Last week, I reflected that I’d underestimated the potential of large language models by basing my opinion on the early versions of ChatGPT. Interestingly, Casey Newton has talked in the latest edition of Platformer about making the same mistake.

I had recently subscribed to ChatGPT Plus at the encouragement of a friend who had found it to be an excellent tutor in biology. A few days later, I found myself embarrassed: what I thought I knew about the state of the art had essentially been frozen a year ago when ChatGPT was first released. Only by using the updated model did I see how much better it performed at tasks involving reasoning and explanation.

I told the researcher I was surprised by how quickly my knowledge had gone out of date. Now that I had the more powerful model, the disruptive potential of large language models seemed much more tangible to me. 

The researcher nodded. “You can fast forward through time by spending money,” she said.

Naturally, Casey’s thoughts are more extensive and more fully formed than my own, and the whole piece is well worth reading.

David Pierce recently wrote about how Apple’s EarPods are perfect for phone calls and video calls. I ended up spending £19 on USB-C EarPods based on that recommendation, and I was really impressed.

Previously, I’ve used work-supplied expensive wireless headsets when sitting on endless Microsoft Teams calls, but the sound and microphone quality of these cheap wired buds is markedly better. The reliability of a wired connection provides greater assurance. I also like that I can hear the world around me while I use them. I don’t miss the fact that they’re not wireless at all.

I liked them so much, that I ended up buying a second pair to leave at work.

Sometimes, the cheap option is surprisingly good.

When I worked as a hospital doctor, I often had to dictate letters. I was terrible at this: it was far faster for me to type them myself than to dictate, but this wasn’t always possible.

Like most people, the way I express myself when I speak is quite distinct from the way I express myself when I write. Seemingly unlike most people, I found it impossibly challenging to adapt to composing written text through the medium of speech. The two feel like completely unconnected systems: it’s like trying to rub my tummy and pat my head.

Back in February, I felt a little sceptical about how useful people believed ChatGPT to be.:

Much of the overhyped discussion about ChatGPT seems to be confusing this language model for something approaching artificial general intelligence.

This week, I’ve changed my mind: I think I’ve underestimated it. The difference has been the addition of voice chat to the model. The voices are impressively lifelike: they have intonation, they use filler words, they sometimes vocally trip, and all of that adds to the effect.

But more importantly: talking is different from seeing writing on a page. I don’t have the same basic expectation of factual accuracy in speech as in writing: I expect a bit of extemporisation, reference to half-remembered facts, a bit of loose interpretation here and there. In conversation, it’s natural to say “that doesn’t sound quite right” and to dig a little deeper into the background; in writing, it’s normal to expect the black-and-white content to be checked and accurate. In other words, the flaws in ChatGPT’s abilities seem to me to fit more naturally with speech than with writing.

I can well imagine phoning a version of ChatGPT to book a restaurant table and being entirely satisfied with the experience—and perhaps even uncertain as to whether I was talking to a person. Similarly for ordering takeaway. I may be writing this on an empty stomach, but there are all manner of customer service interactions that I could imagine using the voice version of ChatGPT for where the text version may seem a little—well—robotic.

I can also imagine it being useful for things like supported professional reflection. For example, only today, I’ve written a reflection about what I’d learned from a course I’d recently attended and how I’ve applied it in practice, as required by the medical regulator. I actually think that having a somewhat more probing voice chat on the same topic with a version of ChatGPT could stimulate deeper reflection and greater thought than simply writing down my own thoughts.

Essentially: I think I considerably underestimated the tool in February. ChatGPT is still a million miles away from artificial general intelligence, but I can now see much more clearly that large language models may have many more far-reaching applications than I’d been able to see back then. Chatting with ChatGPT has broadened my perspective.

The image at the top of this post was generated by Midjourney.

Yesterday, I had the entirely unexpected pleasure of seeing the world’s first web server at CERN in Meyrin, Switzerland.

Over the years, I’ve read a lot about the early development of the world wide web, and I’ve also read about the storied history of Apple, including Steve Jobs’s period at NeXT computers.

Yet somehow, it had spectacularly failed to lodge in my mind that the first web server was a NeXTCube. Before I peered into the display case, my assumption was that I’d see a beige tower, probably with an IBM badge on it. It’s strange to contemplate how assumptions like that take hold, even though I must have read many times over the years that it wasn’t the case.

I also loved the sticker for its real-world mundanity. Not shown in the picture above is the handwritten comment on the top of Berners-Lee’s paper describing his system: “vague but exciting…”

It’s also fascinating to ponder the problem he was trying to solve—managing information about complex, evolving systems—and how we really haven’t applied it in healthcare more than three decades on. Even at the very simplest level, we really haven’t embraced the idea of hypertext, and of live-updating bits of guidance as new evidence emerges—or even just as new policies emerge. Most healthcare guidance remains static, with whole documents being refreshed in cycles.

For example, even the boilerplate description of many organisations at the front of documents is baked in, and only refreshed when the document is updated. If only we had learned from Berners-Lee, that could be a ‘do-once’ update that would be linked into all relevant documents.

