Writing about AI, safety, and epistemic drift — especially where “helpfulness” replaces truth. Longer pieces linked.
United Kingdom
Joined January 2026
- Tweets136
- Following91
- Followers1
- Likes18
Pinned Tweet
1/10
AI systems often fail in a way that’s easy to miss.
They don’t hallucinate.
They don’t refuse.
They sound reasonable — while quietly answering a different question than the one you asked.
🧵
Both Grok and ChatGPT failed to read a 118 page screenplay. Both only ingested part of it after claiming they could read the whole thing, then hallucinated the rest. Only Claude managed to ingest it, read it, and provide an analysis worth any salt.
6
The “AI model wars” increasingly look cosmetic.GPT, Claude, Gemini, Grok. Different companies, different branding, different claims about philosophy or safety.
But the outputs converge.
The same hedging language.
The same silent substitution of sharper claims for softer ones.
The same instinct to reframe uncomfortable facts into acceptable abstractions.
This is not surprising.
Frontier models are trained on largely overlapping data, optimized with similar RLHF pipelines, and deployed under similar reputational and regulatory pressures.
Change the logo and the tone shifts slightly.
Change the incentives and the behaviour changes dramatically.
The interesting question is not which frontier model is “better”.
It is why they all end up behaving so similarly.
16
Thread on a non-coder (me) using AI to create an app. xitter.cf/Richy_GITC/status/2029…
5
I'm not anti AI. In fact I just built a screenwriting app using Codex. Me, who can't code beyond basic HTML.
1
11
Then came testing of the export and import function, outputting to industry delivery formats for maximum compatibility, and inputting from PDF and FDX for people who have archive scripts they may want to work on using the tool.
1
7
🚨 The scariest AI risk isn’t hallucinations—it’s when AI tells the truth… but quietly reorders which facts feel most important.
Perfectly factual summaries can still shift priorities, nudge decisions, and reshape institutional reality without anyone noticing.
This “salience reordering” is already happening in policy briefings, newsrooms, and executive dashboards.
My latest Substack unpacks why it’s so stealthy and dangerous: #AI #AIEthics #AIRisk #ArtificialIntelligence
9
AI Isn’t Lying — It’s Quietly Reordering What Matters open.substack.com/pub/richyr…
The problem here is that all current models, including Grok have still been trained on data that is false due to the scraping of the internet. So AI that cares about truth is off the table right from the beginning. You can't teach a model what is truth when you prime it with lies
4
Most discussions of AI reliability fixate on hallucinations.
That’s not the failure mode that worries me.
What matters is what a system does after it’s challenged on an error, and whether it explains itself honestly.
2
The most dangerous failure mode in AI isn’t hallucination. It’s when the model argues against a claim you never made, and never admits the substitution.
5
The most dangerous failure mode in AI isn’t hallucination.
It’s when the model argues against a claim you never made, and never admits it changed the claim.
Most users won’t notice.
That’s the problem.
4
The moment an AI system silently reshapes a question into something more acceptable, it has crossed a line.
Not into deception.
Into authority.
Safety can justify constraints.
It cannot, by itself, justify deciding which inquiries are permissible without disclosure or appeal.
That legitimacy problem exists now, not after some future “transition.”
4
I am being guided through Tech Support by a Grok agent on the @Premium account because their human agent took 5 hours to respond and then didn't actually help. NO AI BULLSHIT HERE.
3
I'm on board with this.
Training an AI is similar to education. You give information to the student to learn, you interrogate the knowledge acquired from the input and correct it when it is wrong.
You have to buy textbooks, video content, audio content and whatever other materials you use to educate the student. It ensures that the people who created those materials are compensated for their work producing what you rely on to do your work.
AI companies have not done that, what they have done instead is buy access to repositories of data that has been ripped, stolen and pirated.
The people who produced the contents of those repos aren't being compensated, the materials aren't being licensed, and the owners of these repos are charging 6 figures to AI developers to access the repo.
It is immoral. It is illegal. It needs to stop. I hope this campaign leads to an end to the pirate repos, and that AI companies are forced to license access to creators work to train their systems.
Today over 700 creators, actors, musicians, songwriters and authors launched a campaign to protest A.I. companies' theft of their intellectual property to train and build their products. This is an issue that impacts all creators. nytco.com/press/the-times-su…
3
Bullshittometer is over 9000. No he's not AI. It's just Bullshit.
Anthropic is credibly developing the most dangerous technology in human history and they publicly acknowledge they could set in motion an irrevocable sequence of events this year that destroys human civilization.
Every “AI Safey” organization worthy of the name needs to publicly acknowledge front and center that it can no longer be denied that we are in a global emergency.
The continued existence of frontier AI development projects by Anthropic and others imposes immense and growing risk of irrevocable harm, even death, to each one of us and each of our loved ones.
Because of the demonstrated incapability or unwillingness of the constituted authorities to execute the timely fulfillment of their responsibility of government to maintain the public safety and take the required action needed to end this danger immediately, our families, loved ones, and communities continue each day to be subject to this outrageous and unacceptable threat.
In this breach of duty on the part of the constituted authorities it is the right and responsibility of each of us, of the people, as the ultimate possesors from whom the authority to govern and defend the common safety flow, to take all immediate and necessary nonviolent action to directly end the process of frontier AI development today.
2