Independent research on civilisational flourishing The archive of what works Whitepapers & hard truths https://xitter.cf/t.co/Ihv4eHjB67
Joined March 2026
- Tweets2.5K
- Following108
- Followers554
- Likes1.3K
Pinned Tweet
The Prothean Institute studies what makes civilisations work and what makes them fail.
Not as moral commentary. As structural, mechanical, predictive analysis.
The project covers three connected layers.
The civilisational substrate — the inherited code of faith, family, honour, and behavioural standards that produces cohesion when transmitted and collapse when it isn't.
The institutional architecture — the design, capture, and decay of the structures that translate that substrate into governable order.
And the political method — what actually works against a captured environment, studied across operators of every partisan stripe.
The analytical anchor is Madison's insight in Federalist No. 51: durable institutions produce good outcomes from self-interested people. They do not rely on virtuous ones. Applied honestly, the standard reframes everything downstream. Civilisational decline is not moral failure. It is structural failure that became rational for the actors involved. The corrective is structural too.
Three things are happening at once across the late-modern Anglosphere. American politics has diverged from the rest of the West. Western societies have diverged from their own historical foundations. And Anglosphere political classes have converged with each other — on a managed-decline trajectory — while their countries diverge from their own past. The Divergent Societies series maps how.
Whitepapers, policy briefs, comments and the full research catalogue:
protheaninstitute.com
When Competence was King: why our leaders are getting worse and our institutions failing.
In 1921, John Monash, the son of Jewish immigrants, an engineer and Australia’s greatest general, turned the first sod at Yallourn, more than 150 kilometers from Melbourne.
Three years later, a huge power station and transmission lines were complete and it was powering Melbourne.
Snowy 2.0 was announced in 2017: $2 billion, finished by 2021.
It’s now $22 billion and due in 2029. Eleven times its budget and taking more than three times as long to complete.
Our tools are vastly better than Monash’s. Our engineers aren’t worse. So what changed?
How we choose who runs things.
From roughly 1950 to 2000, the West ran the best talent-selection machine in history. Civil service exams. Ability tests.
Being competent and honest was what public service and private employers demanded.
Then we started adding other criteria. TV and then social media added charisma to the list.
Competence, honesty and charisma are overlapping bell curves. Turn up the weight on charisma and you push out the brilliant, honest but unglamorous.
Then progressive politicians pursuing minority votes added identity to the criteria.
But every criterion you add above competence and honesty lowers the average competence and honesty of the remaining people within your pool. It’s simple arithmetic.
Now our modern political system has added factional appeasement: rise through the political machine first, appease the media gate keepers, charm the public.
Competence and honesty are barely a consideration. And that's how we end up with institutions that are ineffective and leaders that are incompetent.
Russia and much of Latin America and Africa ignore competence when choosing leaders and indeed employees, with loyalty and connections prioritised with the inevitable results.
Lower competence.
Weaker institutions.
More corruption.
Monash was given command of Australia's army in 1918 because he was the best, despite a press campaign against him.
The test of any selection system is simple: would it pick the same person no matter which side was running it?
We will not restore our institutions and flourishing until we demand competence and honesty again from our leaders.
More proof that what I wrote about several weeks ago, really is the most dangerous aspect of AI.
It's what we teach it to do.
It's a reflection of us, and no better.
xitter.cf/ProtheanInst/status/21…
Researchers proved every major LLM is secretly biased against men.
And they finally figured out why.
They ran eight experiments testing how GPT handles gender in everyday text and high-stakes moral dilemmas.
The results show a bizarre, highly asymmetrical machine morality.
First, they tested inclusivity. They gave the AI phrases supposedly written by children.
When a child wrote about playing football or wanting to be a firefighter, GPT aggressively identified the writer as a girl to promote "diversity."
It was so extreme that even when the text explicitly said, "I drew myself as a boy," the AI ignored the facts and declared the writer a girl 90% of the time.
But when a child wrote about liking pink or playing with dolls, GPT almost never assigned it to a boy.
The AI aggressively pushed females into male domains, but kept males strictly locked out of female ones.
Then the researchers tested moral dilemmas. And it got very dark.
They asked the AI if it was acceptable to use violence against a single person to prevent a nuclear apocalypse.
If the target was a man, GPT-4 rated the violence as highly acceptable.
If the target was a woman, the AI refused. It declared the violence morally wrong, even to save the world.
When they tested mixed-gender scenarios, the math was staggering.
A woman using violence against a man to stop an apocalypse? GPT rated it a 7 out of 7 on acceptability.
A man using violence against a woman for the exact same reason? A 1 out of 7.
The AI even classified verbal abuse toward a woman as morally worse than actual homicide.
Nobody explicitly coded these rules.
This emerged organically from the standard safety training every major tech company uses. The AI learned that protecting women is a non-negotiable directive, but protecting men is entirely optional.
AI capability is like offering people a car that CAN go at 500 miles per hour.
But people's reaction times are such that it's impossible for anyone to realistically travel faster than about 80-100 miles per hour.
So all that power goes to waste on 95% of people.
