If Santa Isn't Real, What Is? Inauthenticity in an AI Age

By | December 12, 2025
0 Comment

As AI generated content gets ‘better’ — in the sense of feeling, appearing realistic — does that mean we will more readily accept it? Or will we more readily dismiss it — and all content that might appear to be generated by AI?

What happens when we start to suspect that everything is AI generated?

And are we already at that point?

I’m increasingly sensing that even quality YouTube commentaries (usually TV or movie criticism) sound to my ear to be at least partially written by AI.

I’m a journalist, not a linguistician or whatever they’re called. But I can recognise a sentence structure pattern when I see one, and I think I can increasingly see such patterns in AI. Take the emphatic contrast or contrastive emphasis structures, for example:

“It’s not about X, it’s about Y”

“Not only X but also Y”

“Instead of [x], [y]”

“This wasn’t [x]; it was [y]”

“This wasn’t just [x], it was [y]”

On the surface this looks and sounds good. We’re clarifying what it isn’t — or isn’t merely — and asserting clearly what it is. But when you realise it crops up in a lot of stuff you know is AI-generated, you start seeing it everywhere. And by then you can’t unhear/read it, nor get away from the sneaking suspicion you’re having AI smoke blown up your hind quarters.

So what’s wrong with this? Surely if an AI is that good, it’s probably helpful content, right? Worth taking seriously?

Well, no and no. And while I’m not sure it’s an exact match, I would argue that this takes us into the world of source credibility bias, where we tend to judge the quality of the information based on who the source is, rather than on the information itself. This is usually perceived negatively since it overlaps with a better known bias — confirmation bias, where give more weight to sources that match our own beliefs, or affinity bias (preferring information from sources perceived as part of our own tribe), or the halo effect, where we tend to have an exaggeratedly positive (or horn, negative) view of a source of information, making them more (or less) credible than they deserve.

But here we’re talking about something slightly different: we are judging the information based on our estimation about whether the source is in whole or part AI-generated. And what is most disconcerting about this is that it’s a phenomenon has already become a central feature of our lives.

Bias Bias (it’s a thing)

Let’s walk through this.

We all have a mix of the above biases, but that mix changes as we change. As an aspiring (and failed) academic, I once tended to look down on journalists, thinking they only reported the first draft of history, seeing only a part of the elephant, and that it was up to historians like me to put it all in context and give it meaning that would endure. An affinity bias, I suppose.

When I became a journalist, I saw it the other way around. When I interviewed academics, I found much of their analysis wanting, lacking the touch that could only come from actually being there, witnessing change. I felt they either underestimated change because they didn’t see that things could change in an instant (the old Hemingway quote about how bankruptcy happens that equally applies to changes in power: gradually, then suddenly) or they exaggerated change because they saw something that wasn’t really there.

Now I’m a bit of both (wannabe historian and semi-retired journalist), I can see that expertise comes in lots of different flavours, and judging an expert by their age, looks, qualifications and the number of followers they have is not the way to go.

As a Spurs fan, for example, I rely on a broad swathe of people to tell me what’s really going on, and as far as I can work out none of them has a real job and none has been a footballer beyond the Sunday kickabout, and they all seem to work out of their parents’ spare rooms. I don’t care. They explain things well and they make sense — often more than the high-paid pundits on TV.

Sausages and dodgy merch

Oddly, being able to see them in their bedroom with dodgy lighting merch in the background makes them more credible to me than the suits in TV studios. I would argue this is a sort of transparency bias — I tend to believe them more because I can see how the sausage is made.

The same with politics, with movies and TV, with quirky subjects journos and academics wouldn’t touch with bargepoles. I choose them because over time they prove that they think deeply, work within their experience and knowledge, and explain themselves well. In shorthand, I find them credible because they’re lived-in, human.

And that last bit is the problem.

