@AndrewCurran_ So throttle American AI while everyone else has free reign to release what they want? I don't think this accomplishes what they think it accomplishes.
English
Chris Edwards
2.2K posts
Chris Edwards
@ChrisEdwards357
Follower of Christ, Husband, Father, AI Enthusiast
Ohio Katılım Mayıs 2008
450 Takip Edilen180 Takipçiler
The Trump administration has informed Anthropic, Google and OpenAI that they are discussing the creation of new AI oversight procedures that would potentially require new AI models to pass a safety review before being cleared for release.
Mythos has changed things.
English
@mattpocockuk You are doing a great job on your videos. Thank you for making content that is truly useful and a joy to watch.
English
Thanks for 200K subs on YouTube, pals
Ima keep cranking out the good stuff for ya
English
@thekitze I was thinking you were insane til I realized you were using Celsius.
English
@emil_priver I agree. Merge Request is a more accurate description of what is actually going on. Pull Request doesn't resonate as much. I'm not asking for a "pull", I am asking if I can "merge" my code. However, because of GitHub's popularity, "Pull Request" is more popular.
English
I think merge requests is a better name then pull requests
English
@mattpocockuk I think it was Ward Cunningham who said, "Every line of code is a liability." That framing of the problem really hit home with me. It really motivates principles like DRY and YAGNI. Solving the problem with less code is generally better.
English
Writing code is cheap
Maintaining code is not cheap
Anyone who's hired an external contractor knows this
Jonathan Ross@JonathanRoss321
For 50 years, software engineering ran on code rationing. Writing code was expensive, so we rationed it carefully through roadmaps, RFCs, prioritization meetings, and scope reviews. This created a role: the No Engineer. No, that won't scale. No, we don't have bandwidth. No, that's out of scope. No, we need a design doc first. The No Engineer was valuable for 50 years. Every "no" saved real money. Their judgment was the rationing system. LLMs will be the end of code rationing. Code is cheap now. And while the No Engineer is explaining why something can't be done, the Yes Engineer has already shipped three versions of it. If you're a Yes Engineer, the next decade is yours.
English
@mattpocockuk This is great stuff. This is an excellent way to ensure you understand the concepts and can communicate them clearly!
English
@mattpocockuk Is it a requirements vs implementation plan separation?
English
I'm thinking of renaming /to-prd and /to-issues
PRD is super-specific to a certain style of document
Issues are specific to GitHub
The important thing is that the PRD is a multi-session plan and the issue is the plan in that session.
Some options:
1. /to-mission-brief, /to-session-briefs
2. /to-epic, /to-tickets
Or, come up with your own!
English
@mattpocockuk I havent seen that issue. I do get tired of some of the skills during when I don't want them to. I ask Claude about something minor and it tries to go into brainstorming mode. Makes me want to uninstall it.
English
Folks who have used superpowers, does this ring true?
I.e. does superpowers keep trying to grab the wheel and take the conversation/plan in its own direction?
Aaron Layton@AaronLayton
It just always seems to want to add sub phases to a sub phase of a phase. It never really got anything complete. As part of normal conversation, you may mention something, next think you know its created a sub phase. Then your answering a question about the main phase, but then its tacked on 3 questions about this new subphase do this a few times and you're managing the flow instead of this meta framework managing it for you. It's hard to get 1 response that is a few paragraphs about what you are working on, where it shoves in issue F12 on SHA 12345, gives you 3 options to pick from to solve something then immediately starts talking about the new imagined phase from an off hand comment about the original architecture and how it just rewrote everything to accommodate your last off hand statement, and then give you another 3 questions (so 6 total now) asking for you to make a decision on it all. -- seems like i'm rambling, but actually happened today when building an AI Gateway backed by OpenRouter and me asking if we are OpenAI protocol compliant (i.e. can we use the OpenAI Sdk to interface with the new Gateway)
English
@mattpocockuk Feedback Loop makes more sense standalone. Backpressure makes more sense in the context of one service needing another to slow down so it's not overwhelmed. It's a specific type of feedback loop.
English
Which term immediately makes more sense to you?
(definition in the post below)
English
@trq212 Make opus 4.6 selectable via /models again without having to type the model ID in.
English
we're doing a lot more of this, hunting down some of the most annoying bugs in Claude Code
let me know if you have any white whales
ClaudeDevs@ClaudeDevs
In the last four Claude Code CLI releases, we’ve shipped 50+ stability and performance fixes. Faster resume, stable auth, lower memory, fewer hangs: 🧵
English
@RhysSullivan To many trolls on reddit. I used to like it. Not anymore.
