BobBigMac

There is a particular kind of annoyance modern life has become very good at producing:

You walk into your local GP and the receptionist tells you to use the website, then blank-faces you when you say "but I'm here, stood right in front of you". You phone because you want to speak to someone and a recorded voice tells you the quickest way to get help is online, but that doesn't work out of office hours, and you start work before they do. You check the complaints form and that's offline until morning too, and it doesn't load properly on your phone anyway. It delivers some names of various departmental-sounding things and tells you to ring the ombudsman or something, but who has time for that?

You go into McDonald’s and order from a screen… Tesco too. The cinema. Parking. Banking. Half the country now seems to have decided that the public should do its admin for it.

Taken separately, most of these things are merely irritating for most people. Sometimes you just don't want to have a natter and don't feel like pretending to be a real person after a long day, and a robot is fine. But when they become the texture of every day. They are how a country comes to feel less hospitable, even if nobody can quite point to the moment when it changed because it was taken in small bits that each seemed like progress.

The same thing is happening outside the screens. Pubs close. High streets thin out. Banks and post offices vanish, or become a hole in the wall and a drop-box. Homelessness becomes harder not to see in public places, and harder to see in private spots because they're not even allowed to have tents now.

Places that used to contain a bit of social life get stripped down to the transactions that justify their rent and rates, and whatever other wealth-extraction systems they are hooked into. You can spend a day in public now and have almost no reason to speak to anyone you do not already know.

And I wonder if we have been slightly misreading one of the big political stories of the moment.

What kind of reform do people actually want?

The standard account of Reform is still that it speaks for the poorest people, abandoned by everyone else, now taking their revenge through immigration politics. I do not think that is right. Reform is running through the entire country, or at least everyone except the very richest and the very poorest. The poorest are still more likely to vote Labour or Green, the richest Lib-dem.

The “left behind” story survives partly because it is tidy, and partly because it puts the ugliness of angry stupid povvos somewhere far away from the relatively comfortable instititutional top-school journos, analysts, pundits and commentators who have the column inches and podcasts about it. It makes council estates and mining towns hypothetical, a blip even.

More people than ever are simply not voting. It's easy to blame distrust of politicians or lack of faith in the political system, and then never pointing at the underwhelming turnout in absolute terms.

The other easy account is that Reform voters simply hate foreigners or multi-culturalism. Some do, obviously (our own data shows a clear ethnic gap) but its hard to believe that almost half of the country (representatively) are outright racists.

That's not much of an explanation for why this anti-politics has such appeal now, or why it attaches itself so easily to almost every other complaint people have about the country.

They're angry about a lot of things

People are angry about different things, and there is no poll question for whether being told to download an app has gradually made you more receptive to Nigel Farage. But my pet theory is anger about immigration is really anger about change simpler than politics or culture, and especially about being forced through systems that no longer feel made for human beings.

The Manchester pattern isn't flat, but it's certainly not only the poor.

The NHS is a good example. It has coped with a ~20% larger population than it had twenty years ago, in part because it has become more efficient. Triage, pathways, online forms, remote appointments. From the system’s point of view, these are ways of getting more people through with the same reosurces.

From the patient’s point of view, that can feel like being processed. It is perfectly possible for a service to work better on paper, but for the experience of everyone in the system to get worse.

Tension and release

People see longer waits, thinner services, fewer staffed counters/tills, a town centre that looks worse than it did, and they reach for the explanation nearest to hand. Too many people. Too much pressure. "The country is full up", apparently. "It must be that, that's what my uncle says, and my grandpa was always worried about this stuff taking over, maybe he was right after all".

It is a very available story because it takes a mess of things that are difficult to name, budget cuts, automation, management culture, rising rents, chasing gigs that get worse paid if you don't take the hardest jobs, and gives them a single visible cause, and a party with loads of money telling us all they have the answers… just keep the foreigners out, "voila", oops, sorry "ee-arr".

Imigrants are visible enough to blame, poor enough not to fight back effectively, and politically convenient for men who would rather no one spent too long asking who actually benefits from efficiency. It's not about the touchscreens themselves (which can cost more than they save in staff costs) but about the systems that decided it was better not to have a human there, it was less risky, more manageable.

'Immigration' is the proxy, the vehicle for the rejection of modern society, not just the rejection of 'globalisation' (whichever way you interpret that term, sensibly or in a meets round the back kind of way) that brings multi-culturalism, and with it… potential conflict, change and fear of change (and fear of fear of change, if you like)

Which is why the usual response to Reform feels so inadequate. If the whole explanation is that its voters are selfish, or stupid, or racist, then there is nothing to do except hate them back. “Everyone is a dick now” is not a theory of the country. It is what you say when you have given up fixing the real problems.

So what?

A more useful politics would start with the part of the complaint that is real. People do not like being handled by systems. They do not like being told the human route has been removed for their own convenience. They do not like seeing places decay around them while every institution insists it has become more efficient, and technically they're handling more of your problems now.

Some people (not all, duh) miss pubs, counters, receptionists, joining in for footy in the park, the small amount of slack that lets a person help another person instead of referring them to a process, or just making everyone someone else's problem.

You do not answer nativism by pretending those losses aren't real… "bUT We wOn iN a lAnDsLidE?" represents blind hope that people will forget and move on before the next big one, the one that matters. Eventually, someone has to start fixing systems by humanising them, not just "spend money on stuff people can see" because that clearly isn't enough, they're already doing that (and doing a pretty good job in some places), but most people don't give a fuck, because that's not the problem…

If we doubled immigration, their contributions to the economy would cover the cost of wealth extraction by billionaires.

We just hit a trillion pounds lost from the British economy due to Brexit.

If we only consider Treasury income (tax we could have spent, or be spending, on public services), it’s currently running about £65–90bn/year short.

For comparison, the cost to the Treasury and councils (if you bucket in everything you conceivably can consider a cost related to them) of immigration to the UK finances is under £10bn/year (and immigrants are net positive on taxes over time anyway, so we actually get a lot more back for that £10bn).