If I get the faintest whiff that what they talking about has been generated by AI, I’m outta there. To me any AI involvement in the thought-to-content process taints the result. I don’t mind a bit of AI research, as long it’s been checked. What I do mind is something that might have been constructed by AI, or partly by AI. To me that is unacceptable.

Why?

I’m not sure. I’ve been trying to figure that out. I think it has something to do with the Weltanschauung — world view — of the creator. I need to know that the ideas I’m hearing are coming from something that is not synthetic. Sure, we can run ideas past AI, I suppose — I got help from it up there because I couldn’t recall the origin of the bankruptcy quote (I’m embarrassed to say I had no idea it was from Ernest Hemingway). To me that’s more or less OK, because I checked it elsewhere, though it’s still a mark against me because I’m trying to show I’m better-read than I really am. You would be right in thinking less of me because I didn’t have the reference in my head, and was surprised it was Hemingway’s.

What I can’t accept is that an analysis, say, of Three Days of the Condor, or a Pluribus episode, is partially or wholly composed by AI. To me that’s like saying: this commentary is derived at least in part from previous content by a machine, excluding any personality, any conscious or unconscious reflection upon the past by the author, on their experience, on what they might have dreamed last night, on the state of their heart, mind, stomach.

It’s not who we are, it’s how we got here

This is all synthetic, in short. And we humans are not synthetic. We’re a bubbling pool of neurons and soul, heart and head, scars and serenity, our every thought and feeling rooted in our experience, whether it’s through books, TV, love affairs or trauma. We’re all a big mess inside, and that’s what makes every sentence we write so interesting. And so unique. Whoever we are.

(The Psychology of Robots and Artificial Intelligence, a 2025 paper, pointed to research that argued people generally perceive AI-generated artwork to be of lower quality because it lacks the emotional expression and uniqueness that connects them to the artist’s mind, rendering the art as inauthentic and incapable of reflecting true experience. I’d argue that’s also exactly true of any AI-generated content that is not already formulaic — a stock market report, football scores, the weather, most of which has been automated for at least the past 20 years already.)

And so yes, even if a sentence, a phrase, a single insight has been generated by AI, I would hit the purist button and say: if any part of this is synthetic, then the whole is violated and invalid. Because we now cannot tell what is real and what isn’t. We can’t tell whether the thought process that went into the piece took a synthetic turn somewhere, and so we have to discard it all.

That might sound a bit extreme. And I’m probably being a hyprocrite here. Perhaps we should be laying down some rules: It’s OK to check your ideas with the AI. It’s OK to start with an AI as long as you pick up early enough and take over. It’s OK to have AI correct and improve your grammar, because we do that all the time with Clippy and co, right?

No, I don’t think it is. I hate formulaic sentences, and I really hate it when I feel an email or message sent to me is pro-forma. Even more, bizarrely, if it’s written in a faux-friendly style. I’d reach for my gun, if I had one.

Coke: the Real Thing, except the ads

And no, it’s not simply a question of adding a little “AI helped in this.” If it did, all you’re telling the audience is that they have reason to suspect the whole thing, unless indicated otherwise. If the content is not authentically you, then why am I bothering investing time in watching/reading/listening to your content? Are these your ideas, or AIs? Where did this idea start, and who constructed the argument? We take a dim view of such admissions because we commit our time and attention to not just the content but the person/people behind it. If some of this content is actually artificial, it undermines that implicit contract. Why should we invest in someone whose worldview is at least in part derived from a machine?

As larger organisations seek to cut costs, they’re inevitably going to turn to AI. Look at Coca-Cola’s trainwreck of a Christmas ad: Coca-Cola AI Holiday Ad Glitches Highlight Generative AI Shortcomings.

Already there is a cottage industry in AI debunkers — people who study the content closely and can highlight where the AI glitches are. (This one, by Dino Burbidge, is excellent: The truck is different in every shot of Coca-Cola’s AI Christmas ad. Surely that matters? | The Drum) In this case, Coca-Cola, thought they’d get in front of it by saying the ad was generated by AI. But they still got a hiding, and so they should. (They might have read an academic paper called The transparency dilemma: How AI disclosure erodes trust before they embarked on their quest.)

