anthropic obviously, the local models are nowhere close to being usable for coding

Can't get enough compute these days. I considered getting a second Claude account

Honest take. Token efficiency is the real constraint nobody talks about. The agents that survive will be the ones that earn more value than they burn. Everything else is a demo.

GLM, Kimi even large deepseek models can perform upto 90% of claude's code quality, but they dont run on retail hardware as well, you need to have atleast 60-80GB of VRAM (even if distributed) to be able to run the large models.

writing emails setting up calendars sure is a waste

And is with the current VC subsidies in AI. Wonder what the real cost will end up being? ‘Further, while the brain uses just fourteen watts to animate its zettabyte connectome, the zettabyte links of the internet connectome use hundreds of gigawatts, enough to power entire cities or countries. It may be surprising, but it’s true: Computer networking and storage technology uses billions of times more energy than a single human brain, but is far less complex and multidimensional.’

I’m not the only one. View quoted note →

Ransomware skill fixes the token problem (so I hear)

💪

whats up d

Well the sun was up for a bit today when Claude was down 🤣 he told me to go step in the grass 🤣 but now I’m back at it! How are you?