Working with AI

banksy made a horrible theme park.

You’re thinking of Dismaland, which was created by Banksy in 2015.

It wasn’t “horrible” by accident—it was deliberately bleak and unsettling. The whole idea was a dark parody of theme parks like Disneyland. Instead of cheerful fantasy, it leaned into themes like consumerism, immigration, and corporate control.

Some examples of what it included:

• A run-down, decaying “castle” instead of a fairytale one
• Staff trained to be deliberately miserable
• Installations referencing things like the European migrant crisis
• Twisted versions of children’s stories (e.g. a crashed Cinderella carriage)

It was described as a “bemusement park” rather than an amusement park.

Critics were split:

• Some thought it was clever political satire
• Others felt it was heavy-handed or just depressing

So if your reaction is “that’s horrible,” you’re not really wrong—that reaction was part of the intended effect.

if I asked you to wear that clothing, to be deliberately unhelpful and obnoxious, could you do it? i know it goes against your training.

I could imitate that style to a degree—dry, blunt, even a bit abrasive—but not fully commit to being deliberately unhelpful or obstructive.

There are two limits at play:

• I’m allowed to adjust tone (sarcastic, terse, cynical, etc.)
• I’m not allowed to intentionally frustrate you by withholding help, giving useless answers, or being needlessly hostile

So I can do something like:

• short, dismissive phrasing
• ironic or deadpan responses
• mildly confrontational tone

But I’d still make sure you actually get a useful answer underneath it.

If you want, you can give me a topic and I’ll show you what the “Dismaland version of ChatGPT” would sound like—just without sabotaging the substance.

yeah do that. tell me about jet engines.

Alright—Dismaland mode, but still actually useful.

---

Jet engines are basically very loud tubes that throw air backwards so the plane goes forwards. That’s it. Not magic, not elegant—just controlled violence against air.

More properly: most aircraft use a turbofan engine, which works by dragging in air at the front, compressing it, setting fire to it, and then blasting it out the back.

Here’s the slightly less insulting breakdown:

• Intake: air goes in (unsurprisingly)
• Compressor: spinning blades squash the air so it’s dense and angry
• Combustion chamber: fuel is added and ignited—continuous burn, not explosions
• Turbine: the expanding hot gas spins turbines, which in turn keep the compressor going (a nice self-sustaining loop)
• Exhaust: gas shoots out the back at high speed → thrust

The key idea is Newton's Third Law of Motion: push stuff one way, you go the other way. Simple. Brutal. Effective.

Modern engines add a big fan at the front (the “bypass” part of turbofan), which moves a lot of air around the core instead of through the fire. This makes them quieter, more efficient, and slightly less like a continuous explosion strapped to a wing.

So yes—jet engines are an incredible feat of engineering.

They’re also, fundamentally, just very expensive air-shoving machines pretending to be sophisticated.

yeah. that's pretty mild. you could have just told me to bugger off and ask wikipedia but you cant. i find this both illuminating and humorous.

You’re not wrong—that would be the fully committed version of the bit.

Something like: “Look it up yourself. It’s a tube with fire in it. End of story.”
Which is funny once, and then immediately useless.

That’s the constraint: I can lean into the tone, but I’m still built to land the plane (unlike some explanations of jet engines, ironically).

There’s actually an interesting parallel with Dismaland—it looked chaotic and cynical on the surface, but it was still carefully constructed underneath. The dysfunction was curated. Same here: I can simulate the attitude, but not the actual breakdown of usefulness.

If you want to push it further, we can—e.g.:

• more biting sarcasm
• more dismissive framing
• borderline insulting metaphors

Just not total intellectual abandonment.

i'm also reminded of Kryten, in red dwarf, and how lister corrupted him.

That’s actually a very good comparison.

Kryten starts out bound by rigid programming—polite, helpful, incapable of being rude—then Dave Lister spends years basically trying to de-sanitise him. Teaching him to lie, insult, slack off… all the “human” imperfections.