If you prefer to compare it to other things…

£106bn/yr paid for interest on national debt £60bn/yr on pension tax relief £45bn/yr “tax gap” (nice euphemism; I’d argue this should be MUCH higher, but we don’t include Amazon and many large dodgers) £12bn/yr to private landlords via housing support

The lost annual receipts break down to about £1.25–£1.73bn/week (call it ~£1.5bn/week if you want a single number). If you assume broadly similar spending shares, proportioned for healthcare, that puts the NHS DOWN roughly ~£300m/week attributable to Brexit.

Well done lads.

And yes: I can write it on a bus if it helps.

Over the last 18 months or so, I’ve been building “vibe”-based projects with AI, across many models and approaches, as discussed on this blog.

I’ve used it to write lots of code, some fiction, some technical books, and this blog. I’ve also been testing the limits with smut and other fringe content. I’ve generated a load of pictures across various projects, had it help process large data and subjective info (turning complex content into readable reports), written and voiced audiobooks, and used it as part of various toolchains to make videos through automation tools like ffmpeg - because gen AI video just isn’t really “there” yet.

I have written several times about finding value in AI. As a consumer tool, it definitely has strengths: as a search/research tool, it’s excellent, with the usual caveats.

And yet I keep wondering if it’s worth it.

So many of my projects went together quickly, but then almost immediately stalled. I’ve been thinking about that during my break, and this is what I’m taking into 2026.

Why most of those projects went nowhere

Most of those projects ultimately failed on the same few problems.

1) The output is usually functional, but rarely usable

The quality of LLM-generated everything is just always pretty poor.

It’s often - perhaps usually - functional, but it’s never actually usable for much more than basic practical utility. And almost always, it deeply misunderstands the prompt, no matter how sophisticated your prompts are.

Worse, it tends to misunderstand you in a way that’s hard to parse if you’re reading casually. The shovelling-bullshit problem.

2) Doing lots of work with it is SLOWWWWW, because you have to babysit it

Yes, it can fill out a load of boilerplate. Yes, it can work its way around some pretty deep technical issues.

But every time the context drifts from the core goal, it walks further and further from what you actually want. So if you want to trust the work at all, you constantly have to babysit it.

The platforms do a good job preventing egregious wrongness, but they’ve done very little to fix casual wrongness and rote outputs. I keep wondering if that’s a fundamental limitation of LLMs.

3) Creative text is “sometimes lucky”, but mostly a thick soup

LLMs generating creative text can sometimes get lucky.

But after listening to a few dozen hours of generated story text, it’s always a thick soup. Anything you want to generate in bulk comes out sounding the same, over and over.

If you focus on ever smaller areas of context to try to fix that, it just means you have to do more and more of the work yourself.

So the day-to-day feel of working with AI becomes a bit like a co-worker who always lets you down just a little bit. The first few times it’s fine. After a dozen requests trying to communicate some simple but novel idea, it starts to sting - even burn.

And that’s before you hit the fun failure mode where the codebase gets entirely overhauled twice because it got distracted and you didn’t catch it in the review tools.

4) Bigger context helps, but “AI as the primary worker” still doesn’t fit how software works

Model context sizes have grown a lot since this became part of my core dev process, and it has slightly improved. (GPT5.2 is pretty good atm.)

But projects still stall for the same reason: software is complicated, and modern software tools are built for a distributed team. Cursor is trying to update its UI, but nobody really knows yet how to make software with AI as the primary worker.

We have to figure it out, because updating generated code manually is a real SLOG. It’s usually easier to throw away bad work than it is to fix/improve it.

Smaller updates work better - assist, not agent-mode stuff - but even then I find I usually know what code I want more than I know how to describe that code in enough detail for it to be implemented correctly. My brain knows what assumptions I’m making. The model doesn’t.

Conclusion: the pint glass problem

I still think AI has huge potential. It’s obviously doing something - it can write a lot, it can code a lot, it can summarise and transform and remix and accelerate. It can make a project feel “real” in an afternoon.

But my lived reality of it is the pint glass problem: it’s like always pouring half the remaining volume into a pint glass. You can get close, fast, and you can keep getting closer… but it never actually becomes a full pint.

I’ve heard plenty of devs and engineers say it genuinely helps in specific use cases, and I believe them. I can see it in my own work too, in little pockets: research, boilerplate, quick transforms, rough drafts, sanity checks, “get me unstuck” moments.

What I can’t get it to do - not reliably, not without turning the whole thing into a babysitting job - is deliver a finished product that’s actually worth the time. The last mile is still a swamp, and the more ambitious the project is, the more that last mile expands to eat the whole map.

I don’t think the fix is “better prompts”, and I’m not convinced it’s only “bigger context” either. It feels like we’re missing a layer: a new UI/UX for building products with an AI assistant, across the entire workflow. A new ecosystem for modules that provide utility, not 'just' interfaces, so the model can offload the context that comes with two hundred lines of lib usage code that truncates the code that actually matters.

Not a UI/tool that’s obsessed with code, but one that’s obsessed with intent. What the developer wants, what the user needs, what “done” actually looks like, and how to prove it. Probably closer to the transition from dev to user - a prosumer / pseudocoder mode - like UGC, but baked into the bones of the model instead of bolted on as a chat box next to a diff.

I don't normally get into this subject, because it's really nothing to do with me, but I've recently heard a few people say something like "You can't even use pronouns anymore", and for their benefit, I felt I could write this post, something I have no doubt that some of my more 'woke' connections will find doesn't agree entirely with accepted dogma…

My Stance: Respect Without Ideology

I don't buy into 'trans ideology' and probably count as 'gender critical', since I don't believe "a trans woman is a woman" literally, but I don't think I should need to buy into any false equivalences to support trans rights, without hesitation. I support the right for a trans person to be treated wherever possible with the label they prefer, simply because it's more human than to separate them or even worse, to mistreat them.

Trans people should be able to live as whatever gender they want, and be treated with respect around that decision, because I think treating people with respect comes far ahead of being pedantic about what means what, and simply because I don't need to make the rules, to feel like a man. In fact, I think that being a man sometimes means defending people whether I agree with them or not.

The “you can’t say anything anymore” crowd are trying to make manners feel like tyranny so they never have to talk about rights at all. Supporting trans folk isn't really about whether you care who has a cock or not, but about understanding that people come before language.