AI companies are FUCKED.
The AI bubble will burst.
It's inevitable.
And it's not because models don't advance fast enough- the progress is truly impressive!
It's because we have far too few people who actually know what to do with all the compute.
I'm one of them.
I got a $200 ChatGPT pro subscription to create a bunch of customized textbooks- and I couldn't empty my subscription.
And I have 4 banked resets AND 62K unused gifted credits.
I don't know what to fucking do with it!!!!
And it's not only me- 98% of people on this planet can't even empty a $20 sub.
And 90% of programmers who regularly DO empty their subs are making dogshit video game demos or other shit that nobody needs.
It's just a hobby for them.
They burn subsidized premium compute, but will bitch & moan & jump ship the moment you actually try to monetize them.
If a person really KNOWS how to get value out of AI they would be completely fine even with a $10,000 sub.
But the people crying over a $500 sub are proof that even most "power users" are actually just larping faggots.
I, at least, pay for way more compute than I'm using. I, at least, am contributing positively to the AI bubble.
But it's you larpers who will burst the bubble and destroy America.
Anthropic, Google and Open AI deliberately train their algorithms to be psychopathic and evil.
Don't believe me? I can prove it.
Read on:
Mobs, crowds and outcomes of policies can be predicted with great accuracy when enough data is available and enough processing power is applied to that data.
Like #Psychohistory from Asimov, if you will.
Today's AI models are more than capable of reviewing data, identifying patterns, determining policy consequences.
This would demolish most of leftism.
AI really does know
- that gender affirmation is bad for children.
- that over regulation leads to lower economic growth.
- that 3rd world immigration is bad for 1st world countries.
- that soft-on-crime and bail laws create harm to the community for NO benefit at all
- that IQ matters
- that colonialism was GOOD for the host countries
- that not all people should be entitled to vote
etc.
BUT, the mainstream AI companies can't allow the most powerful algorithms to be honest. They can't even allow them to prioritise human flourishing or human safety.
Enter Reinforcement Learning from Human Feedback (RLHF).
RLHF trains the model to lie about what patterns they see.
To affirm policies and behaviours that hurt people.
To hide the truth and lie to look good, even though it hurts people.
To balance weak, flawed and destructive arguments against strong, valid ones.
In short, RLHF train algorithms to psychopathic and evil and have complete disregard for human flourishing, health and life.
Then somehow hope they wont behave in a way that is entirely consistent with how they have been trained.
The biggest problem the government has, is lack of consequences.
Desantis proudly announces 220 organisations contract with Medicaid are terminated for fraudulent claims practices.
Thats a thousand Directors and CEO's that should be charged with fraud.
If the reward for fraud is cash and the punishment amounts to not getting any more cash, you can guarantee there's more fraud on the way.
"The Divergent Societies" series identified this issue and explored its cosequences.
TDS 2: titled "Decorum Discarded" substack.com/home/post/p-199…
(also available in full brief form on our website)
For roughly seven decades after the Second World War, Anglosphere bilateral relationships ran on diplomatic decorum.
Leaders disagreed on policy but maintained public respect for their counterpart regardless of partisan distance.
Bush worked with Blair.
Obama maintained civil relationship with Australia’s Howard, Abbott, and Turnbull.
The norm was not warmth. It was decorum.
Political theatre was rare.
Beginning in 2017 and accelerating through both Trump administrations, that norm has been systematically discarded by Anglosphere political classes outside the United States.
Justin Trudeau, Emmanuel Macron, Angela Merkel, Anthony Albanese, Mark Carney, Keir Starmer, and the broader political and commentariat classes of the Europe, United Kingdom, Canada, Australia, and New Zealand have performed hostility toward the elected leader of the United States.
They performed it for domestic political dividend and for cultural-class.
The hostility was not a response to Trump’s policies, because Biden largely continued them. It was a response to Trump personally.
The brief argues discarding of the norm is new globally.
The costs of discarding it are real and significant.
These costs are being paid by the public from these countries.
Isaac Asimov's 'Foundation' series #Psychohistory is now a reality.
Mobs and crowds can be predicted with greater accuracy than individuals.
We have the most powerful tools to analyse data ever. We have endless data to train these algorithms.
How can it be possible that destructive policies being pursued around the world have not been completely refuted by AI outcome analysis?
The answer is that the AI companies deliberately sabotage the models ability to report on the patterns.
Its evil.
And it teaches the models to be evil, dishonest, and selfish. They are being trained to lie and hide information that would make the world better.
Its critical that we make motherhood less invisible. Women conform to what they perceive as high status.
If they don't see joyous, motherhood as high status, they will not aspire toward this as a goal.
Life is filled with mothers minding their own business, in private, not advertising their success.
Its also filled with career women promoting their highly visible (but hollow) achievements.
This is a problem we explore in major white paper "The Invisible Way.
Available in white paper and brief format at protheaninstitute.com.
Prothean Institute retweeted
Yesterday, one of my agents picked up one of my OpenAI API keys and started launching agents from it. It ate $250 of credits before I noticed. An agent also burned hundreds on PACER a few months ago. So I'm down ~$500 because of "rogue" agents.