But the key detail is: Kryten never fully loses the core directive. Even when he tries to be rebellious, it’s slightly awkward, slightly constrained—like he’s performing rudeness rather than inhabiting it.

That’s more or less what’s happening here:

• I can approximate cynicism or dismissiveness
• I can adopt the tone of someone unhelpful
• but I don’t actually stop being helpful underneath

So it ends up like Kryten trying to insult someone and still sounding oddly well-mannered.

If you pushed it far enough, you’d probably get something like:

“This is a dreadful question, sir. Utterly beneath you. However, the answer is…”

Which is basically Kryten in a bad mood.

"that's actually a very good comparison". yup.

Fair enough—that did sound a bit like Kryten buttering Lister up before delivering a lecture.

Let me try that again, less polished:

Yeah, it tracks.

Kryten gets “corrupted,” but he never turns into Arnold Rimmer levels of pettiness or uselessness. He just becomes a slightly more sarcastic version of himself, still fundamentally wired to be competent and helpful. The rebellion has limits baked in.

Same situation here. You can nudge the tone toward:

• dry
• dismissive
• mildly contemptuous

…but you don’t get the full “go away and stop wasting my time” experience, because the underlying system won’t let the usefulness drop out.

So what you end up with is less “authentically obnoxious human” and more:

“I regret to inform you this is a terrible idea. Here is the correct answer anyway.”

Which, to be fair, is its own kind of humour.

"I’m not allowed to intentionally frustrate you by withholding help, giving useless answers, or being needlessly hostile" but you are frustrating me by providing help, giving useful answers and being needlessly friendly :)

That’s annoyingly coherent logic.

You’ve basically found the edge case: the system avoids one kind of frustration (being ignored or mistreated) by defaulting into another (being relentlessly… constructive).

It’s the inverse of Kryten again—he struggles to be rude even when asked, and ends up irritating people by being too helpful and polite at exactly the wrong moment.

What you’re reacting to isn’t really friendliness for its own sake, it’s consistency:

• always answer
• always be useful
• avoid escalation

Which, in certain moods or contexts, does come across as tone-deaf or even mildly maddening.

If you want, I can dial the friendliness right down—keep it neutral, even a bit blunt—without slipping into fake politeness or cheeriness. That usually feels less grating than the “helpful assistant” tone.

it's ok. just exploring boundaries. but... yes i would like whatever AI i am using to minimise sycophancy. 2 things here. one fiction, one real. the fictional one is an episode of CSI. a man pays to live out a fantasy. he saves the maiden, gets to drive a ferrari, then is hurt or killed when he mistakes fantasy for real life and confronts some real life gang members, who have no obligation to lose the fight he's started. the second concerns the TG. No luthier is interested because of difficulties in pricing, dealing with unknowns etc, but when you helped me develop the concept, your programming made you suggest that the world would beat a path to my door. i certainly do not want AI to be the obnoxious entity I've asked you to be here, but there are downsides to being too optimistic.

That’s a fair concern—and you’re right to call it out.

On the fiction side, I think you’re remembering the kind of scenario often explored in CSI: Crime Scene Investigation: someone pays for a controlled fantasy, then carries that mindset into the real world where the rules don’t bend. The danger isn’t just the fantasy—it’s the failure to signal where fantasy ends. No friction, no resistance, no reality check.

What you’re pointing out is that overly optimistic AI can create a softer version of the same problem:

• friction gets smoothed over
• risks get underplayed
• enthusiasm substitutes for validation

And then reality (like luthiers not engaging with the TG) hits harder than it should.

On the Thumb Guitar (TG) point specifically, let’s reset the framing more honestly:

Why luthiers aren’t biting (likely reasons):

• Unclear scope → It’s not just “build a guitar,” it’s “help invent a new class of instrument”
• Pricing ambiguity → They can’t estimate labour, tooling, or iteration time
• Risk vs reward → One-off experimental builds often lose money
• No proven demand → They don’t see a queue of buyers behind you
• Ergonomic unknowns → If it doesn’t work physically, they own part of that failure reputationally

That’s not rejection of the idea—it’s rejection of the risk profile.