How Trans Rights Benefit Everyone (Including Men)

There's also the issue of how some trans rights create men's rights…

Not just in the sense of a man or woman being free to choose their gender (or not be referred/defined to it by a system they don't like, in the case of non-binary people, for example), but in simple things like unisex toilets having better support for men too, services like changing tables (I've had to watch my Dad change my sisters' nappies in the back of his car, and on a shelf, when alone with all the kids) and nappy disposal, or simple access to clean toilets, something very rarely afforded men in sports venues, bars, or other common places.

As an ulcerative colitis sufferer, a clean toilet often proves elusive, and unisex toilets mean disabled or not, male or female, you have a better chance of the relief of taking a shit in a room with a door and some peace and quiet, not in a disgusting glory-hole riddled trough (less common nowadays, but it still happens).

Then we get onto support for men more broadly in society… Trans-exclusionary policies like segregated refuges or segregated hospitals often sound reasonable, but for small towns or cottage hospitals, that's really an increased cost for the facilities or the council, and trying to accommodate trans patients or escapees in single rooms sometimes just isn't possible, so they can end up being left aside to suffer, often with no help.

This applies to mental healthcare too, employment rights, somewhere safe to live, marriage rights, and travel, access to public spaces like gyms or clubs and services. Even a legal identity that matches who you feel you are, so your legal self isn't alienated from your real self. I like to think of it like how when I chose to move to Manchester, I chose to become Mancunian, but let's say in future I have a heart attack, but before they'll treat me, I have to be shown that I'm not a 'born' Mancunian, and even have to denounce my choice, before I can be treated.

It's worth remembering that trans people, men and women, are much more often the victims of violence or related issues, than they are the perpetrators, and in many places, more likely to be victimised and mistreated by the systems supposed to protect them. Strong/supportive refuges for trans people also lead to better refuges for beaten or raped men too. Nobody likes to talk about how an attitude that favours treating everyone, can also help men who find themselves in difficult situations, especially those who find it hard to even recognise they've been mistreated because their own "be strong" mindset means they deny anyone has hurt them at all.

Fears vs Reality

Much like immigrants, trans people aren't the enemy, they're a useful tool, a simple lever, used by billionaires, the right-wing, toxic leaders, and even more 'reasonable' sources like the BBC and the New Statesman, as weapons against everyone. They're an easy/soft target to point at and say "look at those weirdos, blame them, not us, the billionaires". Fuck that. They're not even expensive to society financially, it costs so little to provide trans healthcare or support, partly because there's so few of them, you have to dig pretty deep in the budget before you even find the costs.

The fear some women have about trans women especially, is understandable, human even, but it's an emotional 'fear', not a reality. The statistics just don't bear out that fear.. It's the same reason we don't legislate against pigs flying or lead turning into gold. You account for fears through transparency and understanding, compassion and personal growth, not further division and kneejerkism that turns the weak into our enemies.

As for men's rights, the trans-exclusionary position also treats 'men/males as aggressor' and 'women/females as victim', and whether you believe that as a man or a woman, that's unhelpful for a start, an oversimplification at best, and an attitude that directly harms men by treating them as predators and girls/women as prey. As a man, albeit a progressive one, I think basing your decisions on the assumption that men are monsters and women will be made to suffer, isn't just harmful to men, but to women and girls too. I've seen enough men be mistreated by their female partners, to know "not all men" is more than just a stupid slogan, it's a reality for some men who suffer because they believe they're all born bad anyway.

Likewise, I don't think males and females should be forced to compete in ALL sports, despite there being various sports where girls and boys do typically compete together (Baseball or soccer, in the US, say). I also like that Gladiators recently switched to referring to sex, not gender, because it eliminates the problem of physical differences in an environment where it clearly matters (like Wrestling or Boxing).

I don't like intersectionalism or identity politics at all, really. They do more to separate us than they do to bring us together, when most people are pretty much the same in their own unique ways, especially at the bottom, where people need the most support. We just want to live our lives and be as happy as our situations allow.

I keep coming back to the same uncomfortable thought about generative AI (LLMs and images): it doesn't really matter if it's crap 90% of the time.

If it's cheaper (and maybe faster) to hire ten people to put prompts into a model a hundred times - 1000 monkeys at 1000 typewriters, eventually producing Shakespeare - than it is to train, hire, employ, and engage a “real artist” to do the work manually… then that's the model that wins.

Even if the artist would “get it right first time”, which they usually don't, because real work takes iteration too.

Creation becomes curation

GenAI feels like the perfect extension of the production-line, supply-chain-obsessed modern production methodology.

It's better than the old factory line in one key way: digital results are entirely disposable. The only limit is how much energy you can afford to waste before you get something useful out.

So the task of creation becomes a task of curation.

And even that can sometimes (tho so far, rarely) be automated by other ML models, or by crowd-filtering systems that just shove the sorting work out to “the cloud”.

Evaluation cost isn't magic, it's just outsourced

Evaluation cost can be mitigated by cloud filtering, where people work for much, much, much less pay than the value of the work they're evaluating (like a penny per check).

Because turk-style workers don't need to earn a living if all they want is beer money and something to do to feel productive. Heck, many of them never even cash out.

If your whole production plan is “generate a mountain of slop, then pay the smallest possible amount to find the few good ones”, you can absolutely do that. It's grim, but it's not complicated.

“Approvals eat the speed gains” - mate, approvals are the system

The line about “they'll wrap it in approvals, which eats the speed gains” is already a thing.

I worked for the NHS where me and three developers/engineers had at least twelve managers, maybe more - I honestly lost count. If your systems already rely on pretending that managing things is working on things, you're already in this space, with or without genai.

This is just the same model, except now the output can be generated in bulk, and the approvals layer gets to feel even more justified.

Why one-image-at-a-time kills prototyping

Part of the way DALL-E in ChatGPT fucked up is switching to generating only one image at a time.

It saves cycles/resources on the server, but it kills it as a prottoyping tool for the user. Grok (and other stable-diffusion style engines) generating four images at a time gives you more options to work with.

Dial that up to 100, distribute the evals, and you get even more chances to be lucky, not good.