English
@appfactory I use cmd+v and it works fine for me in wezterm.
English
why is it cmd+v to paste text into claude, but ctrl+v to paste an image?
English
@dexhorthy @zeeg You are probably right, but I hate email. Signal to noise is so low. To much junk to sift through. Too slow to get feedback... Probably because everyone favors slack and delays checking email.
English
Email > Slack
I’ll fight for this one
Karri Saarinen@karrisaarinen
I have probably never sent an email to an employee of Linear. Only when there is an external person involved. It’s Slack or Linear
English
@cyrilXBT Wow, what an old post. Why are you bringing this up again?
English
ANTHROPIC JUST PROVED MOST PEOPLE HAVE NO IDEA HOW TO PROMPT CLAUDE.
Their applied AI team dropped a 24 minute free workshop.
Not a creator who reverse engineered it.
Not a Reddit thread.
ANTHROPIC.
The people who wrote the weights.
And what they showed is uncomfortable.
There are 6 elements to a properly structured Claude prompt.
Most people are using 1.
Maybe 2.
That is not a skill issue.
That is an information issue.
And it has been quietly costing you every single day.
The outputs that felt slightly off.
The responses you had to rewrite 4 times.
The prompts that worked once and never again.
All of it traces back to the same 6 missing elements.
The people who watch this 24 minute workshop tonight will understand something about Claude that most daily users still do not know exists.
The people who skip it will keep getting 30% of what the tool is actually capable of and wonder why the results never quite land.
I watched it twice.
Then I built a Claude Skill that applies all 6 elements to every prompt automatically.
No more thinking about structure.
No more guessing what Claude needs.
The framework runs in the background every single time.
Full breakdown and skill setup is below.
Bookmark this now.
Watch the workshop first.
Then read the guide.
This is the one that compounds.
Follow @cyrilXBT for the exact prompt architecture, Claude skills, and systems I use to get outputs most people do not believe came from one person working alone.
English
@mattpocockuk @sjivan The detailed plan had all the details. So I don't know if it's context window loss, or just that the to-prd isn't prompting for as much detail. I wanted you to have the full context of the problem though. I'll be using it now and will tweak it and let you know what I find.
English
Got you, thanks for the strong feedback
It rings true for me as I have occasionally felt this myself.
I have been thinking of a /compact-to-file skill for a long time that just takes a conversation and compacts it locally. One specialized for capturing grilling decisions sounds very useful.
English
I want to rename /domain-model before I make a video about it.
It's a drop-in replacement for /grill-me that's better when you're working with code
The mental model is: /domain-model for code, /grill-me for non-code tasks
What should I call it?
English
@mattpocockuk @sjivan I use grill me all the time and I love it. I used to-prd and to-issues for the first time last night and it was a long 50-question session using codex 5.5. I usually use grill-me with opus, and then use superpowers writing-plans skill to write a detailed plan. Fewer questions.
English
@sjivan @mattpocockuk I found this as well. The prd was missing much of the detail that was discussed in the grill me session. I was appalled after discussing so much detail that it just ignored so much of it and wrote a very insufficient prd. The issues that it wrote after that were even more anemic.
English
/grill-me is awesome for interviews however it has a serious drawback. After a long interview session by the time it writes out the PRD files, from context loss the information it captures in the PRD is very very diluted and while you think you went through that detailed interview, the task working off the PRD contains a fraction of the details. One can easily verify this after a long interview checking the PRD only to find its ~30 lines long and missing lots of details.
I modified the skill to have it create a PRD file upfront and following every interview question/answer update the PRD file and at the end of the interview review the PRD for inconsistencies. No real reason one should need to use grill-with-docs. I thought the domain-model was primarily different in that it read the docs in the repo too.
English
@HowToAI_ It doesn't make it 90% cheaper. Not even close. That 90% metric is the reduction in tokens ONLY FOR THE CALLS THAT WENT TO RTK. It's not all traffic. I get maybe 5% savings with it overall.
English
Someone just open-sourced a tool that makes Claude Code 90% cheaper.
It sits between your AI and the shell and compresses command outputs before they hit context. git push, cargo test, ls, grep.. all auto-rewritten.
works with Claude Code, Cursor, Gemini, Codex, Copilot.
100% Open Source.
English
@_trish_xD Been there, done that. Times were very different back then.
English
@thekitze @thsottiaux I agree. They did this in Claude code too. Slash commands are not skills. They should not me skills. Keep them separate!
English
this is such an incredibly dumb decision @thsottiaux
skills get loaded in context, i dont want my 1938413 possible slash commands to be loaded into context, since they aer not relevant most of the time
command != skill
English