Every single LLM I use winds up using European spellings and idioms. Claude, GPT, even my local Qwen Flash Next all wind up speaking to me in Britishisms. It's is strange.
Contributions
🔥 0 day streak · Longest 10🏆 Achievements 18 / 80
First steps 4/7
Explorer 2/13
Engagement 3/12
Content 2/8
Community 1/10
Reputation 2/6
Dedication 1/3
Code Farm 0/9
Quizzes 0/6
Levels 2/5
Milestones 1/1
Qwen Flash Next with Strata is a beast. It's genuinely making me reconsider my max sub with Claude.
I switched from GPT to Claude this week after several of my sessions just stopped progressing for no reason. But the real kicker was when Sol 6.1 was stuck for 30 minutes not doing anything and I see it thinking about adopting a pet.
Also, the difference in quality of result between high and max...
LLMs are not conscious and will never be conscious until they either put them on cloned human brains or figure out real time learning. Once that happens, I'll change my mind. That said...
I've been experimenting with keeping a single session going for a very long time and mixing in several...
It's amazing what Astra can build. It made a dll hook for a 25 year old game that has improved the resolution choices, fixed a mouse look bug, improved its collision handling, and fixed its broken error reporting. And it all works. It's amazing.
Pro tip. If you have Codex working on something in batches, make sure you tell it to work on the largest batch size it can, or else it will split your 3000 step project into 1500 batches of 2.
Astra is now throwing flags about cybersecurity concerns, pausing goals without warning or explanation, for things that are not in any way shape or form cybersecurity related. I'm on the max plan and I have to babysit it to keep it working. This is getting ludicrous. The Chinese models are good...
Been gone for a while but I've been productive.
Basement flooded and led to tearing everything down to the concrete, which was a bummer.
But, in the background, Sol finally finished its reverse engineering/documentation project. I can move to the next step which is reimplementing the...
First time running out of usage with GPT, and I have to say this is really good treatment of the customer. Claude just stops mid task and suggests you do vile things with a cactus.
If you need to do opengl and python, make sure you tell your llm to use pyopengl and to not write its own opengl interop layer. The only reason I caught it in time is because I know what ctypes is. Yeesh.