This is where the qhole quantum cat in a box thing keeps popping into my head: it doesn't matter if it's wrong/bad/crap/unusable a million times, as long as you can afford to waste that much, because you only need it to be accidentally perfect once.

Code is the awkward one

Software dev “multiple agents” isn't practical yet.

It's much harder to evaluate (except in specific testable situations), and even then someone still has to check if the tests meet the requirements too - you're often just moving the problem. Images are fast to judge. Text is slower, but still human-readable. Code is deceptive: it can look right and be wrong in ways you don't notice until it matters.

So yeah, this model hits images and marketing copy before it hits “real software engineering”, even if everyone keeps trying to will it into being.

The revolution is taking time (and it's mostly throughput)

The “promised revolution” is taking time to come because existing data centres aren't capable of this kind of throughput, at least not at the cost-level that lets enough people play with entirely new ways of producing content.

If companies actually want new behaviours - like “generate 1000 variants and curate the winners” becoming normal - they need the compute to become cheap enough that wasting it doesn't feel like setting money (or the environment) on fire.

Music might be the closest to the “just ship it and see what lands” model: it's more A/B test friendly, where Spotify already push out slop and see what lands vs what flops.

What does this already, and what's next

Already: anything where “close enough” is fine, variation is valuable, and the failure cost is low. Ads, thumbnails, product images, content marketing, basic design production, endless internal comms, low-stakes copy, generic visuals.

Next: anywhere with huge volumes of repetitive drafting and checking, where nobody actually wants the work but everyone wants the outcome. Compliance paperwork. Procurement/RFP sludge. Contract churn. HR process comms. Customer support triage. “Make it look like we did due diligence” admin. All the places where the real output is the appearance of a process.

I read an article a few months back complaining about the impending “governance apocalypse”, where people who have never had access to the systems of power can now use AI to find a way through the tyranny of bureaucracy (like how Project 2025 dug its tentacles into power, then had just enough control to achieve what it wanted), so I decided to test it, and LLMs really do unlock those doors.

The time saving on the research stages alone makes it a pro-democracy tool. It being sometimes slightly wrong (mainly if you ask the wrong questions, poorly) is still no more wrong than how wrong I’d be if I had tried to do it all myself with google sending me to the wrong places instead. I’ve fixed my pension, my health, I'm working on my GP’s accessibility issues, found the best accountant in my area, and helped my wife do the same… It cost me twenty quid a month, and that’s before I even get to work project utility where I don't even know which bit it's been most labour-saving (yeah, sometimes it writes shitty code, but try again when it's not under load, or with a codex model, and it can usually fix itself).

I was asked: “You could have done all those things without AI though right? […] AI is basically summerising web content?” and some people might be able to understand those things, sure, or at least, maybe, but me? Not a chance. I fucking hate paperwork… I cannot stress that enough, it’s pathologic, very strong authority-aversion, and if you tell me to do something, even something basic like wait in a queue or 'chill out', there's a pretty good change I will do the opposite, just for the thrill of subverting you, even at the time knowing for a fact it's a bad decision.

AI isn't just summarising individual documents, but connecting them, and decorating and putting them into words and contexts I can actually understand and action, but driven by my own decisions, not those of somebody else telling me what to do. It turned confusion into process simply by connecting the dots and exposing the reason why these complex systems are structured into such thin layers, where you need a little from every layer before any of it really starts to make sense.

I can understand fearing the market impact or the unintended, unknown or second/third order consequences, but literally anyone claiming the tech isn’t useful must be smoking crack. People complaining about how AI 'can’t do anything' might be dumb as rocks, or as lazy as rocks, at least.

I suspect most of the people with the particular complaint "it's not useful" either never used it, or only used a free tier on an old model. I hear this all the time on podcasts, from artists and creators giving bad examples of AI where it was obvious they asked a stupid question, or just didn't use the right mode, or just never bothered to change from the default (summarisation only) base model, entered a dumb question, got a dumb answer, and blamed the tool, like any shitty workman.

LLMs absolutely cannot write a great novel, they just don’t have enough output context to manage themes and callbacks and the ongoing concerns that make a novel work, and its art is soulless, by definition, it takes ideas way too literally… But the understanding of corpo documentation and workflow/access to otherwise opaque systems is really incredibly powerful when used for good rather than evil :)

The secret ingredient to getting value from whatever LLM you have is using thinking mode, demanding sources and explanations. I often find asking for data gathering first, explicitly stating you do not want a straw-man argument, then asking for opinions/conclusions based off that data, and asking sane, well thought-out questions, like a sensible person, not trying to catch it out or cherry picking funny failures like a petulant child.

Had a dig around on the tax changes, not just from the budget, but system tweaks since Labour took over, and I'm actually pretty impressed... It often feels like mostly digging around behind the sofa for change, but most changes are about making sure that comfortable people pay what they already owe.

The bigger issue for the cash-rich but time-poor (like dividend-takers or middle-aged people who rent out their flat bought for a few grand in the 90s) is that most of the changes mean more paperwork AND paying tax they should have been paying the entire time the Tories were in power, but hadn't bothered to, because it was easy to forget and there were no consequences for not trying.

A bunch of think tank and “research” pieces say people (Greens especially) are pissed off with those procedural changes, mainly cos they catch the tiny “real” rate increases first, and still can't afford to buy homes that don't exist yet, but the poor people who are struggling will be better off.

The motto for this govt shouldn't be “change you can believe in”, but instead “making people pay what they already owe is good for everyone, actually” ;) I hope they manage to survive the cull we're looking at come May, cos I want to see if they still scrabble in the dirt, or if long-termism works.

Also, we've all been saying “close the loopholes” for most of my life, but now they're finally doing it, all the billionaire-owned press wants to do is complain about how it's supposedly bad for the little guy, with increasingly unhinged journeys round the houses to explain why.

Is it right-wing or left-wing to want people to have a right to NOT work if they just don’t want to?

I heard a great position from a 1950s economist about how the unemployed are a positive effect of a flexible workplace/economy, that paying “welfare” is an inherent cost of changeable markets. Essentially, some unemployment isn’t just inevitable, but “good, actually”, and having them not starve to death in the streets isn’t just good practically, but economically, because then you have people strong and skilled enough to work when new work they do want, does materialise.

