Thought #22 - Winning Isn’t Everything (But It’s Nice)

Hi lovely humans,

The standout news for me this week is about maths (my favourite subject at school, if you’ll believe it). There was a big international maths competition - and of course the major AI companies wanted to show off their “smarts” by solving the puzzles. Apparently, they were asked to wait a week before announcing results, to give the young people a chance to enjoy their accomplishments. But OpenAI couldn’t even wait for the winners to be announced before claiming gold.

It’s also been a busy week for government and policy - the UK is pushing ahead with new infrastructure partnerships, guidance, and funding, while the EU’s Code of Conduct is making progress (Meta still not a fan).

Our Week in AI

Some Clarity on Coding

My ongoing journey to get Claude’s help with coding had a bit of a breakthrough this week.

1. Claude is great at writing tests for existing code

2. It also handles small, well-defined tasks really well (e.g. “build me a file uploader that chunks uploads”)

One standout moment: I was debugging a PDF uploader that Claude just wasn’t fixing. Slightly frustrated, I googled it - and Stack Overflow delivered a tiny, perfect solution.

AI is good, but not always quicker than old trusted habits.

A Slightly Weird AI Brag

Taught by Humans won Best Use of AI at the Sparkies last week!

We’re not usually ones for trophies, but it’s lovely recognition for the work we’ve been doing to make AI more practical, inclusive and human-friendly. Thanks to everyone who’s been part of it so far.

My favourite part” The application itself was very AI-assisted - back from when ChatGPT was being sycophantic. Much more enthusiastic than I’d ever be on my own… but clearly it worked.

AI New Releases

ChatGPT Gets Agents (Finally)
OpenAI has launched its long-anticipated agent features for ChatGPT. The new tools can now do things - like sort your spreadsheet, book a calendar event, or fetch a document - as part of multi-step workflows. It’s rolling out gradually to enterprise users and may take a bit to reach everyone.

Mistral Le Chat Gets “Deep Research” Mode
Le Chat can now scan academic papers, legal documents, and other long reads in a more structured way. Early tests show it’s more cite-heavy than ChatGPT and better at complex references - though still a little clunky on the UX.

AI News

UK Government and OpenAI Announce Strategic Partnership
A new memorandum of understanding outlines closer collaboration between OpenAI and DSIT, aimed at using AI in public services and boosting UK infrastructure. Critics (including the FT) have called it vague, with concerns about government reliance on a single US firm.
Read their Memorandum of Understanding

Supercomputing Access Opens to SMEs and Researchers
Applications are now open for access to the UK’s AI Research Resource (including the Isambard-AI cluster in Bristol). Designed to give smaller organisations access to advanced compute, as part of a broader push for AI equity.

Civil Service Releases AI Guidance for Applicants
A new candidate-facing guide lays out when and how AI might be used in hiring - and what applicants should (and shouldn’t) use themselves.

UK Plans to Use AI to Verify Migrant Ages
Controversial policy move: the Home Office plans to use AI for age assessments of migrants. Rights groups have raised concerns about accuracy and bias.

EU Releases Code of Conduct for General-Purpose AI
A voluntary code - but signed by Mistral quickly followed by OpenAI and Anthropic. Meta has declined, and Microsoft is expected to sign shortly.
→ OpenAI | Anthropic | Microsoft and Meta

Microsoft Continues Poaching DeepMind Talent
Another sign of the talent wars heating up, particularly for AI safety and infrastructure roles.

Wiley Signs with Anthropic
The academic publisher will use Claude to support education and research products. Details are light, but this could be a sign of mainstream publishing shifting from OpenAI to more “safe by design” models.

Netflix CEO on “Real People with AI Tools”
Ted Sarandos argues for AI as assistive rather than replace-all - calling it “a way to make humans faster, not disappear them.” One of the clearer industry voices on blended teams.

OpenAI’s New $50m Fund
A fund for co-developing AI tools with nonprofits and community orgs - worth keeping an eye on if you work in digital inclusion or social impact.

Google Summaries Reduce Link Clicks
Pew Research finds people are clicking links less when AI-generated summaries appear - which might reshape how web traffic and SEO work, especially in news and publishing.

SoftBank’s Vision of 1,000 AI Agents per Human
Masayoshi Son thinks we’ll soon have “1,000 agents per person” - a very different framing to OpenAI’s “AI teammates”.

DeepMind and OpenAI Win Maths Olympiad Medals
Their models might have won gold, but their timing was off. Despite being asked to delay announcements until after the prizegiving, OpenAI went public early - and Google followed. Not the best look when you're up against 17-year-olds.

OpenAI Infrastructure Updates
OpenAI is teaming up with Oracle to expand its Stargate supercomputing project, and has moved ChatGPT’s hosting to Google Cloud. No changes for users yet, but a strong signal about where (and how fast) they’re scaling.

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

• Can 20 Hours of AI Training Improve Your Job?
  Yes, apparently. A new survey found those with ~20 hours of AI training report significantly higher job satisfaction and performance - but those opportunities aren’t being spread evenly. Especially lacking: women and lower-paid staff.

• Anthropic Argues for Building AI in America
  A policy-heavy essay on why foundational model development should stay within the US. Less flag-waving, more about regulatory influence and national security.

LinkedIn AI Poll

Last week, a lot of news was about voice. So we asked if you’re using voice to chat to AI. As someone who never uses voice for anything, I was intrigued to see if everyone is chatting away.

Turns out no one is using voice all the time (or no one is willing to admit it!). And the majority of people are like me, and don’t use voice ever.

We’ve been playing around with some voice interactions within the team, and will be sharing a blog soon!

Vote in this week’s poll - please!

This week I’m back on my data privacy thoughts, and I want to know what you think companies like OpenAI are actually doing with your data.

Our Blogs and Articles

AI Usage Policies

We’ve put together a guide and a template for writing an AI Usage Policy (something we recommend for all teams using AI).

Assessment for Learning

Carys, our residence education expert, has been thinking about something we love called Assessment for Learning. Something to think about if you are in charge of anyone’s learning or development.

Final Thoughts

As always we hope this was helpful!

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

Next week, could AI get the memo about summer holidays please?

Laura
Always learning