nano banana is really Google’s, you know…

Agents are still too advanced. I feel that before building a true agent, the model’s memory/architecture should be redesigned. Agents are at most an interim solution; the ultimate direction of LLMs should still be integrating other apps.

In the era of self-media, traffic is king, and even negative publicity is good publicity?

That’s quite abstract, still hearing this kind of talk these days 😂

I wonder if it’s possible to automatically push notes to Mastodon using RSS. Speaking of which, the next article could explain how to deploy Mastodon using Docker.

ChatGPT5 is really difficult to use, and think is very slow. I really don’t know why so many people praise it❓

A big problem with GPT5 is that it doesn’t follow instructions and is lazy. You have to say the same instruction many times, otherwise the result is a watered-down version.

It’s completely inferior to GPT4o😮‍💨

I saw a post about connecting desktop pets to LLMs, and it seems interesting, but I haven’t found any fun projects after looking around 😢

It looks like SCUM has a self-hosted version now, wow, finally!!! Let me check it out, the official version seems pretty good.

What!? Scum released the official version? If only I had a Linux Server.