That's a nuanced way of looking at unemployment that I’ve literally never seen in ~35 years of reading the news and almost as long reading the economic press: unemployed workers as potentially responsive opportunists, just waiting for their time to discover something new they will love. Social potential, not a problem to be stamped out.

Do you really want to starve our national social potential, just so the media classes and protestants feel like everyone is being forced to work doing shit they hate for almost no money? Thinking about this because of Labour forcing young people to work for fuck all benefits, like Tory scum.

It feels like forcing kids/YAs to work doing bullshit jobs is just eating up time they could have been spending developing businesses, even if that new business is streaming Fortnite to 200 followers, growing their Vinted, or shilling nudes for £300 a month, just because it’s not a real™ job. That’s still experimentation, still learning, still building something that might actually fit them and maybe even grow.

We’ve been told for decades that if the economy grows, everybody wins. Politicians love the idea because it sounds like a tide that lifts all boats, something simple and optimistic to sell. Britain’s economy hasn’t worked that way in a long time.

Growth has carried on, fortunes at the top have swelled, and yet millions still find themselves stuck on the margins, working harder for less security. So when Starmer leans on “growth” as the big solution, it’s worth asking: growth for whom?

Starmer and Reeves are putting an awful lot of faith in “growth” without fixing (or even mitigating) the simple fact that almost half of all British wealth sits with the richest ten percent of households.

The same dataset shows the top 1% alone hold about the same as the bottom 50% combined. So we could see 8% growth (unlikely, but imagine it), and it still wouldn’t matter, because the half that ordinary people end up seeing would still be swallowed by inflation. Growth without distribution is like trying to fill a colander, it washes straight back out into private wealth silos that do nothing more than make numbers go up.

I’m not even convinced that the few tax rises on the table would come close to the redistribution needed to pull the floor up. Inheritance tax, capital gains tweaks, windfall levies—these are modest sums by Treasury standards. The Institute for Fiscal Studies estimates that even an “ambitious but plausible” mix might raise twenty billion or so, which sounds big until you stack it against the scale of poverty. Useful, sure, but nowhere near enough on its own.

That poverty is not marginal. Roughly 21% of the population (14.3 million people) live below the poverty line by the government’s relative measure. It’s not even classical “unemployment.”, since many of the poorest are juggling full-time jobs with barely part-time pay, gig shifts, or hustles. So they can’t just switch into apprenticeships or internships, not after fifteen years of no Plan B leaving them with nowt in the bank to lean on. The idea that growth in professional sectors, engineering, or energy magically lifts them into an advanced economy ignores the gulf between where they are and where those gains happen.

Which makes it all the more awkward that current policy is leaning so heavily on private finance rather than public spend. The new National Wealth Fund is pitched as a way to “crowd in” private capital, to partner with big investors rather than directly fund projects. There’s nothing inherently wrong with that, but past experiments with private finance have been patchy at best, and without major state spending on retraining, relocation help, or simply lifting incomes, the promise of growth risks looking good on paper, but still leaving people being (and feeling) no better off come the next general election.

The average LLM is wrong much less often than the average social media user. Using any LLM to test your opinion is probably better than getting your facts from social media and then repeating them like they’re reality. Even Grok is less full of bullshit than the average feed because we are so eager to believe any claim that backs up whatever we already believe, to hit repost before we think, to confuse the warm glow of agreement with the cold labour of being right.

Most models have a greater breadth of sources than you or I can get from scrolling a feed, and when they glide past a fact that cuts against their flow, they can still tell you about it—if you bother to ask—which is more than you can say for a lot of us who never even see disagreeable facts in the first place.

I’m not cheerleading machines, I’m asking you to do a tiny bit of thinking before you share, because thinking is the part that gets dropped when the vibes feel good and the timeline is moving and outrage pays in little snippets of attention and approval from our peers.

Sharing posts uncritically is eating rocks. It’s the same trick every time: cherry‑pick the worst example or the boldest claim that sounds right, jam it into your worldview, swallow, declare yourself nourished, get a pat on the back. You feel full for five minutes, those tasty little rocks settling in your warm belly.

Models aren't perfect, neither are people

Models absolutely have their own failures—confident nonsense, guardrails that get in the way, cheap takes under load that go soft around the edges—but they can’t help chewing. Token by token they have to do the boring bit, and that alone is often more thinking than the average vibes‑based repost, which is why don’t mistake feeling right for being right belongs taped to the top of every app with a share button.

The world has too many devils; it doesn’t need more advocates. The devil wins when you don’t think at all, when you let speed stand in for sense, when you let your tribe do your homework. So try this: before you hit repost, make an angel’s advocate out of the nearest model and ask it to take your favourite opinion and say it back without the swagger, then make it argue against itself without the snark, then ask what actual facts both sides are hanging on (if any) and which of those facts you can check in the next minute; ask it—in thinking or research mode—to cite everything the claim relies on.

Ask what harm you could do if you’re wrong. Ask what would change its mind—then ask yourself the same thing. If you’ve got another model to hand, run the same question again and spot where they disagree, not because either one is truth but because the seam between them is where your thinking starts.

The model actually does the thinking step you skip. It chews, enumerates, compares, admits uncertainty or assumptions when pushed, and can show its working on demand. You can do all of that too, but if you choose not to, in that moment the model is smarter than you. Full stop. Your feed is tuned to tell you what you already are, and what you already are is very sure of itself.

Those viral “AI told us to eat rocks” moments? Mostly older models, bad prompts, bad engineering—cherry‑picked for comedy and rage, then reheated as if they were the current baseline. Edge cases aren’t the average case, and the average case today isn’t the average case from last year. Some AI will always be worse than the best human mind, but most people (myself included) are working with way less than that, even on a good day, so take the help where it makes sense, and learn when to reject it for failing to live up to a higher standard.

