🎶 On the 12th day of Christmas, @Google #Antigravity gave to me … 🎶 more model quotas, again. ☹️ Even though I am just using the low-cost 'flash' model, I run out easily. Hey, @Google , this isn't working.
More:
www.linkedin.com/feed/update/...
🤣 "There is a saying that failure is an orphan. But remember: orphans are not necessarily infertile." 🤣
On the fist day of the year, The Economist made me laugh out loud. Thanks, guys.
Happy New Year everyone.
www.economist.com/business/202...
They. Did. What!?
"Lloyds used data from 30,000 staff accounts in union pay talks. Bank's customer insights team compared
financial resilience of lowest-paid staff to
customers as part of salary negotiations."
Contract of employment demands employees have a Lloyds bank account.
This is absurd. And very cool.
A web server with no application logic. Just an LLM with three tools.
github.com/samrolken/no...
Forms actually submitted correctly. Data persisted across restarts. The UI was usable. APIs returned valid JSON.
Friday #thought for the weekend:
"Modelling the world mathematically is as uniquely a human trait as language or art, but you would rarely find anyone complaining of being 'reduced to a poem' or 'reduced to a painting'."
--- Douglas W. Hubbard: How to Measure Anything
🧵 3/3
What steps are you taking to move the #AIGovernance conversation beyond ethics to include the more challenging topic of deep model assurance?
www.linkedin.com/feed/update/...
🧵 2/3
The core issue is #governance. When we deploy #AI systems that are unpredictable by design, we move from managing data risk to managing systemic operational uncertainty. This requires a different, more robust approach to assurance than simply monitoring for output bias. #TechStrategy
A recent survey of #AI advancements provides a useful and broad overview of the field. Two points I took away raises critical questions for corporate governance and #RiskManagement that seem to be understated in the mainstream conversation.
www.linkedin.com/feed/update/...
🧵 1/3
🧵 4/4:
I explore this in my latest article to consider why your data product must become an MCP Server and what that means for your skills and strategy.
www.linkedin.com/pulse/you-bu...
#ModelContextProtocol #DigitalTransformation #AI
🧵 3/4:
The Model Context Protocol (#MCP) is key to making your data products—currently dashboards, models, and more—discoverable and usable by AI orchestration engines. Without it, your information architecture risks becoming rapidly obsolete.
🧵 2/4:
While we're diligently building robust data platforms and dashboards, a seismic shift is underway: enterprise users increasingly demand direct answers, not just static displays, powered by #ConversationalAI.
This means adopting an #AIFirst paradigm for your entire #DataArchitecture.
Are you building yesterday’s information architecture tomorrow? 🏗️
It's a question worth asking as the pace of AI adoption transforms how we interact with data.
www.linkedin.com/pulse/you-bu...
🧵 1/4
Just wrapped up #EARL2025! Fascinating insights on R/Python convergence, pragmatic GenAI, data for good, & the 'last mile' of data science. All driving towards better decisions. Great connections & discussions! How do you stay current? #DataScience #AI
www.linkedin.com/feed/update/...
🧵3/3 But they are making it harder to love them. In any case, a multi-cloud strategy has long been the recommended architecture, despite the obvious overheads and drawbacks. Scandalous that several banks are down.
Good luck to all the ops teams.
🧵2/ brought everything down in (from memory) 2023.
I have been an AWS user for longer than anyone else who is still using it. Huge advocate. I even used to host the Frequently Asked Questions for it, back in the day when AWS was just the advertising API.
web.archive.org/web/20070127...
Some #AWS services #failed. That's unfortunate. What is catastrophic is that the us-east-1 region is STILL a single point of failure for the entire global infrastructure. That isn't good enough. It should have been fixed well before the LAST us-east-1 failure 🧵 1/3
Tools have always shaped how we think. Socrates worried about children reading too much, not being able to remember things anymore and not learning the right (Socratic) way because they couldn't have a dialogue.
An example of unusual data gathering - Compromise! To listen to the weekly Top 40 from the past 20 years would take 96 days and nights Just the No.1 singles would take 17 hours Listening until the first chorus should only take 5 hours So we listened… and got our chart
I have a slide about an unusual data request I once got: can we show that song choruses are appearing earlier due to the economics of streaming (gotta hook people in early to get paid)? I joked that it's not as if anyone's got a database of song structures. But it turns out that someone has...
I wrote a thing about substituting LLMs for survey respondents
notstatschat.rbind.io/2025/08/15/i...
AI generated image of a hallucinating AI.
3/3
👉 Ask not if the #AI is perfect, ask if it is an improvement on what we do now and if you can control the errors, omissions, and mistakes that are inevitable in any decision process, whether made by human or algorithm. 👈
AI generated image of an hallucinating AI.
2/3
Our monkey brains are rubbish at any sort of complex rational thought. And yet we manage to make do and make progress.
Are there any of my readers who have not regretted a decision they made in error at the time because of bias of failure to think it through?
2/3
AI generated image of a hallucinating AI.
Hallucinations and related reasoning errors are a real problem with #AI and something you have to consider in your governance and controls.
HOWEVER, we have lived with 'hallucinations' in humans for millennia, only we call them cognitive and decision biases. Our monkey brains are rubbish … 1/3
2/2
The data scientist standing beside him whispers, “...Now ask it to show you the most valuable customers you're about to lose...”
The senior data scientist adds, “...and what you can do to retain them.”
#PowerBI #Copilot #JustSaying
www.linkedin.com/feed/update/...
AI generated image of a Data Craftsman
An executive proudly shows off their new AI dashboard. “Watch” he says, “AI, show me my most valuable customers.” The screen instantly fills with a list of the highest-spending clients.
The data scientist standing beside him whispers… 1/2
www.linkedin.com/feed/update/...
Kids: it isn't "Big Data" unless you *have* to throw >99.9% of it away.
Matt Strassler has a good overview here: profmattstrassler.com/2024/10/21/i...
2/2
When I started working with commercial organizations, I never understood what all their fuzz and angst about 'Big Data' was all about. Physics, which is my background, now handles 1.5 billion complex megabyte events per second. This article takes me back: spectrum.ieee.org/lhc-radiatio...
1/2
> "To beat the working-from-home alternative, spaces should ‘earn the commute,’ and you do that by creating space where people actually want to work, not just the space where they have to work."
#quote #RTO #WFH #true
Your thoughts?
Source: www.mckinsey.com/industries/r...
An AI generated illustration for the article, based on the idea of an insurer forging their own protection in a complex regulatory environment.
The European Insurance and Occupational Pensions Authority (EIOPA) recently published its Opinion on AI Governance and Risk Management. It’s being welcomed by many as a helpful clarification: no new rules, just guidance.
I am not so sure. Here's why:
www.linkedin.com/feed/update/...
It's the weekend. You need cats 🐈⬛ You need geekery 🧑🏻💻 HTTP Cat has you covered with cats for every status code
http.cat
You're welcome.
AI generated image for the concept of a "Data Craftsman".
Businesses are drowning in data but starving for insights that transform their bottom line. Why? Because we treat data science like a production line, not a craft. Testing does not trump insights.
I've written a piece that explores this:
www.linkedin.com/feed/update/...