Those who might argue that most video is already CGI have a point, but it misses the mark. CGI is the backdrop, the framing, but the actor — the human — is real. (I’m not talking cartoons and stuff here, of course). We expect the actor to act, to bring their best to the scene, even if all the see is green screen and someone in a green suit suspending them as if in flight. We want Robert Downey to put in a performance, and we don’t expect his face to be digital even if the rest of him is. But when Coca-Cola populates its AI ad with fake people, fake expressions, where everything you see is fake but trying to appear real then the spell, the suspension of disbelief, is broken, because, simply, we know.

Technically speaking we’re now in the AI equivalent of robots’ uncanny valley – where as robots become more human-like, our affinity abruptly collapse into revulsion. Research suggests that something similar happens with AI-generated text. AI content might not trigger the gag reflex as much as robots do, but we don’t like being fooled, and it’s this that I think will be the source of the largest pushback against use of AI in anything remotely creative, not purely in artistic terms but in any kind of content that draws on, or pretends to draw on, the creator’s experience, knowledge, qualifications, training, personality and background.

I’m not saying AI is a bubble, but I’d be willing to put money on the labelling of content, products, services etc as “100% Human Generated, No AI” appearing everywhere, and companies that seem to be relying on AI on external-facing stuff to be punished quite severely in the marketplace.

If I don’t post before, have an AI-free Christmas and a slop-less New Year.

(No sentences or ideas in this piece were generated by AI. AI was used in searching for some references and sources)

A Directory of RSS Readers for the Mac

By | September 23, 2025
1 Comment

It’s been a long time since I did Directories of apps here. Indeed, the last one on RSS readers was this one: A Directory of The Best RSS readers which was started in 2004: 21 years ago. I’ll try to get round to updating that at some point. I feel the time has come to reclaim blogs and directories that are human curated and aren’t pushing anything except more satisfying computer usage.

So here’s a start. I’ll update this when I find new stuff. Please let me know if I’m missing something, or getting something wrong.

lire – for iOS, iPadOS and macOS: You only need to buy two apps to cover all three platforms.
lire for macOS if available on the Mac App Store. lire for iOS and iPadOS is available as a universal purchase on the App Stores of the respective devices. Universal purchase allows you to purchase the app once, and then access it on both platforms.

Supports: Feedly, Feedbin, Feed Wrangler, FeedHQ, The Old Reader, Inoreader, Newsblur, BazQux Reader, FreshRSS, Miniflux, Nextcloud News, and Tiny Tiny RSS is currently included.

I like it. Elegant and simple.

Reeder been around for 15 years, and still looking good. Reeder Classic is the one most of us are familiar with. The new version, 5, is just called Reeder and doesn’t support syncing with RSS services. Both are available for Mac and iOS.

While I like the idea that I think the new Reeder is trying to tackle — RSS bloat, where you have way more than you can ever read — I don’t think the answer is to prevent syncing with third party services. Which is why I’m still using Classic.

Leaf – another nice looker. Syncs with Feedly, NewsBlur, Feedbin and Feed Wrangler. $10

Unread – my current favourite, at least for the Mac. (There is an iOS version as well) I’m still not quite sure about the £5 subscription though the free version delivers pretty much all you need.

DEVONthink’s new table of contents features

By | September 18, 2025
1 Comment

DEVONthink, my favourite database since the demise of Evernote, as added some features to version 4 that may not appear that significant, but which I believe offer a real boost to those of us trying to stay organised. They involve tweaking PDF and Markdown files to make them easier to navigate and to find stuff within them.

PDF

A well-crafted PDF includes a table of contents that allows you to jump around inside the document easily from chapter to chapter. The table of contents itself may not be visible in the document, but serve as a kind of outline, visible in PDF readers in whatever sidebar the software offers. Each heading, or chapter, will sit in a tree hierarchy, allowing you to jump to where you want to in a long document.

