Creating a verification skill after i got inspired by @poteto. In Claude Code this shows before and after videos. I have a feeling that i'll use this a lot.
Had quite a lot of fun fine-tuning a model for this. I had @ClaudeDevs manage all the training.
@modal is an incredible platform to train on. It was very easy to set it up and start training on it.
Grok @bot has been draining usage a lot recently. I did about 24% very quickly today. I'm having to use it more judiciously now. I don't give it a lot more tasks because i don't want to run out of usage limits.
I'm seeing quite a lot of slop on Youtube. A lot of the videos i am recommended are just AI generated and are off putting. Some of them don't even try to hide it well. It's going to get really difficult to figure out what's right and what's not.
This is a truly a game changer. Such a neat idea.
I had Deepseek V4.1 Flash build an "AI Chat" video with and without the skill. It's incredible how much of a difference this skill makes.
Stageshow: a skill that helps your agent make better faux UI stills, animations, bento grids, and more.
npx skills add benjitaylor/skills --skill stageshow
I've been enjoying fine tuning models on @modal. You guys have made it really easy to get started. The CLI makes it agent friendly. i've been experimenting with using Opus 5.5 to drive most of it.
Trying to hill climb on a game with Deepseek v4.1 Flash. It's been about 5 hours of continous development. The visuals look okay for now. I'll see how much it can improve from here.
I love the customisability of the dot by @vercel's Geist Serif, so I built something to test it out.
The dots change based on the mood of the sentence.
I've wanted to generate a Cyberpunk style 3D video for quite sometime. I had Sonnet 5.5 high generate one recently.
I think the vibe of the city is there, but it needs some more graphical fidelity. This burned up about 1.3 times my 5 hour limit on the $20 subscription.
Using Remote Hosts on Rex feels incredible. It is so smooth. I forgot i was using a remote host for a while. @mitchellh@almonk you guys did really good.
Most used models on OpenRouter.
Deepseek is dominating for most of it as expected with GPT-5.6 Luna topping one week.
What's interesting is that weekly token usage grew by 2.7x in 12 weeks.
Is Sonnet 5.5 High > Opus 5.5 Low?
I had both models create a video of a Mario-like game with Claude as the character.
I’ve found Sonnet 5.5 High to be more creative than Opus 5.5 Medium recently so I wanted to put it to the test.
While Sonnet costs slightly more, the variety of levels is much greater and the animation quality looks slightly better. Opus leaned in on the stock Mario look for the game.
Diving into @poteto's pstack for the first time. I already know a couple of skills I want to customise for myself. It's a great starting point and I'll share more as I have some skills setup.