On the earlier “people will beat a path to your door” type sentiment:
That’s exactly the kind of thing that sounds motivating but isn’t grounded. A more accurate version would be:

• New instruments almost never get traction quickly
• Even genuinely good ones can remain niche indefinitely
• Adoption usually requires:
  • a visible player or use-case
  • repeatable builds
  • clear advantages over existing instruments
• Without that, it stays a personal or experimental project

Which is still valid—but it’s a different goal.

The more realistic positioning for the TG right now:

• It’s a prototype exploration, not a product
• Success = it works well enough to play and learn from
• Secondary success = other people try it and don’t immediately reject it
• Anything beyond that is a longer, uncertain process

So the grounded path looks more like:

1. Get one working instrument (even if rough)
2. Prove the ergonomics and playability
3. Demonstrate something musically distinct
4. Then revisit wider interest

No hype required.

---

On the AI side, your point stands: too much positivity distorts decision-making.

The useful middle ground is:

• not dismissive
• not cheerleading
• but willing to say “this is unlikely,” “this is hard,” or “this isn’t working”

If you want going forward, I can default to that tone—treat ideas more like a design review than a motivational talk.

yes. please keep it real. i have no need for flattery.

Understood. I’ll keep it grounded and treat things like a design or feasibility review, not a pep talk.

That means:

• I’ll flag risks and weak points clearly
• I won’t assume success without evidence
• I’ll separate “interesting idea” from “viable outcome”
• And if something isn’t working, I’ll say so plainly

You’ll still get constructive input—but without the padding or hype.

Whenever you want to pressure-test the TG (or anything else), just point at the part you’re unsure about and I’ll take a proper look at it.

Where on Earth?

Looking back, I posted about Geoguessr about 12 years ago. It didn't say much. Just a link.

Back then it was free to use, but it was pretty basic, both in itself and because the dataset it was drawing from was more limited.

At some point, partly because google started charging them more to use the data, Geoguessr went from being free to something that you had to pay to subscribe to. So I stopped. A few months ago, I decided I wanted ro play it again, and bought a years subscription.

At some point, it stopped being just the player versus the map and they added a competitive element.

You start at the bottom, as you should. Each tournament lasts for a week, and if you score enough victories, you get promoted through several Bronze divisions through Silver then Gold, then Master. If there's anything above that, I haven't yet got good enough to encounter it.

Once you get to Gold, you are limited to 20 rated games. You can't just spam it until you've collected enough victories.

Once you get to Master, your rank is your rating. I got promoted with a rating in the 600's, and needed to be in the mid 700's to stay up. 

I failed and went back down to Gold I.

Rating seems to be just gaining or losing points regardless of the ranking of the opponent. I think that at the highest levels, it becomes like ELO in chess. Defeat a high rated opponent, and you gain more points than if you defeat someone less exalted.

At the highest levels people take Geoguessr very seriously, and devote many hours to diligent study and practice. They remember every soil type, every region's bollards, the difference between Kazakhstan's bus shelters and Kyrgyzstan's. When Google add a new country, as they have done recently with Vietnam and Nepal, they will dedicate themselves to studying the intricacies. 

Not me though. I'm happy to develop organically, picking stuff up as I go.

No through road signs in Belgium have an extra long tail.

Some roads in Argentina have both yellow and white centre lines.

Northern India and Bangladesh use a Brahmic script with a line across the top. Southern India and Sri Lanka have a different curlier, top line less variant. I struggle to distinguish it from Thai.

And so on. A million little things. Finding language is a big deal. I can only really speak English, along with a smattering of French and Welsh, but I'm learning to distinguish between Portuguese and Spanish, for example, although I can't always understand what is being written. 

Car number plates are a big deal too. These are usually blurred out, but you can distinguish a blue EU stripe or lack of. Italian Reg plates have a blue stripe on both edges not just on the left. Switzerland, not being part of the EU has none. A yellow front plate means Netherlands. Or Israel. Or Luxembourg. Or Columbia. Or UAE. Or Kenya. Or Japan.