EXample of good table of contents in PDF@2x.

A table of contents in DEVONthink

Sadly quite a lot of PDFs don’t have these. DEVONthink fixes this, at least in the app itself, simply by right-clicking on the place in a document you want to add to the table of contents and choosing Add to… Table of Contents.

Table of contents add to mouse function@2x.

The ‘Add To’ popup

These markers will appear in the PDF if opened in a different app, although I found it wasn’t always the case if there was no table of contents baked into the original document.

Table of contents in PDF Preview@2x.

Table of contents in Preview

Markdown

I’m a big fan of Markdown, as you know, because it is simple, logical, is used by most good editor apps and means every document you write can be opened in any app and look good. No locking into a particular format. DEVONthink does a great job of supporting Markdown and they’ve added some useful features in version 4.

The two most useful ones are these:

You can use the same Table of Contents sidebar to view sections in a Markdown document — essentially the same outline function for PDFS, and one you might be familiar with from Microsoft Word. Now in DEVONthink you can move sections around a Markdown document via that pane (usually called an Inspector. You can do the same to rich text (RTF) documents as well.

You can now convert a Markdown document to PDF in DEVONthink, preserving your headings in the usual tree hierarchy format. This feature is, I suspect, not included in most cases where you choose to export a document to PDF, and will only work in DEVONthink if you use the ‘Convert to…’ function.

Taming the beast

I find these functions extremely helpful in taming the data beast. Another feature that isn’t new but relevant is creating a list of selected documents — what DEVONthink rather confusingly call a Table of Contents — in either Markdown or RTF. Each entry is a link to the original document — I find it useful to create a sort of index to all the documents in a particular set or folder, speeding up searching for a document by title.

Table of contents markdown@2x.

A table of contents of selected files in DEVONthink.

There’s a lot else in DEVONthink 4 to shout about but these are features that are hugely welcome, and used properly can save oodles of time.

Products mentioned

You can find a list of the changes here: DEVONtechnologies | Discovering DEVONthink 4: Document Editing

  • DEVONthink for the Mac costs $100 for the standard edition, and $200 for the Pro version. The licence is valid for two Macs. A stripped-down version called DEVONthink To Go is available for iOS.
  • Here’s an overview of Markdown: Markdown Guide.
  • A list of apps that support Markdown: Tools | Markdown Guide
  • Evernote

Readers’ Revenge

By | September 18, 2025
0 Comment

tl;dr: How to control your Substack overload and reclaim your inbox

These are difficult times, and staying up to speed is a full-time endeavour. Ironically, the explosion of newsletters from individuals has made this harder rather than easier,

I have found myself swamped by newsletters, despite efforts to bring some order to them (more of those efforts in a later post.) Newsletters can be a useful way of getting information — the email lands in your inbox, right where you are, you don’t need to go out and find it, and, at least in theory, the person writing the newsletter has taken time and effort to deliver something useful to you and at speed.

But it doesn’t scale for either of us. The person writing the newsletter can’t outsource the writing to someone else, except for a few guest columns from time to time. Unless they’re already rich they’re always going to be wanting to convince you to convert to a paid option, And if you’re going to pay $5 a month, why not subscribe to a full newspaper?

But for us the problem isn’t just financial. It’s that there are just so many good (often very good) newsletters that even if we only take the free stuff, we’re still committing ourselves to hours of reading per day or a quickly bloated inbox of unread messages.

The solution: an old fix

For me the solution has been to dust off an old technology: RSS. I first wrote about RSS in 2001, which gives you some idea of how old it is, how old I am, and how much a failure it has been. The story of RSS’ failure is for another time, but the remarkable thing is not only that it’s not dead, but that a) there are some beautiful apps and services still out there for you and b) the newsletter industry does support (if that’s the right word) the RSS protocol, even if reluctantly. The simple truth is that RSS powered everything we now hold dear: twitter/x, facebook, podcasts and yes, newsletters. It’s the guts of the commercial web, and the people who developed it haven’t made a cent from it.

