Thought #31 - Building Big, Deciding Quietly

Hi lovely humans,

A big week for AI news in the UK. Tech firms have pledged tens of billions in infrastructure, while the government continues to position itself as a global AI leader. At the same time, universities, councils and devolved governments are quietly deciding how AI will show up in day-to-day life - and where to draw the line.

Quite a long one this week, but some to highlight:

1. Oxford is the first UK uni to roll OpenAI out to all staff and students

2. Anthropic has released details of the bugs seen over the summer

3. Italy has introduced a new law which will impact how AI is used

PS if you are remotely near Bristol we are organising two events during Bristol Technology Festival:

1. Actually Build Something with AI - 8th Oct

2. Create an AI Usage Policy for Your Team - 9th Oct

Our Week in AI

AI for Meeting Notes

We've been running at experiment this week at Taught by Humans to try out different methods for making notes from meetings. Basically, we have a lot of knowledge that needs to get out of my head, and ideally into notion.

[I have some reservations about my voice being recorded and the deep fake implications, but that's another story]

I started with recording our tech team meetings with otter.ai

I like otter because it seems to do a better job with my Northern Irish accent than others. We were only recording internal meetings about our tech, so data privacy wasn't such a big concern (as it would be for customer meetings).

Otter.ai tries to make meeting actions and summaries - these were fine but missed some nuance and had a few errors.

But taking the transcript, coping into Claude and asking it to analyse and suggest documents to be made, got us some really good documentation.

The big pluses of Claude output (rather than the Otter.ai) was bits of the conversation were combined and moved around. My favourite bit was the to do list it added at the end - which included some very tiny things I mentioned in the call. We also had something we can directly copy to Notion, which serves as future documentation.

We then tried it for a marketing meeting, and got a really detailed to do list. And a very needed fun list of future actions (as me and Carys tend to discuss very random ideas, we now have a list to come back to!)

There are perhaps other tools or workflows which we'll be testing out in the coming weeks.

But yet again an example of having to use a multi-bot or tools approach to get the right results.

AI New Releases

Google showcases Spline, its new 3D modelling AI
Still early days, but Google's new model can generate 3D object meshes from prompts. It’s aimed at developers and designers - part of a growing trend to apply generative AI to spatial and product design, not just words and pictures.

OpenAI launches Stargate UK
OpenAI’s new UK-based infrastructure project aims to bring its most powerful models onshore. The pitch: faster and more secure local deployment for industries where location and legal jurisdiction matter - like finance, defence, and healthcare.

Teen safety features coming to ChatGPT
OpenAI shared how it’s navigating safety, privacy, and autonomy for younger users. The result: custom safety settings for 13–17-year-olds and a commitment not to serve ads. It’s a tricky balance, but this is a step towards more age-aware design.

Anthropic explains its summer bugs
In a rare move, Anthropic published a detailed post-mortem of three bugs that led to inconsistent Claude responses for some users. Technical but readable, and refreshingly open - especially in an enterprise market where trust matters.

AI News

Tech giants announce £30bn investment in UK AI
Microsoft, Amazon and others have pledged major UK investments in AI infrastructure. It’s being framed as a win for the UK’s global AI standing - but little detail yet on how it will support jobs, skills or regional growth.

NVIDIA and OpenAI partner on high-performance systems
The two companies will collaborate on deploying 10 gigawatts of NVIDIA systems globally - a sign of the growing arms race to power large models. It’s big, but not unexpected.

Italy passes sweeping AI law
Italy’s new law sets out rules on AI use in healthcare, education, and by children under 14 - and introduces formal oversight mechanisms. One to watch as more governments move from strategy papers to binding regulation.

Wales outlines AI plans for public services
A new written statement from the Welsh Government emphasises trustworthy AI in health and social care, plus a focus on building internal skills. Still high-level, but signals that UK nations may diverge in approach.

Anthropic limits how the US government can use its models
Anthropic has refused certain uses of its models, including surveillance of US citizens. Raises a key question: should companies get to say no to governments - and who decides what counts as a “misuse”?

Stockton Council says it won’t replace people with AI
Despite adopting AI tools, the Council has publicly committed to protecting face-to-face services and not using automation for job cuts. A clear position - and one not many others have stated as directly.

Oxford becomes first UK university to offer ChatGPT Edu to all staff and students
A notable milestone - and likely a sign of things to come. Still early days for ChatGPT Edu, but its rollout at Oxford could set a precedent for other institutions.

Cambridge University Press on ethical AI for assessment
A thoughtful paper exploring how AI can be used ethically in English language assessment. The core idea - using AI to support human judgement, not replace it - is increasingly relevant across education and hiring.

Gemini hits gold-medal level in world programming contest
Gemini tackled 10 of 12 problems in the ICPC World Finals - the most prestigious student programming contest in the world. It wasn’t competing directly against humans (it started 10 minutes late), but it still managed to finish at gold-medal level.

Not Quite News, But Worth a Read (or Listen or Watch)

• Why we shouldn’t fear AI – by Tim Berners-Lee
  A calm and balanced perspective from the inventor of the web. Less about hype, more about keeping humans involved in design and decisions.

• Vibe coding: what it is and why it matters
  A short piece from Vercel about how more AI-assisted developers are ‘coding by vibe’. Useful if you’re trying to understand the culture shift that comes with autocomplete tools.

• How AI is shaping lives and work in Kenya
  Many of the people labelling data that trains AI live and work in places like Nairobi. This piece is a valuable reminder that AI’s impact is both global and deeply physical - not just theoretical.

LinkedIn AI Poll

Last week we asked if you were using AI more for work or personal tasks. Perhaps unsurprisingly on LinkedIn people are using it for work. But the majority are using it for personal stuff, at least as much as for work.

This backs up the report released last week about how AI usage has changed over the past year towards more personal use (planning, companion, etc).

I have used ChatGPT to help me shop for work clothes this week, and I was really impressed with the advice and tips given.

Vote in this week’s poll

This week we want to know are you using GPT-5 when using ChatGPT (or are you still going back to GPT-4o)?

Final Thoughts

As always we hope this was helpful!

Feel free to share this with anyone who might find it useful.

Next week, we hope the lack of job replacement news continues.

Laura
Always learning