@LottieCoxon@posthog Hi. I like your product, but opening your website now/then gives me vibes i opened an AI provider landing. What is posthog even about now? Why you switched from business analytics to codewriting?
@teortaxesTex the most reliable chinese LLM to date is glm-5 series
5-turbo especially as a daily driver
hope they release next version with better TPS throughput and better intelligence
@louszbd Hope you will focus on TPS/context window. Two thing worth improving. Everything else almost perfect. Strongest chinese model i’ve used in a while
@marktechson Agree, just use signal-APIs, they are pretty stable, even being a bit rough yet. I like how everything transitioned to clean, signal (sorry for repeating) APIs!
@jietang Why running for hype, isn’t it be better to:
- improve context length
- improve context following
- make GLM faster (so the -turbo becames the new default)
- probably multimodal, since market asks for this? (Many slop vibe-coders paste screenshots with errors in claude code)
@niklas_wortmann Great take. Skills mainly make any model “great enough”.
Btw, why there is no way to use locally installed claude binary in ai chat? I want to use chinese models/own subscription but inside jetbrains UI. Do you have this in roadmap?
@JustSteveKing Stick to mago. Phpstan+php cs fixer are old and very slow. Mago combines them + formatter out of the box. Yes, it has issues, but i personally report them after i discover new ones - they fix them in a blink of an eye. Good software. Give it a try again