Thought #34 - Bubbles, Biases and Big Launches

Hi lovely humans,

It’s been a full week in AI land - and not just because of the OpenAI Dev Day hangover. From the Bank of England’s warning about an AI bubble, to new research on bias, to another milestone for Sora 2 (reaching 1 million users faster than ChatGPT!), there’s a sense of the sector both accelerating and reflecting. The big players are still rolling out features, but the tone from researchers and regulators feels more cautious - and perhaps, more realistic.

A cheeky little ask from me - could you vote for Taught by Humans please? We’ve been shortlisted for the Rising Star as part of the SETsquared Partnership Community Awards.

Our Week in AI

Trying Not to Break a Vintage Sewing Machine

Now my life has calmed down ever so slightly (famous last words), I am back to one of my favourite activities - volunteering at Share Bristol. Where I sometimes get to try and fix things so other people can borrow them. My current goal is to get this gorgeous vintage sewing machine working.

I’ve changed the fuse. It’s passing its PAT test, but the light isn’t coming on.

So I asked ChatGPT. One thing I love about these AI tools is being able to send a picture.

ChatGPT: It looks like the sewing machine in your photo is a Frister + Rossmann Cub 3. Here are some practical steps you can take to troubleshoot why it is not powering on despite having the correct fuse and passing PAT testing:

It gave some tips for testing - I did like that it checked if I’m comfortable carrying out the steps.

ChatGPT: Would you like me to guide you through a safe step-by-step live test with a multimeter? (Only if you’re comfortable working with mains electricity.)

But what I’m most impressed with - I told it:

L: I've been twisting the bottom screw for ages

ChatGPT: Ah yes on this model the back cover screw can be a bit deceptive.

What’s happening is probably this:

• That bottom screw on the Frister + Rossmann Cub 3 isn’t a self-tapper.

• It goes into a metal insert, so it can feel like it’s turning forever without actually coming out.

• It usually needs to be pulled gently while turning, or the cover needs to be eased loose at the same time.

And that is literally what worked.

The other top tip - the manual for this particular sewing machine doesn’t seem to be freely available anywhere. But ChatGPT seems to have access to it - so I’m putting together a guide using ChatGPT!

AI New Releases

Claude rolls out “Memories”
Anthropic has introduced persistent memory for Claude - a long-awaited step that lets the model “remember” past interactions.

Claude Code Plugins
Anthropic is also opening up its ecosystem to plugins, with a focus on code execution - another step towards more capable, task-based agents.

AI News

Sora 2 hits one million downloads
OpenAI’s video app has become its fastest-growing release yet, surpassing ChatGPT’s first-week downloads. The hype continues, though creative professionals are still watching the copyright implications closely.

Bank of England warns AI bubble may burst
The Bank’s Financial Policy Committee cautioned that AI-driven valuations and capital flows are starting to look frothy particularly among firms promising “AI transformation” without clear revenue models. The Guardian’s summary was blunter: a “growing risk the AI bubble could burst”.

Google opens its first AI Hub in India
The new centre brings Google’s full AI stack including Gemini and Vertex to Indian developers, signalling both investment and competition in global AI infrastructure.

OpenAI and Broadcom announce chip partnership
OpenAI is working with Broadcom to co-develop custom AI chips, aiming to reduce its reliance on NVIDIA. The deal underlines the strategic value of hardware in maintaining AI’s pace.

Microsoft faces class action over OpenAI deal
A group of users and investors has filed a potential antitrust suit, alleging Microsoft’s stake in OpenAI artificially inflated valuations and distorted competition.

EU launches new AI uptake strategies
The European Commission has announced two new plans to boost responsible AI adoption, focusing on digital infrastructure, SME access, and public-sector implementation.

Equity warns of “mass action” over image theft
The performing arts union says cases of unauthorised use of actors’ images and voices in AI-generated content are rising sharply and is threatening coordinated legal action.

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

• I asked AI to plan my dream summer holiday
  This light but telling BBC piece shows where consumer AI tools still fall short - and what that says about how we interact with automation when expectations get personal.

• Generative AI’s metrics question – Benedict Evans (thank you to a reader who suggested)
  A sharp, accessible take on why it's so hard to tell what “better” means in AI - and how the lack of standard metrics makes it tricky to separate meaningful progress from noise.

LinkedIn AI Poll

This week we want to know if you use memory in your AI tools (blog in the works explaining why I don’t use memory).

Our Blogs and Resources

ChatGPT vs Claude for Coding

Nicole has been comparing ChatGPT (4o) vs Claude (Sonnet 4) for a beginner trying to learn Python.

Data Privacy Policy Checklist

This week’s we’ve had a lot of chats about data privacy - so thought we’d share our resource on how to read a data privacy policy for new AI tools you want to try out.

Final Thoughts

As always we hope this was helpful!

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

Next week, we sort of hoping Sora 2 is available to more people so we can see what the fuss is.

Laura
Always learning