What you see here is the last week’s worth of links and quips I have shared on LinkedIn, from Monday through Sunday.
For now I’ll post the notes as they appeared on LinkedIn, including hashtags and sentence fragments. Over time I might expand on these thoughts as they land here on my blog.
A few months back, I wrote a short piece on a police department’s accidental data leak and how that is an example of the always/never tradeoff: “The always/never tradeoff in data collection”
When I saw this article, it took me a moment to realize that it was about a different group:
During a press conference, he said the force received a “routine inquiry” freedom of information request, which was seeking to understand the total numbers of officers and staff at all ranks and grade across the organisation. While responding, a police employee embedded the source data, revealing the surname, initial, the rank or grade, the location and the departments for each of our current employees across the service.
(I take issue with the term “breach” here, as that would entail a deliberate infiltration in which someone walked out with the data. This was clearly a “leak” because an internal employee accidentally released data while going about their job.)
The take-away lesson here is that the simple act of collecting data means you risk leaking data.
This would be a good time to double-check your procedures for publishing and exporting data to limit sources of potential incidents. This goes for formal publications of reports, to data products, to raw data dumps.
For bonus points, you can red-team some of your supposedly “clean” data exports to see how they could be misused.
Despite recent tech layoffs, the AI job market still exhibits a lot of demand:
“The $900,000 AI Job Is Here” (WSJ)
Recruiters say they see pay edging up because the available supply of AI practitioners is falling short of demand, particularly for midlevel and higher-level positions.
(The title is a bit clickbait but the underlying message – that companies are competing with each other over AI hires – holds.)
Can we … slow down here? Just a bit?
I’m all for AI being the hot job of the moment! I consult in this space, after all. But I sense this hiring rush is mostly Corporate FOMO™. Companies are hanging their hopes on AI being a game-changer for their business, even though they can’t articulate specifically how it’s going to help them. They’re also unable to temper their optimism with an understanding of how their AI plans may go awry.
My thoughts for the AI practitioners? Good on you! Go make that cash. Be choosy when deciding between roles. Just make sure you don’t fool yourself into believing that this will be a “normal” job market forever.
As for the hiring managers out there: Before you rush to hire that AI specialist, do you have specific, actionable projects for them to take on? Can you draw a clear line from the role to improved revenue or reduced risk? If so, great! Best of luck in the hiring fray.
(And if you need help sorting out those projects and use cases, hit me up: https://qethanm.cc/contact/ )
Hiring in the tech space is weird. Even with the recent FAANG (and FAANG-adjacent) layoffs, the job market is still kind of hot. Which is why this article caught my eye:
“Want to work in tech? Don’t work in tech.” (Insider)
This is a solid read for tech practitioners. I also expect a number of hiring managers are excitedly passing this article around their networks, hopeful to land some ex-FAANG talent. Fair enough!
You might want to keep this in mind:
- The tech layoffs don’t mean that you’ll automatically snap up those hires! You still face competition from other companies.
- Your job posting is also competing against “taking time off,” “launching a startup,” and “going back to school.” Not all of the people who have been laid off are planning for a standard FTE role.
- Further, that job posting should be in competition with “your company doesn’t hire anyone right now.” This would be a great time to review your existing tech teams and the work they’re doing. How, specifically, will additional hires help them? Do you have short-term needs that are better met by contractors (for project work) or consultants (for guidance)?
- Most of all, remember that there are other people applying for your open roles. People whose resumes don’t list well-known companies. They are smart, talented, and more than capable. You don’t want to hold their lack of FAANG status against them.
A year ago today, I published an O’Reilly Radar piece on the then-new Disney+ ad-supported tier: “Ad Networks and Content Marketing.”
Today, I’m sharing my thoughts on the recent price increases of the full-fare, ad-free subscription: “Striking a balance.”
The gist: subscriptions are interesting animals! Especially when you mix ad-free and ad-supported plans. It’s similar to mixing fixed-income and (higher-risk) equities in finance. With the same concerns on how to balance the two.
Every AI model will be wrong now and then. Even if you’ve hired the smartest data team and paid for the best tools, you won’t produce a model that is 100% correct 100% of the time.
Is this a problem? If the model only performs low-stakes work, you’re probably fine! But what if that model’s error leads to a vehicle crash? Or an incorrect injection of medicine? Or a wrongful arrest? You’re probably looking at a lawsuit. And do you really think “because the AI told me so” will hold up in court?
My best answer to that question is: “try to not wind up in court in the first place.”
(This is especially important now, as there’s so little case law around AI-related harms.)
To avoid court cases, you need to manage your AI risk.
That starts early in your AI project planning process: you want to map out the ways your model could be wrong and how you will handle those cases. Consider what the model is meant to do and who is affected by its predictions. The greater the impact, the less leeway you should give the model to decide.
Why am I thinking about this today? Frankly, I think about this every day. But between the recent mistaken arrest due to a facial recognition failure and more students being falsely accused of cheating, well, I figured I should bring it up again.
Fashion house Ganni developed a generative AI bot to speak on the brand’s behalf:
“Ganni’s Copenhagen show put AI on display — but not in its collection” (Vogue Business)
What I find interesting is how they trained the bot, such that this virtual brand ambassador would “speak” with the company’s voice:
Ganni’s AI was trained on data from the brand as well as online interviews with and social media comments from its community of fans and influencers, known affectionately as the “Ganni Girls”. Guests could then interact with the AI during the show, asking any question — speaking into microphones rigged to living trees dotted around the venue — and receiving a response tailored to Ganni’s view of the world.
Generative AI systems have a (well-earned) reputation for saying odd, untrue, or even downright offensive things. So I’d be a little concerned about granting the bot free reign to talk to attendees this way.
If you look closely, though, Ganni took steps to limit the risk of the bot going awry:
- their team curated the training data to reflect the company’s values (thereby reducing the chances that it would say something to the contrary)
- much of that data came from its community (so it should speak in a familiar voice, and on topics that matter to Ganni community members)
- the bot was only available at its event (not on the wider internet)
- we can expect event attendees were all Ganni fans (which means they’re less likely to try to intentionally break the bot)
(Does your company plan to launch a generative AI bot? Check out my piece “Risk Management for AI Chatbots” for details on challenges you’ll face and how to address them.)
OpenAI, the company behind ChatGPT, is finally giving website hosts a way to keep their content out of the bot’s training data:
“How to Block OpenAI ChatGPT From Using Your Website Content” (Search Engine Journal)
On the one hand: This is great. It’s good to see OpenAI actually giving people a choice in the matter.
On the other hand:
- The technique described in the article only stops the OpenAI crawler. Other crawlers are still free to ignore that line in robots.txt. (And if you run a website, you know that plenty of bots do just that.)
- This doesn’t remove content from OpenAI’s existing models. (They’ve been cagey as to where they got that data, anyway… hence the lawsuits.)
- This is another example of a company putting the onus on everyone else to stop them, instead of asking permission in the first place. (Opt-out is weak. Opt-in is the way to go.)
Sum total: OpenAI’s move strikes me as “too little, too late.”