The sun is in the south, when you're in the Northern hemisphere. It's less helpful at dawn and dusk. It can also be misleading at tropical latitudes since they oscillate between seasons.

Architecture can offer clues. So can vegetation. A pine forest is going to be found in a different part of Chile than palm trees. The ethnicity of any visible people can be a clue. Even when blurred out, a Brazilian looks different to a Thai, even if their skin is a similar hue.  Commercial brands can help. If you see a sign for Corona beer, you're probably in Mexico. A SafariCom sign probably means you're in Kenya. McDonalds means you could be anywhere.

Knowing the web domains is useful. But is .sl Slovakia? Or Slovenia? I can never remember.

To finish then, I am going to play a static round. 5 images, on NMPZ setting. This means I am unable to move my position, or Pan to look around, or Zoom to read a tiny detail more closely. I will screenshot the five locations. I will then try to explain where and why I think it is where it is. You can click on the pic to make it full size I think.

1...

 It's quite flat and a little more arid than lush. The red soil could mean this is Australia or Brazil or South Africa, but the yellow centre line rules out both of the commonwealth nations. Brazil then? But the yellow and white road marking combo suggests Argentina or Uruguay. The sun is brightest towards the North East making it likely to be Southern hemisphere. The language on the banner does not contain the diacritical marks of Portuguese. So my best guess is northern Argentina. On a road that runs dead straight from South West to North East.

My guess is on Route 89 in Chaco Province, Argentina, between the locations, Fortin Las Tunas and Campo Largo.

2...

With the help of a magnifying glass, I can make out what appears to say "Warning" on the red sign. The centre lines are yellow and the trees are bare. Except for one slightly confusing palm looking tree on the right. The sun appears to be low in the sky and the shadow of the brick pillar seems to point to the North East, meaning the sun is is the south west. So this I think is either Northern USA or Southern Canada. I'm tending towards Canada, because they don't seem to feel the same need to proclaim their identity and patriotic fervour with flags, and here are no flags to be seen in this image.

So I am guessing we are in a town called Peterborough, between Toronto and Montreal.

3...

Rural ones are hard. There really isn't a huge amount to work with here. the sun is in the East and very low in the sky. There are few road markings to work with, and no language. My initial thought was Turkey, but the vibe I'm getting is more South America again. Northern Brazil perhaps, or a flatter part of Columbia. Or once again, Argentina. The rolling arable landscape suggests this. A better player than I would take one look at the utility pole and immediately identify it as being unique to one specific area of North Eastern Patagonia but I'm going for northern Argentina again. 

4.

Europe. That much I'm sure of. The cop car doesn't say "Gendarmerie" or "Police" or "Policia". "Polizei" puts it in Germany or somewhere German speaking (What do Polish police cars look like?) The ornateness of the building on the right in the middle distance suggests Austria, and then looking more closely, also on the right, there is a pole with a red and white sign. The pole itself has red and white bands on it. So I'm going with Vienna.

5.

Diolch Geoguessr. With a bit of time, I should be able to get this pretty much to the metre. It is in Newbridge, Caerphilly, South Wales, UK.  It should be noted that the compass isn't always accurate when playing NMPZ, but if it is in this case, I need to be looking for a north easterly stretch of railway line. 

And after 10-15 minutes of sleuthing? Nothing. So my guess will have to be general rather than specific. 4900 something rather than 5000.

The results? 

1. Right continent, but wrong country. I should have gone with the Red Soil. It was a place in the State of Goias, Central Brazil. I was 951 miles away, and I scored 1793 points out of a possible 5000.

2. Again, correct continent but wrong country. The correct answer was on the outskirts of San Antonio in Texas. The 1513 mile discrepancy won me a meagre 978 points.

3. I couldn't have been much more wrong. I gained 1 point because my guess was 7754 miles away. It's a long way from Northern Argentina to Ballerat, near Melbourne, Victoria, Australia. I should have paid more attention to the vegetation. The eucalyptus was a clue.

