Mark Carroll's Journal

In writing here previously on generative AI I had wondered what happens to society when people can routinely lose themselves in artificial worlds of their own design. I had been thinking from the point of view of their being able to act out dark fantasies then adjusting to the real world where they don't make the rules and their actions affect others.

Anticipated by many works of fiction over the years, I was slow to consider what may be a good side of advancements in training inference models. There are many people who don't have enough contact with friends, perhaps especially the elderly. We may not be far from a point where they can have some artificial companion, patient and configurable, that offers interesting and helpful conversation on whatever topics the user wishes, even joining them actively in some pursuits, far beyond Alexa who can do little more than reading out the results from web searches.

Such companions may be considered a poor substitute for human contact but I suppose that there are probably funded startups chasing this very market.

In my migrating servers from one provider to another, rather than to the first provider's new platform, it turned out that the first somehow missed that I had ended my services with them and thought that they were all continued into the new platform. Fortunately, they seem to have readily corrected their records, cancelled further invoices, etc. On learning that they had thus overbilled me, they assured me that a refund had been issued. After pressing them a couple of times on this refund's absence, it turned out that actually, there had been another mistake. I did get my refund in the end but, goodness, it helps confirm my decision to migrate.

With secure communication online (TLS, etc.) it is interesting to see how standards develop: older ways are later judged insecure and the community slowly moves onto newer ones. I have wondered if the security services, and others, record some of the more interesting traffic that they can't decrypt yet in the hope that new developments might someday reveal the content of those once-private communications. People move on to different algorithms for actual reasons. Even if past the statute of limitations for prosecution, such records may still yield useful intelligence.

Now, given my job, I think about cryptocurrency more. There are some currencies, popular for payments for illegal services, that are designed to obscure the details of transfers. Even with normal cryptocurrencies, whose transfers are easily observed, there are tumblers which are busy accounts that take in many and various payments, and make payouts differently and rather later, so as to obscure the flows: they make it difficult to match the incoming funds against the outgoing.

I had already been wondering if statistical analysis of activity around tumblers may at least circumstantially reveal repeated flows for habitual users. Now I also wonder if some of the privacy-enhanced cryptocurrencies may be found to be less private than currently assumed, which would be interesting given that the blockchain records all the data publicly and long-term.

In short: as new discoveries uncover historical information, perhaps some people have bad surprises waiting for them.

I think it silly of His Majesty's Government to have switched from issuing physical visas to virtual "eVisas" that one checks by typing an alphanumeric code into a government website. It assumes far too much of mobile data connectivity when one is in some random place where one's documents are being checked. Further, what actually happens is absurd: people get you to show them your visa details in the web browser on your own mobile telephone, something easily faked.

Presumably thrilled with this success, apparently the Government is now moving on to introducing digital driving licences. For the moment, these are to be an optional add-on beyond the physical photocard, presumably most welcomed by those non-immigrants who would also like an easy way to fake their details. I suppose that I can imagine a version that shows a QR code bearing cryptographically signed data that can be verified by some reader, even printed out as insurance against device failure, but, given the eVisa implementation, I am not holding my breath for such.

Finally, I got around to the virtual server migration, rather last-minute. At least, I got the more critical parts, DNS and e-mail, working sufficiently the day before the deadline. I will get around to the other parts over time. It was fun to get back into that kind of work, and interesting to be reminded of how I had set things up previously. Thank you to those who offered suggestions: in the end, Mythic Beasts looked very much to be the right kind of provider and, so far, I am happy with the move to them.

Apart from other services, even my initial ones still need a bit of work: get my zones offered by the provider's DNS too, review the SSL certificates for SMTP/IMAP, sort out DMARC, etc. Still, things are already working well enough that I'll continue at my convenience.

At work, I am on a project that uses large language models (LLMs), an instance of the modern AI fad that brought us ChatGPT and, relatedly, image generators and suchlike. Personally, I am not into gaming and I don't own computers that have reasonable GPUs, and I have little use for systems that can deliver me results that are rather more plausible than they are trustworthy. My preference remains for the more traditional kind of AI centered on knowledge-based reasoning, though I concede that deep learning boasts some impressive successes.

Now that I am digging into modern AI a little for my day job, it becomes more obvious to me what everybody else probably realized years ago. First, I find myself tempted to use it for inane questions like recommendations from restaurant menus for kinds of people. That would be a lot of pointless computation: with each of us using LLMs for whatever comes to mind, just as we might ask our friends what they think, it sounds about as bad as proof of work for accelerating climate catastrophe.

Secondly, in reviewing various available models it became apparent that uncensored models are readily available, there can be guards and such applied subsequently. Many models are probably trained on all manner of material from the Internet, some of it from the sewer. Perhaps one can buy oneself decent graphics hardware, download uncensored models, then privately indulge whatever interactive fantasies come to mind. This goes back to previous questions on the effect of people being able to play violent video games or watch extreme pornography. Are we approaching a world in which anybody can immerse themselves in the particular virtual depravity of their choosing and, if so, what does that mean for society? I suppose that we will find out.

I am a Professional Member of the British Computer Society (BCS) who handle accreditation for my Chartered IT Professional (CITP) status. At the moment, I do not much use my membership except for being able to list it on my resumé. The royal charter granted by the Privy Council to the BCS allows them to award such status so, under the established British conventions, I can reasonably use MBCS, CITP post-nominally after my MA (Cantab).

My latest annual renewal fee was £216 and, right now, money is tight. I went ahead and paid it but I wonder how much employers actually notice and care. I used my membership a little more during the COVID-19 pandemic when more talks were broadcast online but, between work and family, I have little time at the moment for interacting with fellow professionals, etc. and even less for travelling to do so. I suppose that I shall see how things stand when renewal next comes around.

I was taking a look at what reasonable but cheap telescopes exist these days. Some of the marketing is funny, e.g., the SkyWatcher Evostar range is probably better for objects rather nearer to us than stars. Among the first things I noticed in looking at Celestron's products was how many modern telescopes now come wireless-enabled with apps to control them. I would be less sure of finding them still usable decades later, compared with say, my father's refractor from the 1960s.

I would also be sceptical of software quality partly because, back when I worked on software used by professionals who pay many thousands of pounds for their microscopes, it was clear that the microscope hardware was often of far better quality than the software with which the manufacturers lumbered their users. Goodness knows I am not easygoing when it comes to user experience, hence my grumbling ever since smart telephones and televisions became dominant, so I am cautious when it comes to buying software-dependent products.

On one of my work trips to London, EasyJet failed somewhat for the departure back home. They called us to the gate for the delayed flight far too early, especially given that the gate was small with hardly any seats. After having our boarding passes checked, we all got to stand in a small area watching the arriving passengers disembark, then continued standing for much longer while some system loudly played the last call announcement for the flight over and over and over again. Talking to somebody else afterward, they did the same for another flight.

For the same departure, it took quite some time to even get airside because security kept stopping altogether for no obvious reason; the staff and passengers would stand around, waiting for the music to restart or whatever.

An interesting thing I learned on the work trip was about BCPL, an ancestor of C. I may be misconstruing some details but the approximate story is that its simple compact bytecode is such that the interpreter core and your bytecode programme together may run even faster than a straight machine code version of your programme because they fit better into the processor's level-one cache. It seems plausible though I couldn't find a confirmatory reference at a glance.

My server hosting provider's retiring the framework I'm currently using so I get to install onto new servers somewhere. I don't need value-added services, I just need to be able to install onto cheap virtual servers from my chosen image (currently NetBSD), get network and console access, and have my server instances provide services from static IPs. Time to look around for options, simple and UK- or US-based would be ideal.