6 ms·
Interesting. Wasn't Deepseek's founder saying that they had explicitly decided not to focus on multimodal models at all and were going text-only because they be
by v9v 26d ago
Interesting. Wasn't Deepseek's founder saying that they had explicitly decided not to focus on multimodal models at all and were going text-only because they believed it was enough to achieve AGI?
- dakolli 26d agoI think you're thinking of Dario saying this about image generation.
- johndough 26d agoIt was explicitly said that they are pursuing multimodal support. A quote from the meeting transcript: https://github.com/demo-zexuan/liang-wenfeng-investor-meeting-2026-7-22/blob/master/%E6%A2%81%E6%96%87%E9%94%8B%E6%8A%95%E8%B5%84%E8%80%85%E4%BA%A4%E6%B5%81%E4%BC%9A-%E6%96%87%E5%AD%97%E7%A8%BF_1_18_translate_20260723201651.pdf https://github.com/demo-zexuan/liang-wenfeng-investor-meetin... Nevertheless, as a component, we will undoubtedly implement multimodal support — and we are already doing so. We plan to develop relevant models, ensuring that versions like V4 and subsequent iterations will natively support multimodal functionality. Earlier, the following was said, which might match more what you had in mind. Achieving excellence in AI training does not require a global model or even multimodal approaches—by narrowing the scope of AI training and eliminating multimodality, certain tasks may remain unachievable without compromising the algorithm's validity. Multimodal approaches ultimately need to be implemented. It is difficult to tell who said what, since the speaker ids are missing.
- v9v 26d agoThanks, I seem to have grossly misremembered what I read.
- swiftcoder 26d agoWorth noting that deepseek has had a separate vision-capable model for some time, which also powers their chat interface's vision mode