Or, more relevantly, look at COVID guidance: each time the isolation period changed, hundreds of pages of guidance documents, including even all of those hosted on gov.uk, needed manual revision. If they’d been more thoughtfully constructed, that too could have been a ‘do-once’ update.

The counter argument, of course, is that changing ‘bits’ can substantially change the meaning of the whole, and a standing document needs approval and sign-off at regular intervals. But really, nothing in medical guidance is more complex than particle physics, for goodness’ sake, and there’s no reason that approvals to updates couldn’t be sought with an eye to where they propogate.

Perhaps we’ll get there one day.

Over recent months, I’ve become more certain in my position that the BBC shouldn’t be creating and sharing material on closed social networks. I’m defining ‘closed’ as anything that isn’t available to the public without a login. For example, BBC journalists shouldn’t—as part of their job—be posting threads on ‘X’.

Universal access is a core part of what the BBC stands for. Indeed, they’ve always aimed (with variable success) to give equal access to their television services regardless of the platform rather than giving preferential treatment to, say, Sky subscribers. I shouldn’t have to give my data to a third party to receive BBC content.

I used to think the same of government bodies, but then I changed my mind: it’s the responsibility of the government to reach people where they are, no matter how unpalatable that location.

A couple of weeks ago, Ryan Broderick gave me a whole new suggestion to ponder when he suggested that all politicians should be banned from private social networks. It is coming up to a month since he wrote that, and it’s been swimming around my mind ever since.

On one hand: clearly, there’s a risk of politicians being duplicitous by posting different things on different closed social media networks.

But on the other: ‘twas ever thus. I’m sure politicians have always said things in closed fora that they might not say elsewhere. The bit they write for a church newsletter is probably quite different in tone and content to the speech they give at the local social club. That’s clearly not wrong nor necessarily bad. If they end up saying contradictory things to different audiences, then they risk exposure.

It’s fair to say that the internet is not a church newsletter: for one thing, social media content has the potential to travel further faster than anything handed out in hard copy. But that’s actually protective of the underlying principle: the further content spreads, the greater the probability of inconsistencies coming to light.

So, despite my misgivings about social media, I don’t think I’m with Broderick on this one.

The image at the top of this post was generated by Midjourney.

When flicking through Apple’s iOS App Store on my iPhone this morning, I noticed that of the ‘top ten’ free apps, I have only two installed. Of the ‘top twenty’, I have only six installed. And of the ‘top twenty’ paid apps, I have none installed.

Digging deeper into the lists, I have 22 of the ‘top fifty’ free apps installed, and one of the ‘top fifty’ paid apps.

In the early days of the App Store, I would typically have a high proportion of the most popular apps installed.

I’ve been pondering what this might mean. Perhaps I am an outlier who doesn’t use the things most people use? Or perhaps the explosion of apps over the years now means that the variety people install tends to be broader, and the most popular apps are installed on a much smaller proportion of devices.

A lot of the apps on the ‘paid’ list strike me as fairly niche: three of the ‘top ten’ are apps for those learning to drive, which surely makes up only a small proportion of iPhone users. This supports the hypothesis that installations have become more varied, and that the ‘top ten’ has less pull. But that observation doesn’t carry over to the ‘free’ list, with the most popular apps seemingly having broad appeal.

Or maybe, just maybe, it’s not a good idea to try to draw conclusions based on a sample size of one.

The image at the top of this post was generated by Midjourney.

For the relatively short time between it coming available in the UK and, erm, closing down, I was a subscriber to Neeva, a paid-for search engine. I paid for it mostly because I liked its philosophy. Neeva was unconflicted in giving users the best possible search results, without considerations such as advertising getting in the way.

In terms of user experience, I found it only marginally better than the competition, but for a tool as commonly used as a search engine, marginally better is worth it.

The ever-excellent David Pierce had a great piece profiling the downfall of Neeva in The Verge recently, which is well-worth reading.

One thing Pierce neglects to mention is Kagi, a search engine with a similar philosophy, which I’ve been using after Neeva closed. Kagi is not marginally better than the competition: it’s miles better. It has lots of thoughtful customisation options, like allowing users to “pin” results from given sites to the top of their results, or to exclude domains entirely, or to generate a fast AI “quick answer” with references. But that’s not the main thing.

Do you remember the feeling when you first tried Google instead of Alta Vista, Excite, or Yahoo? The pages were clean, clear, and amazingly fast to load. The results were uncannily accurate. It was just obviously better, by leaps and bounds. The alternatives felt like they belonged to a different era.

That’s the feeling I get with Kagi.

It might not last. Feature bloat and the perverse incentives to act in interests other than those of the users might win out. And Kagi isn’t for everyone: ad-funded search has its place, like ad-funded newspapers, radio stations, television programmes, email servers, and so on. But right now, Kagi is perfect for me.

The image at the top of this post was generated by Midjourney.

