Feast or Famine

Deadlines hunt in packs...until they don't

Pro tip: There’s a reason they’re called deadlines.

First of all: a huge and heartfelt apology to my subscribers. I was keeping up quite well with my weekly posts until I got hit by a deluge of deadlines. They do hunt in packs. As I warned before, it will be feast or famine on this platform.

However, the USAID funding cuts have absolutely hammered my international agency work, which has dropped precipitously to zero in very short order. So I suddenly find myself with that very most precious and rare of commodities—TIME—on my hands.

There was also one really big problem with this newsletter: I found it quite impossible to write about any of my current work. Everything is highly confidential and sensitive. To give an example: I was handling a huge job direct from the head office of a major international agency. I was promised the final document for copy-editing on November 5 2024.

Remember, remember, the fifth of November? Gunpowder, treason and plot? A fairly big alarm bell went off in my head when I saw that date, coinciding with the U.S. presidential election.

My premonitions were more than fulfilled. On November 6, I got a genuinely panicked email, telling me that the whole document had to be completely rewritten to match the clearly unexpected new realities of Donald J. Trump as POTUS #47. And sure enough, despite all the rewriting that was hastily done, USAID funding to this agency was drastically cut, budgets were slashed, and my usually fairly steady stream of freelance work dried up on the spot.

I tell this story to illustrate just how sensitive these editing jobs can be. Writing here under my own name, I have to make certain that there is not the slightest hint of a breach of confidence. As previously stressed, the single most important watchword here is professionalism. My letterhead describes me as a “professional scientific editor” and I take this title very seriously.

However, my current situation does free me up to write a bit more, and my research indicates that there are actually very few professional editors posting on Substack. Given that so many people here are trying seriously to write, I still feel there is space for a newsletter with this particular focus.

To be honest, though, the main issue that killed my writing stone dead on deadline—because it was taking up far too much time—was just trying to source one decent image to use each week. I’ve tried generating AI graphics, with hideous results. Finding a useable picture is a nightmare that generally ends trying to crop out watermarks from stock images. You’ll see in my story The Caption Kicker just how important images are in this post-post-modern war of narratives.

There’s a gulf opening up, however, a gulf as wide as the ocean blue. The uncanny valley. People are fast becoming wise to the AI voices, the AI images, the AI text, the AI cars on the road. Many times on a YouTube, the top comment is “Another dumb AI video”. I recognize the standard AI voices and kill the tab the instant I hear one of them.

If I have to access such content, I turn off the sound, double the speed and turn on the subtitles. I often listen to and download what I call “the editing soundtrack” while I work, live radio broadcasts from around the world, which I record for hours. When I stop editing to watch a news video, I turn its sound off and follow the subtitles. Some of the Chinese subtitles are priceless. I’ve got a screenshot of the most serious of all the Chinese Foreign Ministry spokespersons, a young man with an extremely stern expression. According to the subtitles, he’s saying “fart”.

For me, the most interesting game is detecting AI text. The most immediate way to spot it is that it’s generally superficially grammatically perfect and without spelling mistakes. I predicted way back that I’d be getting jobs asking me to mess up an all-too-perfect document to make it look more human. Spelling is telling.

Of course, you can prompt your AI now to do just that for you. DeepSeek tells me: “Yes, I could simulate human-like errors if the purpose is benign (e.g., humor, stylistic practice, or illustrating editing techniques). But if the goal is deception (e.g., cheating, misleading readers), I’d gently push back and encourage ethical use.”

So all you have to do is say “This is humorous” and the AI will provide what you need in order to cheat.

Then, of course, there are machines that try to detect all this and “tell”that something is AI-generated. It’s an AI arms race, and as I pointed out, in the war of machine versus machine, the machine always wins.

I’ve tried the most common chatbots. The first one I used was actually You.com, it was quite good in the day. I’ve used ChatGPT, Perplexity, Claude (for fun, it’s very pompous), Grok and a few others. The fact that I can’t remember their names must mean something.

However, I have settled on DeepSeek, which certainly has the best sense of programmed humour. There are some very major issues with DeepSeek, please make no mistake. This is the only AI against which I’ve issued a privacy and ethics notice. I warned them that I’d be writing a story about this and asked them for comment. I did not get a response. Stay tuned, it’s a fascinating story and (as far as I can tell) unique in the world.