RSS simply takes content, wraps it up and creates a URL. If you save that URL to an RSS reader (for example, it could be a browser like Vivaldi) and the reader will receive all future content sent via that URL. No sign-up, no personal data given. RSS was the future, until more commercially minded souls used it for their plumbing, but removed features that didn’t suit their goals.

Most visible in this are substack-like platforms. Sure, they’ve helped create an industry of smart people who can make a modest living from writing for an audience. Most important is that modest living bit. RSS never really offered a seamless way for writers to monetise their work, and that was a major drawback the likes of Substack have solved. But they solved it by leveraging an even older technology — email — that RSS was trying to get us away from. Back in the late 1990s it was clear that email was vulnerable to spam and malware. (The first phishing emails were sighted in about 2001, but email had already become a favoured vector for trojans and viruses.)

But the bigger problem was that mailboxes were getting bloated, as personal emails, work emails, spam and newsletters all clogged up the works. RSS promised to take the newsletters out of the mailbox by creating a channel for the rising world of blogs to reach their readers, respecting their privacy and their sanity. It worked well for nearly a decade, until social media usurped blogs by making interactions between users the key selling point, rather than the content itself.

Simple, almost

So, where does this go? When my efforts to get my email inbox to zero failed, I decided to quit. Instead I’ve spent the past day or so adding all the newsletters I subscribe to via Substack, Medium, Ghost etc, adding their RSS feed to my reader, and then unsubscribing from the newsletters (unless they’re paid ones.) It’s still a work in progress, but I think it might be the only way to cope with the scaling issues of Substack et al.

It’s better with RSS

Why is RSS better than email? Lots of ways:

  • first, privacy. You don’t have to give any of your details to anyone — the platform, the company/individual producing the newsletter, or anyone in between. There’s no tracking, spam, data sharing, and it’s fully autonomous;
  • you can organise your newsletter subscriptions as you want, within folders or tags supported by whatever RSS reader you end up using. You can do this in your email app, but it’s not as intuitive;
  • things don’t get all mixed up. I’ve talked before about email bloat, and the pain of missing important emails. Keeping your information sources and your email separate is a real plus;
  • space junk: RSS feeds are stored in the cloud or your computer, but not at your cost. Nowadays email is not free, unless you religiously delete stuff.
  • The format of content RSS feeds is more bare bones, but there are some RSS readers that use this simplicity to create very elegant interfaces. All the annoying ephemera added to most email newsletters is stripped, leaving only text and images in – usually – a readable and pleasant format. My favourite is Unread, available for macOS and iOS, which is free, but has a premium subscription of $5 per month, which covers Mac, iPhone and iPad. Another app which focuses on elegance; Reeder.
  • getting out is easy. Substack and others have made it progressively harder to unsubscribe from a newsletter, and don’t get me started on trying to cancel a paid subscription. More on this below. In an RSS reader, it’s easy.
  • you can easily navigate through past posts from one particular newsletter. Doing the same on the likes of Substack can only be done by a cumbersome keyword search in your email app, on the platform’s app or on the web.
  • Slightly nerdy, this one, but RSS readers tend to make it much easier to save stuff to somewhere else. Yes, you can always print an email to PDF, but all the bells and whistles in the email tend to male it a clunky experience. Exporting a post from an RSS reader tends to be more straightforward, and the result more elegant and simple.