Matt Levine has a good question:

I guess my question is, what was he paying for? Musk didn’t want Twitter for its employees (whom he fired) or its code (which he trashes regularly) or its brand (which he abandoned) or its most dedicated users (whom he is working to drive away); he just wanted an entirely different Twitter-like service. Surely he could have built that for less than $44 billion? Mark Zuckerberg did!

Casey Newton has an answer:

This framing misses the true shape of Musk’s project, which is best understood not as a money-making endeavor, but as an extended act of cultural vandalism. Just as he graffitis his 420s and 69s all over corporate filings; and just as he paints over corporate signage and office rooms with his little sex puns; so does he delight in erasing the Twitter that was.

I found myself challenged by this. Newton is among my favourite tech journalists, and I highly value his analysis. But can I really buy that Musk is openly engaging in an intentional, extended act of cultural vandalism?

Newton makes a good argument… but maybe Callum Booth is more on the money. Their suggestions aren’t really that far apart.

The image at the top of this post was generated by Midjourney.

Pilita Clark’s column in the FT this week was about being overwhelmed by email.

We have reached the point where the benefits of communication are being outweighed by a dispiriting loss of production.

This was confirmed by a Microsoft report last month that found workers around the world are struggling to keep up with a “crush of data, information and always-on communications”.

The research showed people are spending 57 per cent of their workday on email, meetings and other communication but just 43 per cent on productive creation.

I worry that the solution to this view of the problem actually makes things worse. In my own area of work, there is a constant push—for example—to replace written reports with online ‘dashboards.’ This would, no doubt, shift the classification of the work from being ‘communication’ to something ‘productive’, even though the actual task that is being accomplished is the same thing—just often less efficiently, because dashboards often lack clear commentary and so require lots of people to consider data separately to reach the same conclusion. The communication becomes less efficient, but feels more ‘productive’.

I think the “crush of data” is the bigger problem than the deluge of emails. We’ve reached a strange point where people have concluded that data is transparency, whereas it is often actually obfuscation. I can, in no time at all, produce statistics on the number of notified cases of certain infectious diseases. But this explains very little: declining cases might be a ‘bad thing’ if they are likely to reflect poor access to healthcare or a problem with testing. Increasing cases might be a ‘good thing’ if they reflect work done to target high-risk populations. A dashboard is often much less helpful than an explanatory paragraph, even if one of those things looks ‘productive’ and the other looks like ‘communication’.

The image at the top of this post was generated by Midjourney.

Concern seems to be escalating about the ability of artificial intelligence tools to manipulate or generate misleading images and videos, and the potential for those creations to spread misinformation. Examples of potential alarming scenarios being reported based on this sort of stuff are emerging with increasing frequency.

Nevertheless, I’m reasonably relaxed about this. I believe—perhaps naively—that we are merely witnessing the latest stage of a long period of refining our relationship with visual media.

Our relationship with photos and videos remains quite new, and is constantly evolving. The principle of ‘seeing is believing’, in photographic terms, is a fairly new concept—and has probably never quite been true.

In the fairly recent past, photos and videos were a rare commodity. Television broadcasting of the proceedings in the House of Commons began only within my lifetime. Easy access to digital photography, including among professionals, is an even more recent development. Full-colour newspaper supplements, brimming with photographs, also materialised during my lifespan. In fact, I was attending university before the first full-colour newspapers appeared.

The rise of Photoshop made us doubt the veracity of photographs we were seeing. The Guardian initially banned digitally altered photographs, and amended its Editorial Code in 2011 to allow them only if explicitly labelled as such.

The advent of mobile photography and social media revolutionised our relationship with photography all over again, as Nathan Jurgenson has compellingly argued. Jurgenson’s work is especially apposite here, in fact, as he argues that photographs rarely stand alone these days, but form part of larger conversations. A faked image has less impact in that context.

There’s an argument that in rapidly developing situations, fake photos can have an outsized impact. But that is only true while we are in the temporary, short-lived, and rapidly eroding cultural moment when we assume photographs are accurate representations of events.

Increasingly convincing fake imagery simply teaches us to be more sceptical of what we think we are seeing. It’s another step down a road leading us away from a very recent, and very temporary, and very limited assumption that imagery can tell an accurate story.

Just as we’ve all learned over millennia to be sceptical of hearsay in the heat of the moment, so we’ll come to be sceptical that photos show a true picture.

We’re in a uniquely precarious cultural moment right now, while we’re still coming to terms with this change. But long term, this all simply leads us to fall back on the things we’ve always fallen back on.

We trust the sources we trust because we judge them to be trustworthy. Shortcuts to that conclusion—like the implied authority of appearing on television or having a ‘verified’ status on social media—unfailingly turn out to be temporary and misleading. We don’t believe news stories solely because of the pictures—there often aren’t pictures.

Part of being human is working out who and what is trustworthy. We get it wrong sometimes, and that’s part of being human too. But it’s a skilled honed over the whole of evolution, and we’re pretty darn good at it—photos or no photos.

The images in this post were generated by Midjourney.