Rambeaux

@rambeaux.bsky.social

Software/Data/AI/ML Claude Code Addict

More Opus 4.7 thoughts: I have found myself bloating my app-specific guidelines with rules that are aimed at getting the LLM to do its job, rather than understanding my app. Eg: "Do not refuse to develop a feature because there is not yet a consumer of that feature."

Anther example of Opus 4.7 going to shit. It is struggling to distinguish basic scopes of user interaction. It confuses our requirements as a developer/agent pair with that of the eventual post-deployment user of the system. I have to explain the difference... We haven't built it yet, Claude! 🫠

My thoughts on Opus 4.7: Obviously brilliant at architecture, planning and coding. However, when it comes to understanding intent and reading between the lines conversationally, it is from another planet. Always getting wires crossed. Feels like a huge backward step from 4.6.

LLM coding is making software creation accessible to a shit ton of people that otherwise would never make software — lots of underrepresented types of people/creators will start to make things we might not have seen otherwise thanks to a greater diversity of perspectives that is a very good thing

Big "I will contact attorney general" vibes here from someone who rage screenshot-block-posted me yesterday. 1. I use the GPU for work. 2. I run it locally on upcycled hardware (slower, but cheaper, more ethical and fun) 3. Inappropriate? huh? You posted it! More deets in the alt. #AI #databs

An anti-AI ( ya think... ) rage post against me for asking the model of a GPU that was offered to the world.

Just sayin', but:
1. I use the current GPU to run open LLMs locally for my work.
2. I run it locally on upcycled hardware - a MacPro 5,2 no less! Slower, but cheaper, more ethical and definitely more fun to get up and running than paying OpenAI for their API. A 1070 is *just* powerful enough to be useful. Only just.
3. I put myself last ("if no other options...") 
4. I definitely expected to pay for it.
5. No shame in gaming, but still... work probably qualifies as "actually needs and uses it" more than playing fortnite. 
6. The poster couldn't have known any of this, but... they didn't exactly ask...
I was blocked, screenshot quoted in a matter of minutes, escalating faster than Part 1 of Lubalin's amazing internet dramas  https://www.youtube.com/watch?v=zr1h2Z11oTI 

lol

It has been a bit dry in Victoria lately so not as many saffron milkcap mushrooms as past years, but the quality and form of those we found today were the best we’ve ever found.

Saffron Milkcap mushrooms (lactarius deliciosus). Free if you know where to look.

If you were in doubt that Australia and Canada are at two ends of the same wormhole: Two opposition party leaders lose their seats in each election a week apart. Conservative and NDP. Liberal and Greens. That’ll never happen again!

Seems the data ecosystem in the Canadian election is pretty abysmal. In exploring riding results there is no: - comparison with votes from the last election - metric for percentage of (likely) votes counted It is basically impossible to get any context.

Feeling a little bit nervous watching the Canadian election with two seats flipping conservative early. Difficult to find an election tracker that shows the swing at a riding level though to get context.

The thing that finally, totally 🤯 about the tariff calculation is that it completely ignores the goods involved and only looks at the $. If I sell you my iphone for $1000, and you sell me your pencil for $1 that’s a couple of fair trades.