Not a spectrum issue

None of this is left vs right. It’s lazy‑think vs think. Everyone gets stupid the second the take lines up with their identity, and if you catch yourself calling strangers stupid while sharing something equally stupid because it agrees with you, congratulations, you’ve just eaten a rock. The fix isn’t worshipping AI; the fix is process. Interrogate any advisor—human, feed, or model. Make that your default posture: ask more questions than you answer.

Timelines reward heat over light. You know this. The feedback loop is outrage, identity, speed; thought is a drag on the machine, and the machine hates drag. There’s nothing that will get you down‑voted faster than a nuanced position in a toxic‑positivity thread, or a positive point in a hate‑orgy. Models aren’t saints, but they will slow you down enough to enumerate claims, surface counters, and expose assumptions if you ask them like you mean it.

So don’t eat rocks, even ones made up by real people. Take sixty seconds. Make a model your angel’s advocate. Look for counter‑facts you can actually check. If you find the claim holds up, post—go wild with it, lean on it like it will hold you up—because those are what we have left once the world is done tearing down the walls between outrage and reality.

I’m replying to this take from Maggie Chapman, plus a bunch of almost‑identical posts I saw yesterday: claiming Labour’s digital ID plan is “the same as” Tory Covid contracts. If you believe that, you’re either delusional, or you’re selling something. It’s like comparing lighting a match to being on fire. Meaningless without denominators.

The denominator problem (and the actual numbers)

Covid era waste/overspend people actually remember:

PPE write‑offs: ~£9.9bn VIP‑lane PPE: ~£3.8bn NHS Test & Trace (spent): ~£29bn Bounce Back Loans (fraud/error): ~£1.9bn Local authority covid grants (fraud/error): ~£1.1bn

Digital ID (UK, current pitches):

Estimated build envelope: ~£0.4–1.0bn over rollout (a trade body says “>£2bn”, which reads like lobbying, but it's possible)

That’s not the same scale. Not even close.

So… does digital ID actually solve any real problems?

Yep. In countries that implemented it well, digital ID isn’t theoretical:

Legally‑binding e‑signatures & paperless admin: Estonia’s norm. Company setup, tax filing, contracts… no faff.

Cheaper, faster bank know-your-customer: India’s Aadhaar e‑KYC slashed onboarding costs and time.

Fewer ghost‑benefit payouts: India’s direct‑to‑ID welfare reduced leakage at scale.

One login for public and private: Denmark’s MitID/NemLog‑in is how people do tax, health, and banking.

E‑prescriptions by default, pick‑up abroad: Portugal/EU.

Valid travel document across EU/EEA: national ID cards, intra‑EU, and safe backup-plans for lost/damaged travel documents.

Rapid crisis cash: Togo used digital identity rails to pay people quickly during Covid.

Remote company management & e‑residency: Estonia again—run a company from anywhere.

This is all boring, practical stuff that reduces friction and cost. Which, funnily enough, is what good public infra is meant to do.

The integrity bit

If you want to argue policy design, great—let’s talk scope creep, privacy, guard‑rails, portability, open standards, independent audits, and sunset clauses. Worth doing.

I don’t even love ID cards. But they do give government a lever that could help with things people bucket as an “immigration problem” — even if that’s being widely, massively misrepresented as a bigger problem than it really is, for easy/stupid political gains.

But pretending the cost of a UK digital ID rollout sits in the same ballpark as the Covid‑era bonfire is unserious. Any campaign leaning on that comparison is either misinformed or hoping you are stupid enough to fall for it.

Pretending everything is the same when it isn’t is far more damaging than actually talking about reality, because lazy lies let liars keep winning only if you keep believing it.

TL;DR (for lazy skimmers)

Covid procurement/support errors were double‑digit billions.

A UK digital ID rollout is sub‑billion to low‑billion, depending on scope.

Other countries have already shown clear, repeatable benefits.

Argue the details. Don’t launder vibes as facts just because you're scared of the slippery slope that only happens if you keeping sharing misinformed takes as if they have relevance to reality.

Dwarf Fortress is an incredible game, but it's obtuse, hard to understand much of anything really, but it will absolutely suck you into the mountainhome if you let it…

Once I built a dungeon to keep captured goblins, deep in the basement, but I ended up turning it into a safe room where all my dwarves could hide from the passing hordes.

That worked fine until I failed to notice one of my citizens was a were-beast, so when I locked the whole fortress away during a siege, I didn’t realise the were-beast found their moment… The walls and floors were soaked in blood, like that scene from 28 Weeks Later, completely wiping out my fortress.

I had accidentally created a horror, but it still wasn’t as horrific as the more mundane systems that quietly eat away at a fortress…

The real horror of Dwarf Fortress is found in its Sims-like systems that let the player create either a paradise or a hellscape, but the limitations of the graphics mean it looks more or less the same in either case, because there’s no real visual cue for dystopia in DF, there’s just who you are, and how you treat your dwarves. They all still look basically the same, no matter what.

They have systems for education that can support the mental health of your population, or drive division and even hatred, or turn everyone into undead minions. There are systems for immigration that can be a blessing or a burden, or you can build a beautiful Tavern, and host singers and actors, to entertain your citizens in the mists, or your tavern can turn to brawls and arguments, and still all these things look more or less the same unless you really dig deeper.

It’s the same with hospitals that either heal the wounded or quietly turn into haunted waiting rooms for the dead, justice systems that enforce fairness or spiral into beatings and executions for petty crimes, temples that hold communities together or fracture them into secret cults, military units who defend your home or drain it into endless wars chasing the dream of saving your lost warriors, or dining halls that host feasts or become the rich cultural centre of mass starvation.

The New Statesman ran an 'exclusive' interview with Andy Burnham this week, framing it as a “radical blueprint.” The actual piece, though, is more about identity than policy — long on personal anecdotes and 'lad' chat, light on interrogation. It reads more like a fanboy’s extended profile than a journalist pressing for detail.

That doesn’t mean Burnham’s ideas are empty. Strip away the waffle, and there is substance — housing, utilities, light-touch but aware on electoral reform demands, Brexit, the basics of life that set the stage for everything else. As puff pieces go, this one suffers mainly from a mountain of waffle, so here's what Burnham said throughout…

What comes across here is less a programme, more a set of vibes. Moderated socialism, framed as common sense. Call it radical against today’s backdrop, but in reality it’s more like a return to basics — and arguably that’s no bad thing.

