Fair play to Dario and @anthropic.com for putting their foot down against the Pentagon. Dario should get the Nobel Peace Prize for his stance. Legend.

Interesting watching things like Being Gordon Ramsay on Netflix. What ingredient is it that forces that continual drive? Such success and yet it’s clearly still not enough yet.

If the LLM capability exponential turns out to be an s-curve and ‘good enough’ open source LLMs integrated properly into a company with relevant contextual data is the way right way to go (which I think it probably is), then the frontier labs are in for some stormy seas.

I help run a reactive repairs and maintenance company. We've just deployed our own on-premises AI system. Local LLMs. RAG on our own data. Automations. Data sovereignty. Full GDPR compliance. This is what AI adoption actually looks like for SMEs. I'll be sharing the journey.

When I grow up I want to work at CERN.

Most white collar work isn't done in a well structured, self contained, version controlled repository with an instant feedback loop. Most companies are messy, siloed, human, nuanced, and the feedback loop is measured in weeks/months. That's far from an ideal environment for today's SOTA LLM's.

‘AI is replacing programmers. Software engineering is hard, ergo white collar jobs will be eliminated too.’ Software is well structured language in a version controlled repository with instant feedback. Perfect for AI. Organisations are messy, siloed, human and nuanced. They’re safe for a while yet.

1988: 8 year old me waiting for screeching cassette tapes to load on the #Amstrad CPC464.

2026: 46 year old me feeding a ROM into AI and playing an original 8-bit inspired browser game in less time than the tape ever took to load.

That curve is hard to grasp.

Where has all the RAM gone?

Open WebUI and Ollama work fantastically well together. Crazy enough that I’ve got a local LLM running on a home server. Even crazier that I implemented RAG and an SQL agent this evening, for those tools, using those tools to help me.

.md files in a GitHub repo + Claude CLI = the future of corporate documentation.
HR policies, finance procedures, legal docs, all version controlled. AI reads, suggests edits, and every change is tracked through git.
No more SharePoint sprawl. Just clean, auditable, AI-enhanced documentation.

Voice assistant! It can’t just be me that wants to wear one of these on my shirt - Star Trek combadge style.

The Raging Moderates and Prof G Markets have completely taken over my podcast bandwidth. Jessica and Ed are brilliant. After that chat with @kyla.bsky.social if @profgalloway.com doesn’t encourage her to join the team soon, it’ll be an epic miss.

Simplify. Structure. Automate. Then Add AI. The Real Path to ROI This article is an expanded version of my first post in the Simplify Structure Automate newsletter on LinkedIn, updated for readers here with added context and examples. The Hype vs. Reality Companies are throwing billions at AI. But how many can honestly say they’ve seen a measurable return? Too often, leadership teams start with the wrong question:“Where can we add AI?”

Simplify. Structure. Automate. Then Add AI. The Real Path to ROI

This article is an expanded version of my first post in the Simplify Structure Automate newsletter on LinkedIn, updated for readers here with added context and examples. The Hype vs. Reality Companies are throwing billions at AI. But…

Benefit of having an electric car when it’s warm. Drive through the French doors, park in the living room, open the windows and turn on the air conditioning.

I’ve been looking around for a while for an old butler sink for the outdoor kitchen I’m building. Yesterday one of my neighbours got rid of one, which would be great on its own, but this sink is originally from the British Museum, making it awesome.

The only logical answer is to focus on two things:
1. Energy innovation, get the cost of energy to trend towards zero.
2. Automation, automate as much as ‘humanoidly’ possible.
With the cost of production trending to £0, wealth inequality evaporates and we can live in a world of abundance.

Jony and Sam fall out with each other due to differing strategic objectives. The former wanting to focus on autonomous health care, the latter autonomous warfare.
Elon relocates to Mars and uses #Neuralink to maintain humanoid presence on Earth.

iBrooch makes use of iPhone tech + OpenAI tech + Vision Pro tech. It utterly decimates phone sales.
Focus shifts to humanoid robot production using iBrooch tech. Apple goes head to head with #Tesla and their #Optimus robot.

#Future prediction: Sam Altman and Jony Ive take over control of #Apple as it continues to struggle with Government pressure on the location of its manufacturing supply chain, and an uninspiring product development cycle.
They start with an AI enabled brooch that we all wear on our chests.

In light of what is happening in America in 2025, it’s staggering how accurate Back To The Future II really was.

I’ve spent the evening reading about how spacetime might not be fundamental. I’m calling it, us being in a simulation is the only plausible explanation.

That’s where government is meant to step in, subsidising or incentivising the beneficial yet unprofitable solutions, or taxing the profitable yet suboptimal alternatives.

So much criticism about Apple’s #liquidglass. Presumably it’s a matter of time until someone vibe codes a better mobile phone… 🙄

This is no doubt going to be unpopular, but I’m saying it anyway.
There are too many #cat photos on Bluesky.
I need a Catblocker.

This still looks like a solvable problem.
The energy requirement is going to be the major limiting factor to AI for quite some time, not the capability to reason.

4. Hold a weighting of fundamental principles. When faced with a seemingly insurmountable problem, which fundamental principle should be scrutinised first.
It’s in this area that the science community often operate.

3. Know when to decrease the ‘learned accuracy’ of the pattern, increasing randomness to try and find an entirely new or optimal pattern.

2. Know when to increase the breadth of that similarity. Applying known patterns to unrelated problems, and how far to push that boundary.

Surely the difference with human reasoning is the ability to:
1. Apply known patterns to ‘broadly similar’ problems.
What weighting to apply to known patterns to optimise which to try first, and how long to keep trying.