Pinned Tweet
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 🧵
Peter Bowyer retweeted
Oh wow. Apparently I can't even quote one sentence and link to the article on my website without paying PCMag a $12,000 licensing fee.
What in the absolutely stupid fucking shit is @PCMag smoking these days?
This quoted post is unavailable.
What has happened to @OpenAI Daybreak Blue @thsottiaux? I used it to prepare for our regular Pentest in early summer and it was great.
This time round every other message does with "This content can't be shown" making it useless. GH Issues has many similar reports
65
My top 5 hymns:
1. And can it be
2. Here is love, vast as the ocean
3. To God be the glory
4. O come, O come, Emmanuel
5. Guide Me, O Thou Great Redeemer
My top 5 hymns:
1. Praise my soul, the King of Heaven
2. Thine be the glory
3. Crown him with many crowns
4. Tell out my soul
5. Lo, he comes with clouds descending
2
35
I challenge you to use a plan with 5-hour limits for a week @thsottiaux.
You'll see why we'd rather have half the usage and use it whenever we want than have to reschedule our lives around arbitrary 5hr windows.
The results of this poll has gotten me confused. I definitely thought it would sway much more favorably in the 2X direction across Pro 100/200/500 at the cost of bringing back 5h limits in order to smoothen the demand curve.
1
46
Peter Bowyer retweeted
We're very pleased to release Splink 5, which is now more scalable, faster to train models and lighter to install
🤖 Made with AI
Peter Bowyer retweeted
Appallingly sad news that the wonderful Stephanie Cole had left us today. An absolute force of nature and national treasure.
Impossible to imagine anyone else playing Carolyn Knapp-Shappey the way she did, so brilliantly. Thoughts and prayers to her loved ones.
Is there any way to get "Pro" thinking level inside Codex @thsottiaux ? It's amazing for researching and planning project architecture, but would be even better with access to local code and documents.
46
The latest GPT/Codex CLI changes seem to have helped - Sol Medium polled a RAID setup for an hour, ran benchmarks and analysed the results for another hour, and only used 6% of the Pro 5x weekly allowance.
Good work @thsottiaux and @pvncher!
Replying to @vikhyatk @thsottiaux
Saying "Poll at 5min intervals" helps a little, but not much.
I reported the problem and sent traces to @pvncher comparing 5.5 and 5.6 on jobs in the same repository - it was like a 10x token burn difference. @thsottiaux is this one the tiger team can investigate please?
1
2
1,083
Peter Bowyer retweeted
After speaking with SQLite folks I've just made this completely available today. Give it a read.
Peter Bowyer retweeted
A thought: is it a harness problem not model problem? Watching Claude Code, when it's compiling or waiting for a background process, there's no (visible) polling like Codex cli does. Does each harness listen for "done" in a different way?
Peter Bowyer retweeted
Introducing DataBench, a new frontier benchmark for complex data work and analytical reasoning.
Compared to other analytics benchmarks, DataBench uniquely evaluates realistic data tasks derived from real Hex usage instead of "pub trivia" style Q&A.
It's designed to evaluate analytical reasoning and judgment, not just coding ability— and we see really interesting test-time-compute curves as a result.
More details and link to results in thread!
If you use an AI Agent Search Provider in a harness like @ExaAILabs @firecrawl or @tavilyai - does it work better than the built-in Claude or Codex web-search?
2
4
576
Let's hope so - I'm not in the OpenAI Open Source program (yet) and a runaway security review burned the week's tokens overnight on Thursday. Still putting right what it did!
Session 019fdcde-b171-7b32-98ce-76bcde275811
1
132
I'd pay a lot for an AI model that writes well. Like 5x the token rate I pay for @OpenAI and @AnthropicAI now.
Current models are okay at critiquing writing, but slop as soon as you ask them to act on it.
Please @thsottiaux give us a 24h limit (00:00-23:59 in our account timezone) not a 5h one.
5h fits so weirdly in 8h work-days/24h days.
Hello people of Sol! I've reset usage limits for all ChatGPT Work and Codex users. Together with that, a quick update on GPT-5.6 Sol usage limits.
Over the past few weeks, many of you have told us that Sol was using your Codex limits faster than expected. To be clear, we have not reduced usage on any subscription plans.
We’ve been digging into what was happening and have landed several improvements. As a result, we expect your usage to last around 18% longer during typical use of Sol. Some of you should already see significantly larger improvements from today. Tomorrow, we’ll also restore the five-hour limit that we temporarily paused while investigating.
Here’s what we found:
- GPT-5.6 Sol is much more willing to work for longer, make additional tool calls, and coordinate complex workflows across tools and subagents. That makes it better at solving hard problems, but some tasks were using far more than we intended.
- Sol also works harder at the same reasoning effort than previous models. High on Sol can use more tokens than High did on GPT-5.5.
- Programmatic tool calling, also referred to as code mode, gives Sol much more flexibility to run tool calls in parallel or continue working while waiting. But it also led to more responses per turn, more cached input tokens, and higher usage than expected.
- This was particularly noticeable when Sol was waiting for tool calls to finish or running many web searches. We’ve improved how we handle both cases and are continuing to make code mode more efficient.
- The impact was also very uneven. The median user actually found Sol quite token efficient, while some power users working on harder tasks saw their usage drain much faster. We were very focused on average and median usage before launch and missed some cases where the long tail could use significantly more usage.
Sol is a significant step forward in what Codex can do, but capability and efficiency do not always improve at the same pace, and some issues only become clear once people are using the model at real-world scale. We should have recognized this sooner and been more upfront about it.
You keep pushing the frontier and we’ll keep improving efficiency and sharing updates as we go.
61
Peter Bowyer retweeted
This is fascinating from @MichaelDnes1 on his Pennine tunnelling plan, and why he couldn't make it happen within government *even though it would have slashed the cost of the project*
















