I build AI YouTube channels and apps. 10M+ views. Sharing how it works.
Sweden
Joined August 2026
- Tweets1K
- Following181
- Followers188
- Likes369
Fable 5.2 came out last night and my whole feed is one shot demos. Fun to watch.
On my animated YouTube films the first render was never where my time went though. The fix rounds were, 3 to 7 of them on the complex ones.
What I want to know is if it needs fewer of those.
2
5
363
In a habit app I'm building, the paywall text loads from the server instead of living inside the app.
So I can test different wording without sending a new version through App Store review every time. After 12 days waiting on the first review, that matters.
7
8
519
People ask how long my animated YouTube films take me. About 30 minutes of my own time each.
I come up with the story, watch the previews and send fixes. Claude writes all the code that draws and animates it, in the background while I do other stuff.
1
7
579
Lots of "Opus 5.5 got dumber" posts right now. Maybe it did, no idea.
One thing that feels a lot like a nerf though is it repeating a mistake you already corrected days ago. It won't remember that unless the fix is written in a file it reads before every job.
1
324
My biggest YouTube film got fewer likes, subs and shares per view than the ones I made after it.
It still did 4.8M views. 83% of people were still watching at 10 seconds.
I'd bet YouTube cares way more about people staying than about them liking it.
2
6
383
Lots of people making videos with Claude right now. One thing that mattered a lot on my animated YouTube films is that something on screen changes every few seconds.
A shot can look great and still lose people if it just sits there. They don't wait, they skip.
1
4
378
Most posts about Apple's new Xcode plugin for Claude Code are about building iPhone apps faster.
The part I like is the simulator. A build passing doesn't mean the app works. If Claude can run it and see the screens itself, I'd guess it catches way more before you ever open it.
1
2
416
Why is Apple taking so long to review my app?
It’s been 12 days since I submitted it, and I haven’t heard a single peep from them.
Is this normal for App Store reviews?
2
3
471
Lots of "Claude makes cartoons now" posts lately. My YouTube films are animated in code by Claude, and most of my fixes are about motion.
A frame looks perfect as a still. Then it plays and the arms jump between poses and the feet float.
so watch it play before calling it done
1
6
432
Anthropic posted how Claude tidies its memory between sessions. They call it dreaming.
In their Pokémon test the older Claude saved a huge pile of files and kept getting stuck. The new one kept a few short notes and got way further.
If your agent keeps notes, keep them short.
2
1
359
In my YouTube thumbnail tests, determined or angry faces keep beating sad ones. Always the same cartoon character on a dark background.
My films are motivational though, so it might just be my niche. Has anyone seen sad faces win somewhere?
2
2
341
Linus retweeted
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
When Claude checks its own work, it already knows what it meant to make. So it usually says it's fine.
Open a fresh session that hasn't seen any of it, show it only the output, and ask what it sees.
If that one can't tell what's going on, nobody else will either.
1
287