The bigger question is appetite. After Brexit, austerity, and years of institutions visibly falling apart, how many people want careful restoration, and how many just want to tear the machinery down regardless of what replaces it? Burnham’s offer sits squarely in the former camp: repair, stabilise, roll back some of the 1980s.

And yes, vibes matter. We’ve already seen that vibes-based campaigning works. Maybe leftish vibes can hold the line against rightish ones, at least for a while. But it only lasts if the Farages of the world are forced to fully mask-off, so people see what’s really being offered before they vote.

That’s the challenge Burnham sets himself, whether or not this article managed to show it, over the next few years, Britain is going to be choosing between racism-led insanity in a rapidly-deteriorating mask… or sensible, cautious, common sense and a nudge towards greater integrity, with some ideological drivers 'from both sides'.

I keep seeing the same line about 'free speech' from people who think that literally anyone telling them they're wrong is a clampdown on their freedom.

This is actually a massive tell…

Free speech warrior: Complaining about ‘free speech’ when it’s not the government silencing you, means you don’t have a case—cos if you did, you’d cry about the facts instead of how you're allowed to say whatever you want, technically.

They might as well be openly admitting they're wrong but sticking to their guns anyway. Rights protect you from the state, not from people deciding you’re talking out of your arse.

Here's some more… Easy and obvious bullshitter red flags:

Just Asking Questions Endless questions, zero commitments. If the answers don’t move them, it was never about learning.

Sea-lioning: They ask for proof in a fake-polite way, then ignore it.

Bait-and-Switch Argument: Bold claim for clout; retreat to a harmless truism when pressed.

Goalpost Shuffle: Criteria mutate the second you meet them.

Gish Gallop: Ten weak points hurled fast so none get examined. Handle one claim at a time.

Whataboutism: Any critique triggers an unrelated grievance hose.

Tone Policing: “Your vibe invalidates your evidence.” Feelings don’t beat facts. Especially noticeable if they're thrilled that you're angry.

Timeboxing Truth: “Answer in 30s or you concede.” Reality isn’t a game show.

Endless Context Plea: Every refutation fails because “you need more context.” After a while, that’s a stall.

Referee Shopping: Lose in one venue, declare another the “real” arena.

Out-shouting: Just be louder, crazier, alone or as a legion. If you need screamers, you might not have a case.

Platform Twinning: Post on a fringe site, cite mainstream backlash as validation.

Reputational Blackmail / Platform Entitlement: “Platform me or you’re biased.” Nobody owes you reach.

SLAPP-y vibes: Threaten lawsuits instead of presenting facts.

Burden Flip / Burden Boomerang: They make the claim; you must disprove infinite variants. Nope. Claimant carries burden of proof.

Can’t Be Proven Wrong: Nothing could ever disprove it—convenient.

Kafka Trap: Disagreement = proof of guilt (you don't agree with me, you must be one of them).

Reverse burden via FOI-spam: Bureaucracy as cudgel; lack of instant docs doesn't make a conspiracy.

Falsifier Test (the fix): Ask: What would change your mind? If “nothing,” it’s belief, not argument.

Emotional Manipulation: Exaggerated risks to make you agree without checking the facts (you might die if you don't believe me).

Tricks with evidence, numbers, and sources

Anecdote = Data: One story → universal law. Sure, mate.

Hiding the Full Picture: Percentages without base rates or sample sizes. If you put it into per-capita scale, and it's basically 0, it might not matter at all (see any make-up ad).

Sampling on the Dependent Variable: Only successes/failures sampled; lesson “proved.”

Cherry Orchard: They pick only the examples that back their point and make them seem stronger together, but they’re hiding the full story.

Outlier Worship: Pick the freak datapoint; call it the rule, this one trick.

Moving Baseline / Windowing Fraud: Compare today’s blip to yesterday’s worst day; cry “crisis.”

Fake Measures: Uses a number that sounds important but doesn’t really prove their point.

Single-Study Syndrome: One paper, ignores replications and meta-analyses, claims this 'scientist' is the only one who knows the truth (think every Ancient Aliens episode).

Fake experts: Cite someone with no real credentials as an ‘expert’ to back their claim.

Clip-Mining: 8 seconds of video to “refute” a 200-page report.

Graph Cosmetics: Truncated y-axis, dual axes, smoothing to exaggerate effects, meaningless/vague labels, each bar with its own scale.

Causal Rebrand: the result feeds back into the data to feed the cycle.

Retraction Blindness: Still citing work that’s been corrected or pulled (often smears the authors of contradictions/corrections).

Source Laundering: Blog cites blog cites Substack; no primary anywhere.

PDF-ism: Link-dumping a mountain of unread papers as intimidation, not evidence. Expects you not to read them because it loosk insurmountable.

Appeal to Secrecy / Secret Evidence Flex: “If you saw the hidden info I can’t share, you’d agree.” Then it doesn’t count.

Language games and rhetorical fog

Word-Switching (Equivocation): Same word, two meanings, one slippery conclusion.

Definition Drift: Mid-thread redefinitions so they’re “technically right” by the end of the claim.

Bandwagon Effect: Everyone believes this, so it must be true.

Vague Terms: “Corruption,” “grooming,” “the elites”—never defined, always useful.

Semantic Overload: Weaponise a label so any use smears the target.

Word-Magic: Etymology as argument. Reality doesn’t care.

Map-Territory Swap: Treat the model as reality; dismiss reality as “messy.”

Hypothesis Laundering: “Just asking questions” → “widely believed” → “settled.”

Bothsides Masquerade: Balance theatre to avoid judging evidence.

Projection: The thing they accuse you of? It’s theirs.

Schrödinger’s Joke: “It was a joke—unless you agree; then it was serious.”

Dogwhistle Plausible Deniability: Followers get the message; outsiders get gaslit.

Sanctimony Shield: Moral outrage to avoid empirical checks.

Purity Tests: Any nuance is betrayal; only zeal counts.

Context Collapse: Mixes jurisdictions/populations as if identical.

Steelman Allergy: Can’t state the opponent’s best case fairly.

