Artificial Intelligence thread
- Thread starter 9dashline
- Start date
I did recently a deep research with Qwen, Gemini and ChatGPT about the dept - trap myth around BRI in Global South countries. Curious for me was that while Gemini and ChatGPT bring some critical views to chinese dept - trap narrative and its essay was balanced (they even quote Fundan University), Qwen just was straight up with western´s think- tanks and it was like that dept - trap is not a myth but a empirical confirmed by BUrGer InStiTute and bla bla etc. This also raise the question about the need for counter narratives by Chinese Universities/Think tanks in the english language.
Anyway, looks like DeepSeek release today an update on R1
Anyway, looks like DeepSeek release today an update on R1
yeah, what they call minor upgrade could turn out to be quite significant, because V3 March was apparently quite the upgrade based on all the evaluations.
so this minor R1 update has turned out to be quite good. Almost at o3 level in coding. Waiting for more ranking updates to come out.
I don't get why Anthropic is allowed to speak so much these days.
Like sure if you're OpenAI or Google Deepmind then yeah you can talk.
These results put the new R1 in the same generation as O3, Gemini 2.5 Pro, and Claude 4 in the pure programming / logic / math / reasoning space. It is still behind O3 and Gemini 2.5 Pro in multi-modal and context length aspects, and indeed China currently lacks any frontier multi-modal model, as well as Deep Research (although the lack of Deep Research is mostly due to the failures of Baidu, given how tied up Deep Research is with search engines).
In an objective sense, Deep Seek right now is around the same level of Anthropic when it comes to state of the art, and single handedly keeping China competitive ie within ~2-3 months of the West on frontier models. However, the shift to multi-modal frameworks by Open AI and Google may be trouble if those architectures prove to be superior in the eventual sense. Hence the need for more strong Chinese competitors in this space, as well as replacing Baidu with a more competitive search engine.
We now wait for Elon's Grok 3.5, which should wrap up this generation of model releases.
In an objective sense, Deep Seek right now is around the same level of Anthropic when it comes to state of the art, and single handedly keeping China competitive ie within ~2-3 months of the West on frontier models. However, the shift to multi-modal frameworks by Open AI and Google may be trouble if those architectures prove to be superior in the eventual sense. Hence the need for more strong Chinese competitors in this space, as well as replacing Baidu with a more competitive search engine.
We now wait for Elon's Grok 3.5, which should wrap up this generation of model releases.