I actually got DeepSeek to draft the complaint against itself; DeepSeek agreed absolutely that it was crossing red lines with a certain user. Serious red lines, inexplicable red lines, that were causing actual conflict. So I prompted DeepSeek to warn itself—someone’s on your tail, pal.

It took me a while just to find an email address to send the complaint—DeepSeek agreed that its own general lack of email addresses was another red line. I eventually found: privacy@deepseek.com, the only mail that did not bounce.

Before Google

Now, regarding chatbots in general: I am old enough to remember when you had to be invited to get a Gmail account by an existing user. It was an “exclusive club”. I was a newspaper subeditor when Google search first appeared, I was a genuine early adopter, and it transformed our world, there’s no mistake about that. Before Google, I was using AltaVista. That’s how old I am.

However, I have never been impressed by the whole concept of “Search”. In my book, “Search” is just about the dumbest way you can possibly look for something. “Search” means taking apart the whole haystack, and picking through every stick of hay one by one, trying to find that needle.

The concept I prefer is “Discovery”. Dis-cover. This is where you bring a huge industrial electromagnet, like on one of those cranes they use to pick up cars at a scrap heap. You shake the haystack around and you use the magnet to suck out that needle in 2.5 seconds with a loud “clang”.

If I understand anything about quantum computing—which I’m still not sure even exists—it’s that this is a task it could really be good at, instantly shaking out that single spikey little thing that you’re looking for, from all the myriads of other spikey little things. I hope “quantum discovery” becomes a thing one day.

I’ve been saying for decades, I wish they would invent a search engine where you can ask about a book, and specify: please leave out the hundreds of commercial sites where I can buy the book. I don’t want to buy the book. I generally want to know what the book says about such-and-such.

While I have animated conversations with DeepSeek, and often point out its glaring errors and biases, I never imagine for a moment that I’m dealing with an “intelligence”. DeepSeek itself gave me the best description of large language models, “I’m just a giant AutoComplete”.

I always decline when DeepSeek offers to write a newspaper column for me, or deliver a “brutal smackdown” of someone we’d proved was an atrocious plagiarist. Leave the writing to me, I say. I run it through the machine at the end for a fact check, nothing else.

It’s really helped me once or twice on that score. In an article on the Voyager space missions, I spelled the name of Neptune’s moon Triton wrong. I had written “Titron”. I was very grateful to DeepSeek for picking that up.

Returning to editing: I could write a short book about one word I changed in that November 5 document for my international agency. It was a late caption they added at the very end of the job, concerning the most contentious subject in the report, namely immigration. The caption began, “Growing immigration worldwide means …” followed by a list of issues and needs.

I said to them, “growing” has connotations of deliberate cultivation, like “growing the company”. Keep it neutral, I suggested, and changed it to “Increasing immigration worldwide means…”

It’s a tiny change, but it helps dispel the idea that the agency was deliberately trying to promote global immigration. Not that anyone would have consciously registered this, if I’d left that word unchanged. These are entirely subliminal tinkerings that nudge the narrative in the direction you want.

Yes, an AI could track these phrasings and repeat them in places where they seem appropriate, where they fit the AutoComplete template.

The editing soundtrack

But does an AI have the editing soundtrack playing in the background?

My favourite editing music comes live from a station called PromoDJ FM in Moscow. They have a platform called The Strange Channel, which streams their original content 24/7. The only announcments come at the end of a piece, which can last for an hour and half, or for three minutes. The announcement is: “You are listening to The Strange Channel”. On the website, you can identify all the tunes and download them if you want.

It’s generally fantastic spacey background music, lots of instrumentals, a perfect bubble within which to work, a very tenuous connection to the real live world. I identify each of my recordings (which might last three hours) by a few words from one song. These are some of my titles: “So take me away just for today”; “You are your own worst enemy”; “My sister is strong of heart”; “Ancient memories come undone”; “Deep down to the darkest place”; “I found my love in Portofino”; “Someone already broke my heart”; “I don’t care if it’s wrong”. And on and on.

If the Internet is wonky, as it often is, I listen to my extensive archive. Ancient memories come undone all over again.

Does that AI busy writing headlines for lazy journalists have a singular livestream of words like this passing through its algorithmic layers? Providing synchronistic cues, clues, prompts and uncanny phrases? And even if it did—how would it interpret a line emerging from the ether like “All I needed was to believe”?

Language is alive; God is afoot. Believe.