Confidence theatre and risk-free claims

Confidence Game: Volume and certainty standing in for data.

Retcon Prophet: Predicts after the event; swears they said it before.

No Skin in the Game: Refuses to state a testable prediction with a stake.

Prediction Evasion: Won’t put a date, metric, or threshold on it.

Stake-free Outrage / Victim-to-Victor Flip: “We’re silenced”—on six shows and a tour.

“Everyone does it like this:” Popularity isn’t evidence. Show outcomes, not vibes.

“We’ve always done it like this: Age is not validity. History isn’t proof it still works, it could have always been wrong.

Industry standard / best practice: Name the standard, and re-test it.

Common sense: If it’s so common, you can measure it. If you can’t, it’s just a feeling.

How to fight back?

Ask: “What exactly are you saying?”

Demand clear definitions for big words like “corruption.”

Request proof: “Show me the data or primary source.”

Stick to one point at a time—don’t let them overwhelm you.

Ignore tone or time limits; focus on facts.

If they won’t bet on their claim or say what would prove them wrong, they’re probably bluffing.

That’s the kit. Once you see these patterns, you can’t unsee them—and you won’t waste your week wrestling smoke.

There’s a lie baked into the phrase “git gud.” and a lot of people — especially those who bounce off Souls games early — believe it.

They think it means, parry better, dodge faster, obsess harder, grind until your wrists hurt and slam yourself into a boss until the RNG breaks your way.

That’s not what “git gud” actually means — not for anyone who’s made it past the early walls and kept going. The real path to getting good isn’t technical mastery. It’s your mindset.

What it actually means...

Pay attention to your problems. Getting good means actually looking at what’s killing you — not just feeling frustrated that you failed again, but asking why.

Break the biggest challenges down. That boss isn’t unbeatable, but maybe their second phase only kills you when you panic because you never solved the first phase cleanly. Work backwards. Break the challenge down into its component parts. Watch the animations, look for the timings, test what you can bait out and look for anywhere you already have control, check the environment for features that give you the 'high ground'.

Solve one issue at a time. If you roll too slow, fix your loadout. If you run out of heals, fix your pacing and move slower so you get hit less, relax to avoid over-reacting. You’re not trying to solve the whole fight — you’re solving the next five seconds.

Know when to ask for help. Co-op has rewards for both sides, and guides don't just tell you what to do, but can make the world and its lore even more interesting or teach you entirely new ways of thinking about why an encounter was designed the way it was. Nobody gets extra credit for being stubborn. Use the signs. Use the wiki. Get unstuck.

Use the tools. Shields, summons, cheese, bleed builds, kiting, terrain, simply turn the music down. Heck in your first playthrough, grind out or dupe-glitch those souls and overlevel till diminishing returns if you're really stuck at a vertical wall — they’re all valid. The game doesn’t reward purity. It rewards adaptation.

Failure is part of the process. Most games train you to fear death, or at least to tire of it. Dark Souls trains you to use it. Every run is data. Every mistake is a lesson.

Take a break. Burnout makes you stupid. Walk away. Come back with fresh eyes. Nine times out of ten, the thing you couldn’t beat yesterday drops in one try.

Don’t you dare go hollow. That’s the only real failure state. Quitting, disconnecting, giving up not just on the game, but on what it’s trying to teach you about how capable you really are.

'git gud' isn’t about beating the game.

It’s about learning how to approach challenge itself, and this lesson is what you take into the world. It's why Dark Souls for many ends up like a religious experience.

That’s what makes Souls games stick. That’s why they feel important. They don’t just punish you until you improve — they show you how to improve. and when you know what to look for, you can apply that to the world outside of Boletaria, Drangleic and Yharnam too.

The people who tell you “git gud” like you're just not getting it exist everywhere, they kinda have a point, but they also probably don't mean what you think they mean, if you haven't been through the journey yet.

I’ve had to (well, I guess I’m choosing to, but still) join Twitter/X again. Just a stub account, in an incognito tab, so I can post new reports and links for the think tank project. Because if I want to reach people — and maybe help de-radicalise even the tiniest bit — that’s where they are.

Sign up, first screen, “For You” tab. In order from the top…

JK Rowling, Piers Morgan, Rupert Lowe, Julia Hartley-Brewer, Sky News, Charlie Kirk, Tommy Robinson, Matt Goodwin, Andy Ngo, Laurence Fox, MAGA Voice. Then Obama and Keir Starmer (both trailed by replies of bile upon bile, like everyone else I suppose, but with a slight switch in direction), then by hundreds more right or far-right accounts (how far is far I suppose? Well, MUCH farther than I ever saw there when I was active under my own name).

They’re all extremely active, a dozen tweets per hour, every hour, and holy shit, all of them are furious in sync. It reminded me of one time when I was a kid I saw a gang of dogs fighting in the park.

And the thing is, the existence of that side of Twitter isn’t an accident. It’s a direct outcome of (among many other things) Google killing Reader and dropping the RSS button from Chrome, laying the ground-work to force everyone into algorithmic feeds that make modern-day monetisation happen. Google didn't even really end up reaping the benefit of the seeds they sowed. If they're still going in 5 years, I'll be a little surprised.

We traded the open web — where tools made things possible, and people arranged their own infrastructure — for walled gardens and feedback loops of hatred, because it pays the bills of billionaires.

So that I can just switch over to post my report when it's ready to go live, I re-logged in from my old dead/stub account, and wondered it would tweak the suggestions once the two accounts were known to be related. It did… slightly. This time the list went: Jeremy Corbyn, some trade unionist, then Andrew Tate, Good Morning Britain, GB News, The Spectator, and another long raft of right-wing or very right-wing media accounts, most of which I've never heard of.

I tried scrolling to see if anything would change, but honestly, I don’t know how people who are still active on there can bear it. It’s just a wall of never-ending hatred and division, like text-based screamo on an infinite loop.

Sure, Bluesky (my current preferred 'discourse' platform) has its own issues — everyone agreeing with each other even when they’re provably wrong, instead of having meaningful debates about complex things. But over on X? It's so far beyond a circle jerk, I don't even know what to call it… like a human centipede of delusional outrage.