Edwin Hayward
@edwinhayward
Author: 'Slaying Brexit Unicorns' (see link) and commentator. Tweets: AI & fun geeky stuff, Brexit & UK politics. Expect facts, stats, sarcasm & gallows humour.
Thinking of changing suppliers to Octopus Energy? You can get £50 in free credit when you switch by quoting referral code 'bold-lane-539' or using the link below. share.octopus.energy/bold-lane-539 I'll get a bonus too. Thank you. That magic £50 link again...
"It's Your Party, and You'll Cry if You Want To"
Jeremy Corbyn says his new political party, which he will co-lead with Zarah Sultana, will be 'very different' to the Labour Party under his leadership. The party name will be decided in a vote by members, so for now, they're referring to it as Your Party.
Google Gemini 2.5 Pro has decided that my app's code is as optimised as it's possible to be. That completes Phase 1. Phase 2: start adding functionality without breaking all that lovely speed!

This is astonishing. 1.2 billion views and counting. Motto of the story: You never know what will catch fire, other than petrol and a match.
Share a piece of lore about yourself
PROMPT: Do you have any secrets you've been holding out? Anything that, when you look at the code, screams "I know how to do this better! Why doesn't he ask me about it?" Gemini 2.5 Pro's such a tease! And its insight was really really good. Will speed up part of my app no end.

Feel the acceleration!
>>> Qwen3-Coder is here! ✅ We’re releasing Qwen3-Coder-480B-A35B-Instruct, our most powerful open agentic code model to date. This 480B-parameter Mixture-of-Experts model (35B active) natively supports 256K context and scales to 1M context with extrapolation. It achieves…
Kimi K2 is a superb AI model, but sometimes it can tie itself in weird knots. At the moment, while bugfixing 5000 lines of code, it has the problem narrowed to a 20-line function. It is now rewriting the function again and again and again. 22 iterations and no end in sight...
Best settings for coding using Google Gemini 2.5 Pro in Google AI Studio: Temperature: 0 Set thinking budget: on Thinking slider: 32768 (max) Grounding with Google search: off (You may want to leave grounding on, and experiment with the difference it makes.)
One problem I find with vibe coding using AI is the temptation to keep going in the same chat window, e.g. when working through a long series of bug fixes. The issue: since all previous exchanges remain within the AI's context window, its responses gradually get worse and worse.
We will never rejoin the EU until we start trying to rejoin the EU. After that, it may be quick or it may take a long time, but we'll be on our way. But if Labour drag their feet the rest of this term, that's 4 more years minimum (and likely decades) until we even make a start.
I wonder which frontier agentic AI model will be the first to empty an unwary user's bank account?
I used to oppose 16 years having the vote but talking to my 17 year old daughters particularly their desire for UK to Rejoin the EU has convinced me it's a good thing. A future EU Rejoin Referendum will benefit from the youth vote for sure.
Exactly!
Every time more bad news comes out about the UK economy, it makes Labour's "steady as it goes, with tiny tweaks" approach to Brexit more and more absurd. Reversing Brexit is the ONLY way to inject decent growth. It may not be easy. But it's NECESSARY. Nothing else will work.
Prompt tip for AI vibe coders: "Give me a technical and a plain English answer in pairs, ie techical answer 1 and plain English answer 1, technical answer 2 and plain English answer 2, etc." Add that, and it will give you the right vocabulary to Google for more info if needed.
Every time more bad news comes out about the UK economy, it makes Labour's "steady as it goes, with tiny tweaks" approach to Brexit more and more absurd. Reversing Brexit is the ONLY way to inject decent growth. It may not be easy. But it's NECESSARY. Nothing else will work.
When you using an AI via an API for tasks such as coding, you pay for the amount of data you send to the AI, and the amount of data the AI returns to you. But its thinking is free. One trick to getting better results is to prompt it to think more: "Think step by step. Wait."
Possible explanation for Epstein scandal... Republicans: "There's OCEANS of evidence! Warm up the buses. Hordes of people are going to jail." [Weeks of shredding and deleting.] Republicans: "There's ZERO evidence. Why are we raking over ancient history? So lame!"
Massive if you're vibe coding. Kimi K2 is great for planning, and you can now get answers faster than you can read them! Means AI response time isn't a limiting factor on your brainstorming sessions any more.
We’ve ben seeing a lot of demand for Kimi K2 on @GroqInc. Happy to say that it’s now available on the Groq API at 185 tokens per second, 6x faster than any other provider (AT FULL CONTEXT)