I've jotted down common traps that companies face during their AI transformation: qethanm.cc/2026/01/16/t... (Folks on LinkedIn are getting this piece-by-piece. Since you follow me on a site that doesn't discourage external links, you get to read the whole set at once.) #dataBS #aisky

Lines up with a point I raised last year: newsletter.complex-machinery.com/archive/027-...

Lines up with a point I raised last year: newsletter.complex-machinery.com/archive/027-...

I admit, I did not have "genAI-induced psychosis" on my risk bingo card. But I will say: one way to uncover genAI's downside risks is to forcefully cram it into every possible crevice and see what breaks. Which is exactly what's happening.

Reviewing some old issues of Complex Machinery and tempted to rename the newsletter to "I Fucking Told You So"

I get it -- genAI companies don't truly believe in these examples. They are constantly playing a game of Cook Up Barely-Plausible Use Cases So We Have An Excuse To Keep Pushing GenAI. Yes. Yes, and ... ... by this point, they're grasping. It looks desperate. Game over. Go home.

NEW: The UK is playing “whack-a-mole” with big tech regulation. After the Grok backlash, ministers rushed through an important ban on nudification apps, but it may not even apply to Grok, or stop wider AI harms. Me + @catneilan.bsky.social for @observeruk.bsky.social observer.co.uk/news/nationa...

I get it -- genAI companies don't truly believe in these examples. They are constantly playing a game of Cook Up Barely-Plausible Use Cases So We Have An Excuse To Keep Pushing GenAI. Yes. Yes, and ... ... by this point, they're grasping. It looks desperate. Game over. Go home.

It's almost like we need strong, unambiguous laws that prohibit certain uses of data...

In the latest Complex Machinery: As with any gold rush, the genAI space has attracted its share of newcomers. That's created some interesting downside risk exposures ... especially when inexperienced players wind up in key roles. #dataBS #aisky newsletter.complex-machinery.com/archive/053-...

Reputation laundering is a form of data tampering: modifying source records to influence downstream decisions. Such edits to Wikipedia are especially pernicious, given the volume of people and machines which rely on it. (h/t @bondhack.ft.com ) www.thebureauinvestigates.com/stories/2026...

The security/convenience tradeoff strikes again:

What I've been reading this week: - the economic pains and gains of the climate crisis - unintentionally bad data - intentionally bad (poisoned) data - the "warehouse model" applied to food and more! Read online or subscribe: inothernews.complex-machinery.com/archive/010-...

Is September the month of the best 🍔 ? Time for #dataBS to find out...

Is September the month of the best 🍔 ? Time for #dataBS to find out...

OK but AI's impact on excitement and Corporate FOMO™️ has been through the roof. It's almost like there's ... some kind of disconnect? I wonder ...

Tapping the sign again (and again, and again):

Seems a good time to dust this one off again ...

Tapping the sign again (and again, and again):

A painful example of the knock-on effects of bad data: Due to incorrect provider listings, some patients shell out for pricier out-of-network care. (Now, imagine this same incorrect data several levels deep in some automated system/predictive model. Welcome to a Kafkaesque hellscape.)

A painful example of the knock-on effects of bad data: Due to incorrect provider listings, some patients shell out for pricier out-of-network care. (Now, imagine this same incorrect data several levels deep in some automated system/predictive model. Welcome to a Kafkaesque hellscape.)

Seem to be a bunch of new joiners to BlueSky. Here’s my old FinTwit starter pack. go.bsky.app/5m8gZri

if you're new here this is a good bunch go.bsky.app/GhSgfrR

01 November 2024: writing about genAI chatbots, I pose the question "Why would anyone ask a probabilistic bullshit artist for a factual answer?" 09 January 2026:

Thirteen years later, This Is (Still) Fine

ladies, it's ok because only paying subscribers can make/distribute deepfake porn of you and your children now www.ft.com/content/c24d...

So: - please connect your medical records but also “not intended for diagnosis or treatment" and then "HIPAA doesn’t apply in this setting" Seems legit.

Need to put this one on repeat:

So: - please connect your medical records but also “not intended for diagnosis or treatment" and then "HIPAA doesn’t apply in this setting" Seems legit.