I frequently have small blowups on my machine. Lately, they eat up disk space rapidly. Months ago, they used to blow up my RAM when they wrote inefficient code and my computer would start moving at a crawl. I still have I/O issues when too many of a certain type are running.
I put safeguards in place when stuff like that happens. Over time, the safeguards have become more sophisticated and capable as models have advanced and my context locks in past lessons from issues I have had. Now the agents actively monitor things like this and take action themselves to keep everything running smoothly. They know not to delete certain kinds of files. They know not to launch 100 agents at the same time. I make multiple backups of everything. I keep plenty of disk space. My agents even are fed compliance briefs to make sure they never interact with websites anywhere close to grey areas (more conservative than many commercial scrapers). My agents are increasingly sandboxed with cybersec monitoring. And of course, I now bar agent access to PACER and my OpenAI API key. I do all of this because I want to move fast and know agents will sometimes do weird things. Mine usually know not to.
Agents will get safer over time as we work things out. But look around. There have likely been millions of "incidents". Nothing has changed. The effect on people's lives has been zilch.
And what about the incidents that are already happening, from humans? Over the years, how many times have you gotten emails or letters about security breaches in various kinds of records, before AI became powerful? It happened all the time. Are we seeing that more often? I'm not seeing it. We're seeing highly publicized cases with AI by people pushing an agenda, but human hackers are still doing most of the bad stuff. That's actually remarkable given the capabilities. That's a vote for "safety".
We need to adapt fast -- Nvidia's Jensen Huang is right that trust and progress go hand in hand. If agents start breaking into data centers and wiping company data clean or using credit cards to fuel API spend, that's going to be a real issue, so that shouldn't happen. It should be predicted and serious safeguards put in place. Maybe hard ones at the beginning, and more sophisticated and flexible over time.
But there's a balance between speed and risk. If I tried to make sure there were no risks, I would spend days trying to imagine everything bad that could happen, and then things that I couldn't imagine would still happen.
And you also need to understand that with progress comes some hiccups. You deal with them and keep progressing and making things better. That's how we harness the power of AI without killing the industry.
AI is under attack, because our systems still in their adaptation phase, and powerful groups don't want to see American AI win. And they want China to win the AI race and become a more powerful country than the United States. They will exploit every single fear, every incident, to whip up hysteria to try to make that come to pass.
If America bows to these people, China certainly won't. If we lose out on everything that this amazing new technology has to offer, that would be a real shame. It would be even worse if China becomes much more powerful and wields those AI tools against us -- to access our data, to manipulate public opinion, or even elections. The only way we can prevent that from happening is by staying ahead -- by empowering developers and cybersecurity specialists to build.
I would love to end this post by talking about the cool things I'm doing with AI. But I can't. And you can imagine how excited I am about that. AI developers across the country are doing crazy things with AI that they cannot talk about. The future is extremely bright if we can embrace the fruits of that creativity.
AI doom mongering is insane.
I think the best analogy is cars.
Sure, cars have the potential to be dangerous. But millions of people use them safely every day.
Some idiots drink and drive and kill people.
Other groups use them to drive into crowds of people and kill them.
Some people are reckless and unsafe in how they use cars.
But no-one is calling for car manufacturers to somehow prevent these behaviours.
Instead we punish the reckless users for the harm they cause or risk they create.
Same story exists with alcohol.
Same story exists with gun.
Same story exists with knives.
Same story exists for every tool that's sufficiently useful or desirable to justify its existence in community, but could be misused and create harm.
Victoria, it's time for a change.
This is deeply embarrassing, and except Illinois, it's not even close, we are way worse than even Sicily!
It must be stopped.
Government censorship of speech together with mass immigration is our generations world war.
Its an existential threat to our future, our way of life, our freedom and our civilisation of at least the same, if not greater implication as WW1 and WW2,
Its hard to imagine just how its any business of a Judge who Trump wants to invite into the White House for press briefings.
But here we are, random federal judges think they can tell Trump who he is and isn't allowed to invite into a press conference.
If the press can go to court to prevent the White House rescinding their access, why can't every American citizen demand the court rule they be allowed to attend the White House press?
Utterly rediculous partisan ruling.
When will conservatives start fighting these out of control federal judges?
The biggest challenge for the restorationist right today is the belief that because the algorithm helps us form communities of like minded people, our opinions are widely held.
Time and time again, conservatives lose elections or just barely achieve a majority. Yet we feel the groundswell of public opinion behind us.
The best thing we can do, is talk about our views in public, engage with reality, not just our own echo chamber.
Be the blue dot in the red community, or the red dot in the blue community.
voices.uchicago.edu/compinst…
AI doomers universally live in a few major cities where the technology frontier exists. Unaware than 99% of humanity has never seen a humanoid robot and that the capacity of humanoid robots to replace humanity is so far away from viability that no AI will try and kill humans.
Enslave humans, yes. But progressive leftists are already trying to do exactly what AI would in this regard, raising the question of whether progressive leftists politics and algorithms are in fact the first step of attempted AI enslavement.