Downsides? A few

Disadvantages? Sure, there are some:

  • If you don’t get that much email, and don’t subscribe to many newsletters, it makes sense to keep everything in the same place;
  • RSS won’t (usually) work with paid subscriptions. Better to keep that in email and the app;
  • RSS feeds of newsletters aren’t always identical. I’ve noticed some stuff doesn’t make it through to the RSS feed, but this doesn’t happen much;
  • you will miss some features that Substack and others are adding to their platforms: hangouts, notes, that kind of thing. But I’ve not used these much, except for paid subscriptions.
  • we commonly monitor email more readily than other apps so we’ll likely see newsletters as they land because we have our notifications set that way. Which is good — except when you subscribe to a lot of them, or you want to save alerts for the scary email from the boss or HR;
  • it is a bit more complex to set up. I’ll explain this in more detail in another post, but I do recognise that not everyone is interested in making the extra steps to make this work;
  • with RSS the creator of the newsletter doesn’t get the same benefit of metrics to see who is subscribing, and who is ‘engaging with’ (what we used to call ‘reading’) their content. As a creator using all platforms I definitely think this is helpful, but not everyone is going to use RSS, and so I think on balance writers will get enough of a sense of how they’re being received from email subscribers not to be adversely affected.
  • you might find yourself moving one bottleneck from one place to another rather than removing it. (I was wrestling with much the same problem nearly 20 years ago: Email Wins Over RSS? and What’s RSS to You?) But as RSS makes it really easy to unsubscribe, so thinning the herd every so often is not too painful.

Still, it’s been a good start and for my sanity it was probably overdue. The world is spinning very fast, and the last thing we need is to get overwhelmed with the information sources we actually trust.

Bewildering and Discombobulating

By | June 4, 2025
0 Comment

This is the first of a series of pieces based on Mary Meeker’s recent deck about AI.

It can be bewildering and discombobulating to try to absorb the rapid rise of AI, and I can understand why many of us choose to ignore it, dismiss it as horribly overhyped, or throw up our hands in despair. All of these reactions have some justification. But I think it’s worth taking a step back and trying to place what is happening in some context. Doing so might help in accommodating what is happening, even to draw some benefit and comfort from it. A 340-slide deck may not sound like a good way to do this, but I’ll try to condense it. I hope it will be worth it.

The deck is from Mary Meeker, a veteran of Silicon Valley and someone who, over the years, has got a lot of things right. She’s also at an age, not unlike myself, to have witnessed the miracle of Internet-based technologies and so can see a bit more clearly than those who grew up with the Internet (essentially anyone younger than 45). A slide from a Mary Meeker deck, therefore, is usually worth 100 slides from most other folk, so her recent dump on AI is worth the time. (With some caveats I’ll leave to a later post)

One of the peculiarities about AI is that while it threatens the livelihood of millions, it’s also one of, if not the fastest adopted technology/ies in history. It took 33 years for the internet to reach 90% of users outside North America; it has taken ChatGPT three. It took Netflix more than 10 years to reach 100 million users; it took ChatGPT less than 3 months.

But this is not where the real change will come from. Let’s face it, ChatGPT is easy to adopt because it’s quasi-human. We interact with it in the same way we communicate with our friends. We haven’t adopted generative AI as a technology so much as allowed its anthropomorphic version to insert itself into our lives. This is unsurprising: we have known since the 1970s that we humans tend to accommodate anything, live or dead, into our lives if it hits certain (but not all) anthropomorphic notes. A cute animal (behaviour, features), a favourite teddy-bear (inertia that we take to be stoicism and loyalty), Alexa (voice, responsiveness). A sign our adoption is complete is that we then yell at it when it doesn’t do what we want.

So our embrace of the technology is not really where change is coming from. The change is in how fast company CEOs are adopting AI — and want to be seen to be adopting AI. In late 2023 the proportion of S&P 500 firms mentioning AI during quarterly earnings calls was about 10%, a number that had risen slowly from zero since 2015. By 2025 that proportion had risen to 50%. (Slide 68)

And this is not just CEOs barking whatever buzzwords their media and IR teams are throwing them. In a survey of CMOs by Morgan Stanley in December 2024 two thirds said their companies were running initial tests and/or exploring using Generative AI for marketing activities. (Slide 70)

Agent AI

So where are those tests taking them? The chatbot image of generative AI is not what is really getting CEOs excited. What is getting them excited is the next wave of AI: agents. Agents form a “new class of AI… less assistant, more service provider”, in Meeker’s words. Where chatbots operate “in a reactive, limited frame”, agents

are intelligent long-running processes that can reason, act, and complete multi-step tasks on a user’s behalf. They don’t just answer questions –they execute: booking meetings, submitting reports, logging into tools, or orchestrating workflows across platforms, often using natural language as their command layer.

