Sebastian

[Sebastian @Danmoreng]

@thsottiaux Complex tasks. When I create a new API and rename a parameter after some iteration, it shouldn't put in "legacy" compatibility for the old name. 🤣

[Sebastian @Danmoreng]

@szeloof fab1 was the garage? 😃

[Sebastian @Danmoreng]

@giffmana Ideogram4 for comparison.

[Sebastian @Danmoreng]

@HamelHusain Except you cannot do the work of 20 people. Moving much faster doesn't come free: if what took 3 weeks now takes 3 days, you are missing out of 18 days thinking about it and that shows in the implementation. AI speeds up prototypes. Good production systems still take time.

[Sebastian @Danmoreng]

@simonw It would also be so easy to filter out the bots for the platform - but that's something they simply don't want because it would reduce their user count by a wide margin. @nikitabier

[Sebastian @Danmoreng]

@alexinexxx Would be nice to have competition for NVIDIA, but I’m pretty skeptic. This take could become reality. https://t.co/F9pCNkVFC8

[the tiny corp @__tinygrad__]

@Etched This is the next Theranos.

[Sebastian @Danmoreng]

@antirez @SourceCodeplz @ElevenLabs Or just use the raw cli https://t.co/8O59L7KfHm

[Sebastian @Danmoreng]

@antirez @SourceCodeplz @ElevenLabs You can use something like Qwen3TTs with voice cloning. I made a simple desktop app with cpp gguf implementation under the hood. Around 5x realtime on CUDA. https://t.co/BFmzAvs4k5

[Sebastian @Danmoreng]

@ngxson No import fees at all for small packages wasn't good either IMHO, but I'm with you that a flat 3€ fee is a horrible solution. 🤦🏼‍♂️

[Sebastian @Danmoreng]

@mitsuhiko On iPad/desktop this looks good imho. Mobile is another thing though...

[Sebastian @Danmoreng]

@weswinder Where did you get that graph from? When I visit the website, the graph looks like this.

[Sebastian @Danmoreng]

@badlogicgames Im happy with my Samsung S25, but that’a not stock Android of course. For comparison of phone specs I can highly recommend gsmarena. Really nice side-by-side view of multiple devices, especially useful for size comparisons.

[Sebastian @Danmoreng]

@badlogicgames Hey! I just spent the weekend optimizing the repo. Should be 10x faster now - on my RTX 5080 I get 5x realtime speech streaming. Only tested with Windows/Linux and CUDA though, I don't have MacOS. Should work though since its GGML based.

[Sebastian @Danmoreng]

@jh3yy I was thinking somewhere in my corporate intranet. 😆 Component-wise button hover sounds reasonable to me.

[Sebastian @Danmoreng]

@antirez Tbh best thing one can do is block such accounts, regardless if human or bot.

[Sebastian @Danmoreng]

@MichaelThiessen For my simple-markdown-viewer I use this library which works well https://t.co/n43LMQAwYp

[Sebastian @Danmoreng]

@LyalinDotCom And after this post you have at least one more 😂 tbh I just ignore people who I don't know in person.

[Sebastian @Danmoreng]

@leegmoore @mitchellh Qwen3.6 27B Q8 is Sonnet 4.5 level - just takes 40-50GB for this precision and KV cache. To be fast this has to be VRAM. So you basically need an RTX Pro 6000 which is 12k atm or 2x 5090 for 8k.

[Sebastian @Danmoreng]

@badlogicgames Meanwhile: babysitting GPT 5.5 to get every UI feature exactly the way I want it, because otherwise the result is a hot pile of garbage.🙃