Peter Bowyer

So you want to know what uses energy in your house? £16 of tools and a few hours free time is all you need. Here's how to find out. A 🧵

@rowlsmanthorpe I'd have jumped at that when younger (and had fewer ties). Perhaps they'll let me do it when early retired!

Interesting report on the Number 10 innovation fellowship: a scheme to recruit highly technical people for a short "tour of duty" in the public sector

@jamsandwich @annavanpraagh Portrait orientation vs landscape orientation - I've found no record of both being valid

@peterbowyer @annavanpraagh Is there a wrong orientation of an England flag ?

I’d love to put an England flag up outside my house but don’t want a brick through the window. How can we normalise pride in our flag, being patriotic and supporting Britain on the world stage again?

@pvncher @thegenioo I can send you an equivalent gpt-5.5 one if helpful (will be another 40 min or so until it finishes)

@pvncher @thegenioo I fed back earlier, but assuming the session id is the feedback id: 019f5576-81a8-7df2-9ea5-18fbb3de1eb1. gpt-5.5 high uses 10-20% of the tokens for this

Honestly done with how horrible limits are for even 5.6 Sol (light) on Plus account, I give a small review task and it burns through my limits like a madman How the hell am I supposed to even use it when it just burns through all my limits in one response? at light?

@KentonVarda It's a personal project, there's no requirement to issue CVEs is there? Fix, release, move on

Feeling the pain a bit of the deluge of AI security reports on Cap'n Proto. 10 bugs reported in the last three months, which is more than the entire 13-year history of the project before that. None that are extremely critical (I think) but there are some near misses. Be sure to upgrade to release 1.5 (or to tip of the v2 branch) if you use it. Don't get me wrong, I'd much rather have bugs reported and fixed than not reported. But it's a lot of work. Although the Cloudflare Workers team heavily uses and contributes to Cap'n Proto, I still technically own it as a personal project and have been personally responsible for releases and security advisories. That used to work perfectly fine, I spent like two days a year on it... but I think it might not be viable anymore. It's a lot of work to write up every bug, request a CVE, etc. Rules say you must issue a separate advisory and get a separate CVE for every bug, but I just don't have time, so I issued a rollup advisory with no CVEs. Sorry. Gonna have to figure out something new going forward. Ideally the whole process could of course be automated with AI, but then I have to spend time setting up that automation, including prompting it to write advisories in a way that doesn't suck, which I... also don't have time to do right now, too much going on. github.com/capnproto/capn…

@thsottiaux Writing. Write like a human, not a robot. A custom skill takes the rough edges off but doesn't fix it all. No AI lab has made a model that writes well, so you be the first!

At different times you complained about speed, code slop, frontend quality, ... With each release we improve and GPT 5.6 Sol is ✅Fast and token efficient ✅Hardcore at back-end dev ✅Great at front-end ✅Does not use useEffect everywhere What is next?

gpt-5.6-luna high is no match for gpt-5.5 high. Still looking for what is (without burning tokens at Sol's rate)

@pashmerepat It is confusing with Sol/Terra/Luna and effort levels, on top of new models. In future, please release an "For previous users, the equivalent of 5.x-low is ..., 5.x-med is ..." so we don't get surprised. Was esp important this generation because 5.5 = terra or sol?

A few things to clear up here. 1. Reasoning buckets These are rough product labels, not fixed apples to apples token budgets across model versions. 5.6 spans a wider range, so 5.6 xhigh isn’t equivalent to 5.5 xhigh and can use more tokens. 2. Token efficiency When we say token efficiency, we mean that 5.6 can solve the same problem using fewer tokens than 5.5. But if you run it at a much higher effort ceiling, it can spend more in order to do more. And like I mentioned in point #1, the effort ceiling is not 1:1 across buckets in the different models. I agree this is quite unintuitive, but hopefully this clarifies

@rasbt Sorry, but I find this so confusing. If the Luna High is better than a Sol Medium, why does the Sol medium exists? I'm honestly so confused with all these models and levels.

@Dirbles_ Interesting, thanks. Until I get my head round this I'm going back to 5.5 or 5.4 subagents if it'll let me (once my usage resets 😭). The 3 levels is too confusing

@peterbowyer im thinking about just not using terra, theres a model thats cheaper for the same intelligence 5.6 terra low -> 5.6 Luna Medium 5.6 Terra Medium -> 5.6 Luna High 5.6 Terra High -> 5.6 Luna xhigh or 5.6 Sol Low 5.6 Terra xhigh -> 5.6 Sol Medium 5.6 Terra Max -> 5.6 Sol High

There’s currently an issue on Codex for 5.6 Models, they are unable to select models and reasoning effort for spawned subagents To fix it, add this to your config.toml: [features.multi_agent_v2] hide_spawn_agent_metadata = false tool_namespace = "agents"

@thsottiaux The 5h usage window is buggy with 5.6? Using Sol and Terra subagents and it's burning away. 5.5 xhigh didn't use it this fast! The weekly allowance seems right for the tokens used.

@kellabyte Can't believe the replies you're getting defending the attacks. He is making personal attacks, it's a terrible look for Zig and Andrew keeps 🪏🕳️

The personal attacks on Jarred have not stopped by the way. They have continued on the Zig mailing list.

@danshipper @tedescau Do you pair it with any skills, for example to make it mimic your tone of voice?

GPT-5.6 is a much better writer than Fable. It consistently one-shots marketing emails for @tedescau that every previous model would fail at. Fable is too verbose and liable to fall into using sentences in its own private language. If you use AI for writing, 5.6 is a fantastic model for you.

Hi @trq212 Is there a way to enable API billing for Fable 5 in Claude, but use my Claude Code subscription for everything else? Thanks!

@thsottiaux Expand my pasted text so I can see and edit it bfore submitting the prompt.

What is something that you feel is surprising that Codex still can't do well and we should have gotten right a while ago?

@wightmanr @giffmana Good to hear it's not just me needing to back and forth now with Fable. I'm back to doing multiple review rounds w GPT-5.5 xhigh to catch Fable's mistakes (missed race conditions, logic errors etc). Still better than Opus tho!

WDYT of it? Pre ban I had a half day with it and though it was remarkably thorough. It was one of the only times I'd worked through a non-trivial feature and it covered all bases, no obvious issues from my own review or a codex review. Trying it after return though, I seem to be back to the usual back and forth game to get to the quality I expect... with many oversights that are typical of Opus 4.7/4.8 or Codex w/ GPT5.5 xhigh. It no longer feels like a significant leap. Hmm.

tfw none of my Fable5 chats has been downgraded...

@m_ashcroft @Kirsten3531 Exactly right! Add in the expertise and flexibility we bring (and fire-at-will) on top of the hidden employment costs, and we are still good value.

@Kirsten3531 it helps to factor in employment costs that you don’t see! You’re not benchmarking against your salary but total seat cost, which can be a lot more than salary add in shared services, employers’ NI, etc etc but yeah, it’s good to factor all that into day rate

I just calculated what I should charge for freelancing and oh my GOD how does anyone afford to hire people who aren't full time employees?!! salary + pension + PTO + a little extra for the coordination costs of finding multiple clients and suddenly your day rate looks ludicrous!

@llqdave Mostly agree - the other situation is where they focus-group a policy, and still have to U-turn because their group wasn't representative...

@peterbowyer Why do you think we are subjected to so many u-turns these days? Someone pays (erm, I mean "lobbies") for a policy to be implemented, and then the general public have to point out the stupidity of it.

I'm amazed how little policies are wargamed before being introduced. From the outside it looks like they never think beyond the top expected effects. Or the downsides worked through from a SME perspective.