Kalin Fetvadjiev
@kfet.dev
Software Engineer, interested in AI/ML, modern languages, likes sailing, fan of Nintendo consoles and games (Metroid and Zelda in particular)
🇨🇦 🇧🇬 🇪🇺
* fosstodon.org/@kfet
* threads.net/@kalinf
* github.com/kfet
🇨🇦 🇧🇬 🇪🇺
* fosstodon.org/@kfet
* threads.net/@kalinf
* github.com/kfet
I’m very sceptical of the idea that Deepseek would hold out on the release of a model which is ready.
Not sure if they used Mistral’s work though.
Deepseek only got a good MoE model way later. I think that’s why I got the impression they followed Mistral’s work.
Not sure if they used Mistral’s work though.
Deepseek only got a good MoE model way later. I think that’s why I got the impression they followed Mistral’s work.
December 12, 2025 at 9:04 PM
I’m very sceptical of the idea that Deepseek would hold out on the release of a model which is ready.
Not sure if they used Mistral’s work though.
Deepseek only got a good MoE model way later. I think that’s why I got the impression they followed Mistral’s work.
Not sure if they used Mistral’s work though.
Deepseek only got a good MoE model way later. I think that’s why I got the impression they followed Mistral’s work.
Mixtral of experts | Mistral AI
A high quality Sparse Mixture-of-Experts.
mistral.ai
December 12, 2025 at 8:52 PM
I think it’s the other way around, Mistral invented MoE, then Deepseek used it to build their models efficiently.
December 12, 2025 at 8:27 PM
I think it’s the other way around, Mistral invented MoE, then Deepseek used it to build their models efficiently.