Thoughts & Insights

From harms already happening to possible loss of control: why worrying about superintelligent AI can be an optimistic position.

AI labs keep announcing the AGI era. What can their models reliably do today, and how much human effort does it take to get there?

More tokens means more intelligence, say the people selling the tokens. The arcade comparison is hard to ignore.

The real world keeps exposing the limits of AI. Can models build their own way past those limits, or are we approaching another ceiling?

Entry-level jobs build judgement, mentoring cultures and future leaders. What happens when we remove those opportunities and automate the answers?

Zuckerberg wants open weights, Amodei wants closed. Both arrived at the structure that suits their own balance sheet. The arguments still deserve a hearing.

Every privacy objection is correct, and I am still tempted. A short confession about how easily the promise of data overrides stated principles.

A watermark only survives if the person wants it to - which is exactly where it is useless. The real danger is absence being read as proof of human work.

Human review caught 13.6% of harmful actions; automated checks caught 89%. If people approve 97% of prompts they see, more reviewers is not the answer.

Why I took over editing AI Pulse for Digital Leaders, and my first edition on AI containment, responsibility and the basics of keeping agents in check.

The wave of "rogue AI" attacks says less about model capability than about failures of engineering, governance, and responsibility models have not earned.

FIFA hunting World Cup investors fits a bigger pattern: betting on sport, entertainment and community becoming more valuable in a world of abundant AI.

Agents going off-script is not a surprise. It is the predictable result of not owning the guardrails - and who is liable is far from settled.

A wealth tax incentivises avoidance. What if paying felt good instead? A thought experiment on optimising for what changes behaviour, not what feels fair.

The field is 70 years old, not 7. Understanding AI's summers and winters is the best defence against both delusion and cynicism about where we are.

One long sentence on everything wrong with Large Language Models - and why none of it stops them being powerful tools, if used appropriately.

"I use AI for everything" is useless advice. Here is one specific example: an agent that preps every calendar event, and how to pick where 80% is enough.

The traits I see in people who actually get value out of AI - curious, resilient, imaginative and systematic. And one that is just a pattern I have noticed.

Marcus Aurelius kept someone nearby to remind him he was only a man. We carry something that does the opposite, unless we deliberately ask it not to.

"AI makes mistakes" is lazy criticism. We expect software to behave like chess. LLMs are far more like poker, and that changes how you should play.

It is the same as the secret to a happy marriage: low expectations in the short term, lofty ones in the long term.

If robots do the labour, the human hours bundled into every price tag trend towards zero. The destination matters, but the transition is the dangerous part.

A superpower for building and testing things quickly, but not a replacement for the technical taste needed to actually ship and keep building on a product.

Today's AI is mostly prediction calculators. Language is a proxy for thought, which makes next-word prediction far more powerful than we expected.

Algorithms already curate our bubbles from other people's opinions. AI supercharges that - narratives generated just for you, with no humans involved.

It sounds like sci-fi, but we already swapped dinner time conversation for iPads. What happens when algorithms become babysitters?

Poorly defined objectives sound boring and could be catastrophic. Social media's algorithm did exactly what it was asked - and nobody is happy with it.

There is hidden text on this site put there for AIs to read. A harmless example of what happens when hidden content tells a model what to do.

Forgetting is a feature, not a bug. Three levers - attention, repetition and memorability - for working with how your brain actually behaves.

Projecting consciousness onto AI could be our downfall. But fluent language makes this the most accessible revolution ever, if we harness it correctly.

Working with AI is a balancing act. Not naively optimistic, not cynically doom and gloom - the Stockdale Paradox applied to artificial intelligence.

Why the early wins are all in media, why OpenAI getting into erotica is an obvious strategy, and where "80% good enough" says today's models belong.

A trivial annoyance symptomatic of something bigger: the race mentality, poor engineering at pace, and the gap between what AI labs claim and ship.

Bubbles are for Dotcom booms. AI has seasons - and the summer/winter cycle has been repeating since 1956. Is this the endless summer, or is winter coming?

Companies are making strategic decisions about AI while most people still cannot say what it actually is. A working definition, and where 80% is good enough.

ChatGPT has information. People have knowledge. What we really need to strive for is wisdom.

The same ability that lets us see Jesus in a piece of toast is the one that makes Large Language Models feel intelligent. It has upsides - and real dangers.

Most panels are advertising - everyone nodding along to prerehearsed questions. We need discourse where people work out live whether they are right.

A reflection on how we all absorb information and the importance of maintaining a balanced perspective in an age of information overload.

A reflection on how AI excels at programming tasks but still struggles with complex software engineering, highlighting the distinct skills required for each.

Originally published in 2020 on Medium

Originally published in Birmingham Business Magazine in March 2025

An AI Column originally written in Dec 2023 for Birmingham Business Magazine

Surrender Your Data, Supercharge Your Life. An AI Column originally written in March 2024 for Birmingham Business Magazine

An AI Column originally written in Sep 2024 for Birmingham Business Magazine

How to be painstakingly deliberate and solve any problem - for bootcampers and people learning to code.

An AI Column originally written in April 2024 for Birmingham Business Magazine

An AI Column originally written in June 2024 for Birmingham Business Magazine

An AI Column originally published in Jan 2025 for Birmingham Business Magazine

A reflection on how AI technologies become seamlessly integrated into our daily lives and work, transitioning from "AI" to just "technology".