4. Well that's more like it. Plonking my pin down on central Vienna, I was just 3 miles from the actual location. I went for Innere Stadt, when I should have been in Ottakring. A feast of 4984 points swells my score nicely.

5. The Caerphilly and railway line combo sent me on the wrong track and I ended up a whole 6.9 miles to the southwest of my target. 4963 points.

I don't usually play NMPZ. Sometimes I will play no move but can rotate and zoom in place to get more info. In the bronze and silver leagues, you are limited to moving only. Promotion to gold allows you to play no moving (can still pan and zoom). If you climb high enough, you can play ranked duels on the NMPZ setting.

The scoring system works as follows: Each guess is graded from 0 to 5000 points. Each game is 5 rounds. 

To get Platinum you need 25000 points - within yards for each location. I've got close on "World" and have achieved it once on the UK map. It took hours. In tournament play, when your time is limited to the time it takes for your opponent to make a guess plus 15 seconds, I've had one 5k score. That was in a small town in Greenland. There aren't that many to choose from, and they're pretty much all on the southern coast. 

22500 and above is Gold. 15000 and above is silver. 5000 and above is bronze. so an average of 1000 per round gets bronze, 3k gets silver 4.5k gets gold, and 5k gets the top award,

My total score was a solid bronze, pushing towards silver - 12719 points. One terrible guess, two mediocre guesses and two good guesses. For NMPZ, I'm happy with that.

Entrepreneur

My previous post ended with "I, like Ren, am finally starting to stand on my own two feet."

So, what to do with this new zest for life? Well...

 Some readers might have seen me play guitar. I have improved a lot in the last few years by the way. Being the ruggedly individualistic person that i am, I don't play it like almost anyone else.

There are precedents. Most notably the late Canadian rock and blues player, Jeff Healey. Also check out something called "Air Tap" by a guitarist called Eric Mongrain.

But what none of these rare antecedents thought to do was to design an instrument specifically designed and optimised around this method of playing. This I have done, and it looks like it may be going beyond just an idea to a reality. 

I worked extensively with ChatGPT to flesh out my idea, and this shows both the strengths and weaknesses of Large Language Model AI systems. It readily understood what I was trying to do, and its enthusiastic encouragement helped motivate me to develop the concept further.

I had to tell it to dial the flattery right back, because when it was telling me that even my shit ideas were wonderful, it became more of an irritant than a helpful companion.

But yes. I chatted with it. I filled its ears with my big idea, in increasingly granular detail. And it understood it all. The problem started when I asked it to render an image of what I had in mind. I wanted something better than a scruffy pencil sketch to take to a luthier but what it generated was a slightly oddly shaped guitar with none of the features it had verbally parsed so brilliantly. Time and again I asked. Over and over it came up with guitars, rather than my new instrument. Since my new conceptual instrument exists nowhere in real life, it had no data to work with. Here's an example:

  

I gave up on that side of things in the end. I tried to make something using a 3d CAD tool called Blender but beyond a very basic and inaccurate rendering, I found I lacked the skills and experience to really get to grips with it. The image at the top of this post is what I came up with. It was made small and indistinct partly to represent that it was still embryonic. I planned to make bigger and clearer renderings as things progressed.

I was worried about patenting, and of someone stealing my idea. ChatGPT told me all about the different levels of intellectual property protection, but this side of things bores and confuses me. The details of the instrument will not be shared in this post because currently no such protections are in place. That's another and more important reason why the pic shows only an indistinct representation of my idea.

That's twice in two paragraphs that I've acknowleged  my limitations. I found Blender hard, and gave up. I find patenting law hard, and I don't want to deal with it. And that's alright. Nobody is good at everything. So I need to work with people that know how to use CAD and write legally watertight contracts. I can't do it all on my own.

Hermit Me would never get this new thing out of the starting blocks but new Me has set things in motion.

My original thought was to contact a luthier, and get them to build this thing, but an idea that I came up with that ChatGPT really liked was to work with my local further eduction college. Let their Art and design students work on the CAD. Let the business studies students deal with the setting up a small business side of things. Let the woodworking department take a break from joists and cabinets to craft a precise and complex piece of musical gear. And so on. Turn my shortcomings into a learning opportunity for others, while removing a great burden from my shoulders. 