It's Wednesday! Here's a selection of what I've read this past week: - tech-enabled Lego bricks - Yann LeCun's FT interview - ummm .... "that" - Hasbro's leadership simulation - Meta's "playbook" for dodging regulators … and more. inothernews.complex-machinery.com/archive/009-...

An entire playbook to dissuade regulator scrutiny around the globe. It's like Meta read about Uber's "greyball" tool and said "oh we can top that." This is a long and solid read. Here's just one of many quotable excerpts:

Working in this field, there's a fun game of "Find The Feature" -- trying to reverse-engineer a search / ML / recsys match. The weirder, the better. I have a hunch about this one ...

Interesting thread on matching and match strategies in games. Lots of factors to consider, and ideas that can apply to other areas. (This is one case where trading is on the simpler end: "I am willing to buy/sell N shares at price $P. Any takers? No? OK I'll change my price and try again.")

An entire playbook to dissuade regulator scrutiny around the globe. It's like Meta read about Uber's "greyball" tool and said "oh we can top that." This is a long and solid read. Here's just one of many quotable excerpts:

The sign says Wegman's "shares" but does not "lease, trade, or otherwise profit from the transfer of biometric identifier information." Such wording may still leave the door open to: 1/ analyzing the data for internal purposes 2/ sharing raw data (no profit) 2/ selling _derived_ data products

Solid thread overall on MetroCard implementation. Highlighting this post in particular because I'd picked up on the value-on-card bit some time ago, and it really stuck with me. It's a systems design consideration that we see elsewhere in tech. Such as "compute-on-device vs central-server."

Last month, I mused that Meta likely built its "superintelligence" group because the incumbent AI team wasn't telling Zuck what he wanted to hear. newsletter.complex-machinery.com/archive/051-... Today, Yan LeCun's statements in the FT hint that I was on the right track.

For the first Complex Machinery of 2026, I'm going through the top pieces from last year: newsletter.complex-machinery.com/archive/052-... #dataBS #aisky

Last month, I mused that Meta likely built its "superintelligence" group because the incumbent AI team wasn't telling Zuck what he wanted to hear. newsletter.complex-machinery.com/archive/051-... Today, Yan LeCun's statements in the FT hint that I was on the right track.

A well-designed game can provide more than amusement. It can double as a form of simulation, a safe space in which to test ideas and work through scenarios. Case in point is this game for Hasbro leadership. A 🧵:

A well-designed game can provide more than amusement. It can double as a form of simulation, a safe space in which to test ideas and work through scenarios. Case in point is this game for Hasbro leadership. A 🧵:

Starting a new thing in 2026, a long running AMA! Ask me a business, tech, or random question and I'll answer them on my personal blog. Happy to give you my take on your work situation, career advice, etc. Bring it on! forms.frankwiles.com/ask-frank

Bonne année et meilleurs vœux!

And three decades after the MetroCard was designed and implemented, account-based AFC systems are increasingly the norm! (More on this distinction at kurtraschke.com/2021/11/loop...)

Solid thread overall on MetroCard implementation. Highlighting this post in particular because I'd picked up on the value-on-card bit some time ago, and it really stuck with me. It's a systems design consideration that we see elsewhere in tech. Such as "compute-on-device vs central-server."

Ages ago, I built a running list of news articles related to data ethics (to keep some training materials fresh). I quickly progressed from: "hmmm will I even find enough articles?" to: Oh wow there's a lot." and finally to: "Why is Uber in so many of these?"

A selection of what I've been reading this past week, including: - AI-generated slop video is $urpri$ingly good business - why people keep falling for narcissistic leaders - do people prefer AI-generated art? Read and/or subscribe here: inothernews.complex-machinery.com/archive/008-...

Running through my 2025 best-of list: the pieces I most enjoyed writing over the past twelve months. If you're in need of some year-end reading, here's a mix of my usual: risk, AI, and a splash of finance history. 🧵:

Running through my 2025 best-of list: the pieces I most enjoyed writing over the past twelve months. If you're in need of some year-end reading, here's a mix of my usual: risk, AI, and a splash of finance history. 🧵: