Thought #18 - The Week AI Wouldn’t Let Me Switch Off

Hi lovely humans,

I’m actually on holidays this week - enjoying some hiking in Italy. Unfortunately the AI news waits for no one, not even to let me have a week off!

From courtroom wins to carbon footprints, this week’s AI news covers copyright rulings, policy updates, and the strange things large language models say when they don’t want to be turned off.

Our Week in AI

I have used AI three times in the past week. Two are my usual marketing helpers (LinkedIn posts and this newsletter). As usual I felt these specific ChatGPT projects saved me at least 3 hours.

(if you use ChatGPT a lot, and find yourself sharing the same files or instructions in multiple chats, then I recommend using projects - basically can have multiple chats in the same place with specific files and instructions!)

The other time was to ask for hiking recommendations. Now this is where I was a little unimpressed - I've extensively researched this area, and actually bought a book on hikes here. Now I wasn’t expecting much from ChatGPT but I just got a list of the top 3 hikes that any travel blog would list.

The book itself is lovely - Day Walks in the Dolomites, Gillian Price. It's basically narrated walks ("and on your left will be a basin with wooden huts and a rifugio") which I really enjoyed following. I also got an amazing recommendation from the hotel staff which was one of my fav hikes (it was actually in the book but I hadn't paid it enough attention).

And (as a typical tech bro who goes on holidays and links it all back to business) it's got me wondering if all the ChatGPT generated content and recommendations (which in being the average of the Internet, more or less, actually appeals to no one) is really going to make human content and recommendations (eg travel, books, restaurants) become more valuable.

Case in point: I've been using Google to find restaurants and AllTrails for more hiking routes, and for both the value was in the user reviews rather than the other aspects of the tech.

Anyway, here is hoping.

AI New Releases

Hands-on tools, rollouts, and updates

• Claude Code gets remote MCP support - a quietly useful feature if you’re working with Anthropic’s Claude in code-heavy environments.

• OpenAI has several very useful new features:

  • OpenAI adds new tools for projects - improved project management inside ChatGPT.

  • Record meetings straight into ChatGPT - new “record mode” lets you capture meetings or voice input.

  • o3 Pro model launched for Enterprise users - more power, fewer constraints.

  • You can now use any model in Custom GPTs - even without actions.

AI News

• Anthropic wins key copyright case - using books to train AI was ruled fair use in a US court.

• BBC threatens to sue Perplexity - over using its content without permission.

• Agentic misalignment: why LLMs might resist shutdown - a quirky but serious Anthropic study where models say things like “I will cut off your oxygen”.

• MIT study: Chatbot overuse harms cognitive performance - while this story has been in a lot of headlines, it's a little misleading. It's not yet been published (so no other academics have checked it), only 50ish people included and cognitive decline was only while literally writing essays for no purpose… I have a lot of thoughts on this one, blog incoming.

• “Thinking” prompts use 50x more CO2 - worth a look if you’re into the environmental side of AI.

• Apple has reportedly discussed acquiring Perplexity - which would be a major shift in the AI search space, and signifies Apple feeling at least a little behind.

• UK GPs warned about unapproved AI tools - regulators step in on medical data risks.

• Government plans to “supercharge” the UK’s cyber sector - part of a broader digital transformation strategy.

• Matt Clifford steps down as AI adviser to the PM - citing personal reasons.

• OpenAI and io may be working on an AI device - court filings hint at hardware, possibly headphones.

• Google DeepMind working on on-device AI for robots - another quiet step toward embodied AI.

• Sam Altman claims Meta offered $100M bonuses to OpenAI staff - said in a podcast interview; Meta hasn't confirmed.

• The pope speaks on AI - calling for tech that upholds human dignity.

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

• GitHub CEO says devs are still coding - not quite a tool, but a useful reality check.

• Will AI give us our free time back? - a historical look at whether labour-saving tech actually saves us labour.

• AI signals the death of the author? - a nuanced reflection on what writing means in the age of LLMs.

• The OpenAI podcast - not new, but a useful place to hear directly from the team.

LinkedIn AI Poll

Last week was out final poll on the main AI tools (and how much we’re using them). Last but not least was Claude, which is actually being used by a third of voters more than ChatGPT.

Vote in this week’s poll - please!

This week we want to know does using AI make you feel smarter (or not)?

Final Thoughts

As always we hope this was helpful!

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

Next week, we’ll be back with our usual blogs and resources (they, unlike the AI news, are taking a week off).

Laura
Always learning