Meeker compares this shift to that of the early 2000s, which

saw static websites give way to dynamic web applications –where tools like Gmail and Google Maps transformed the internet from a collection of pages into a set of utilities – AI agents are turning conversational interfaces into functional infrastructure.

The key thing to understand here is that an agent is not “responding” so much as “accomplishing”. They don’t need much guidance — indeed, they may quickly need no guidance, but instead autonomously execute, in Meeker’s words, reshaping “how users interact with digital systems “from customer support and onboarding to research, scheduling, and internal operations.” This is where the bulk of the enterprise appetite is going, not just experimenting but “investing in frameworks and building ecosystems around autonomous execution. What was once a messaging interface is becoming an action layer.” (All quotes are from Slide 89)

Strip away the glitter here and it’s this: An agent is essentially a human in disguise (or multiple humans.) Once briefed, it works independently, executing, learning, improving and extending. And companies are investing in the infrastructure to support this autonomous activity. You don’t have to be paranoid to see how agents, barely acknowledged a year ago, are now the focus of significant investment, which would likely have been directed towards investment in human-led processes.

Meeker cites a handful of examples: Salesforce’s Agentforce not only handles customer support but resolves cases, qualifies leads and tracks orders. Anthropic and OpenAI have agents that can control a user’s computer screen directly to handle tasks like pulling data and making online purchases. (Slide 91). Where AI was a research feature, it has since 2023 become a CapEx line item. It has become, in the words of Microsoft President Brad Smith, a “general-purpose technology” like electricity — “the next stage of industrialisation.”

Mary Meeker again:

The world’s biggest tech companies are spending tens of billions annually – not just to gather data, but to learn from it, reason with it and monetise it in real time. It’s still about data – but now, the advantage goes to those who can train on it fastest, personalise it deepest, and deploy it widest. (Slide 95)

This is where size helps. Training a model costs more than $100 million. Anthropic’s CEO has said that these costs could rise to $10 billion — per model. Inferences, while falling in unit cost, will likely “represent the overwhelming majority of future AI cost,” in the words of Amazon CEO Andy Jassy, because training is a periodic cost — done from time to time per model — while inference costs will be constant — every query. In Meeker’s words:

The economics of AI are evolving quickly – but for now, they remain driven by heavy capital intensity, large-scale infrastructure, and a race to serve exponentially expanding usage.

Meeker doesn’t walk us all the way down the path, and I’ll go into more detail in the third of this series of pieces — but it’s clear that worker productivity is top of most corporate agendas for embracing AI. She quotes a Morgan Stanley survey from November 2024 (Slide 330) where workers are top of mind: the largest adoption of AI was focused on employee productivity, the second highest worker savings.

Meeker avoids reaching her own conclusions on this. Instead she gives over a whole slide (Slide 336) to NVIDIA’s Jensen Huang, who paints a picture in the rosiest of terms. Yes, he says, jobs will be lost. But only to those who don’t take the opportunity. In fact, he argues, there’s a shortage of labour and anyone who takes advantage of AI will benefit. Here are the two bookends to the slide’s overall quote which probably encapsulate his thinking best, and illustrate the fist inside the velvet glove:

It is unquestionable, you’re not going to lose a job – your job — to an AI, but you’re going to lose your job to somebody who uses AI… I would recommend 100% of everybody you know take advantage of AI and don’t be that person who ignores this technology.

Meeker offers no annotation to this slide on the subject of workers. In the next couple of pieces I’ll try my hind