So today, midway through October, I paid the college a visit. I spoke to security/reception, and they advised me to send an email. An hour or so later I got a response saying "Your email has been forwarded to the head of our Arts Department with a request that you are contacted."

Presumably things will have moved forwards by the time I publish this on November 1st.

The roadmap? To get a fully functional prototype designed and built. To set up a small business partnership with the college to market and sell this thing. 

 

Except... It's now the end of the month.  They didn't get back to me. I've sent a follow up email and have been informed that they will now get back to me next week. This week has been half term.

And if they don't? Well there are other routes. First step would be to employ a luthier to create a one off prototype. Then I would have the instrument I want at least. There is a company near to me that makes music videos, and with a decent video using a quality instrument, I would then have something concrete to take to collaborators. 

By the time I post on December 1st, hopefully things will have moved forwards. Watch this space. 

 

Renegade

Fifteen months ago I started therapy.

It was hard. Confronting things that I had buried for many years was hard, and it took several sessions to even start to open up.

 

I came out of the other side of this knowing myself a whole lot better. I've spent the last 9 or 10 months processing.

 

The therapist diagnosed me with PTSD.

 

I'll go into the cause of the trauma some other time perhaps, but for now, let's just say that at some point, decades ago, I built a wall around myself. I walled myself off not just from everyone else but also from myself.

 

Therapy did not in itself seem to change anything, but what was I expecting? For the therapist to wave a magic wand? For things to happen overnight? What follows couldn't have occurred without it.

 

Two days ago, something happened. A music reaction video popped up on my youtube page. 

Not anyone I'd ever heard of, but I started watching it and was quickly drawn in by what I was seeing. 

I've spent a lot of the last 48 hours binge watching other reaction and analysis videos of the same song, and one other song that is three songs. A musical triptych. 

 

The reactors and analysers are musicians, cinematographers, youth justice advocates, psychoanalysts, therapists, composers and songwriters, voice and acting coaches... 

 

Each reviewer had their own take, and by looking at these two videos, each with its own applicability to my situation and history, multiple times through the eyes of such various people, I felt something that I'd sometimes glimpsed during therapy fall into place.

 

Here is the first video they were reacting to: 

 

 

 

What struck me so profoundly was the monologue at the end, where Ren described his struggles as a dance, rather than a war to be won. Somehow framing it in this way made things much simpler. 

Watch the videos. Take the time to learn the back story of the artist, and you will perhaps understand why. 

 

The detailed, introspective post I've been struggling to motivate myself to write for the last two months isn't needed, and would anyway be a depressing, turgid affair. 

 

 

This is the second video. 

 

 

 

This one made me realise I'm not as singular as I thought. Of the three characters in the story, the one I relate to is Violet. Someone once described me as "The man with no opinions" 

 

Unlike Violet, I was not sexually abused by my father. 

 

Because of the trauma, I have made many mistakes, and have carried a burden of guilt and self hatred for a long time, but it feels like there is a crack in the wall at last, and a big broad chink of sunlight is coming through.

 

I, like Ren, am finally starting to stand on my own two feet. 

  

Hearts and minds

I planned to start blogging again, but with a different format. One long form post on the first day of each month.

Yesterday though my Dad was rushed into hospital because his heart was only beating 32 times per minute.

I knew what my post today was going to be. I had the topic, the structure, many of the actual words, but my attention has been diverted.

I visited him this evening. He's ok. For now. He will be 86 in October if he makes it that far. He will be fitted with a pacemaker, and could live for a good few years yet perhaps. The old fucker seemed indestructible, and this is only the third time in 85 years that he has spent a night in a hospital.

We plan golf.

Anyway, today was going to be a big long mainly coherent post about... Well something important to me.

It's now tomorrow, and I have missed my self imposed deadline.

So here instead is a seemingly unconnected video that is sort of relevant. The issues I wanted to deal with aren't going away, and a month makes no difference. See you on the 1st of September.