perception and experience that…

A few weeks ago I posted about using encoded human expertise to process an astronomical image. It got a good reaction, but the more interesting part happened after I hit publish, when I…

An AI agent ran an IBNR reserve analysis on a healthcare claims dataset. Five thousand three hundred twenty-six claims. Nine accident years. One data file.

An AI agent read a prior authorization policy document. The document was long, about 580 lines. It had an FAQ section, CPT code tables, a policy change log, forwarded emails from a medical…

An AI agent read a due diligence framework. The framework referenced an appendix, so it read that too. The appendix contained a precedents section with worked examples from prior…

I just corrected an AI after it kept insisting a given date was a Wednesday. The correct answer, trivially verified, was Thursday.

The economics of LLMs are nightmarish. And enterprises may be sleep-walking into a trap.

McKinsey's internal AI platform, Lilli, was compromised in a controlled security test by a startup called CodeWall. Their autonomous AI agent found a very old kind of bug in a very modern…

There are now more than 10,000 LLM-based SaaS startups in the United States. The number varies depending on how one defines the category, but the broad point holds. The market has filled…

The traditional flow of using analytics tools with data is already feeling dated.

If you've been using ChatGPT or Claude or Gemini for a while, you've probably noticed they're getting better at remembering you. That's intentional. Memory is one of the biggest features…

Model lock-in is the new vendor lock-in. And it may be worse.

ClaudeCode, codex, opencode, etc. - these tools are about way more than just "code". Or equivalently, we vastly underestimate the versatility of code.

We're drowning in tools. As AI makes it trivial to build software for extremely narrow problems, we risk creating a tool zoo that demands more attention than it saves.

Welcome to the upside-Dodown!

A friend at a party might tell you to buy that hot new stock or IPO. Or the stock market might tank tomorrow and you may want to pull your money out. Or you might be tempted into active…

We're witnessing the rise of a peculiar phenomenon in AI: agents that throw massive amounts of compute at problems that could be solved with a fraction of the tokens. It's like watching…

In the rush to embrace AI tools that promise to compress weeks of work into days, we may be overlooking a crucial side effect: the cognitive toll of hyper-accelerated workflows. While…

As actuaries, we spend our professional lives trying to quantify uncertainty, predict the unpredictable, and model the future. Yet perhaps the most important lesson I've learned in my…

Syed Muzayan Mehmud September 2024

Once in a while you come across some great content on Netflix. The show Connected presented by Latif Nasser is one such example. In particular, the episode Digits had me wondering whether…

There is a branch of mathematics called optimization theory that is behind all sorts of algorithms, including AI. These algorithms search through the space of possible solutions to find one…

Fearing an omniscient AI is a trend. Countries are (allegedly) racing towards developing general AI, and whoever gets to the ring first - will rule them all!

The science (or art) of healthcare claim-based risk adjustment is about three decades old. For the most part, the information used in these models has remained the same: diagnoses and…

This is the summary of a very long blog post,

Lorem ipsum dolor sit amet...

This post is another in my series of posts on healthcare risk adjustment. As I promised in my last article on RADV, this one will be a less tedious read!

This post is another in my series of posts on risk adjustment. My goal in this one is to provide a high-level overview of the Affordable Care Act (ACA) Risk Adjustment Data Validation…

There is a revolution happening in analytics and Artificial Intelligence (AI) is at its center. My goal with this post is to catch you up on the 80-year history of AI by the time I finish…

The Wall Street Journal broke the story on the eve of July 6 that it was expected that the Trump administration would halt "risk adjustment payments" to Affordable Care Act (ACA) plans. The…