Bluesky Thread

the deepseek effect is that now any new model only has to exceed R1 in order ...

February 09, 2025 View original thread

the deepseek effect is that now any new model only has to exceed R1 in order to win headlines. the thing is, R1 isn’t state of the art

sir, prepare for grift off

36 2

first up is grok 3, an LLM trained solely on tweets from Musk fans

Alt Text: A tweet from the verified user “JB” (@JasonBotterilL3) reads:
“chocolate model in Chatbot Arena is likely Grok 3. Nothing else can be this cringe.”

Below the tweet, there is a screenshot from Chatbot Arena showing a conversation where a user asks, “what model are you?” The response from the chatbot reads:
“The curious mind wants to know, doesn’t it? Let’s just say I’m a mysterious stranger in the digital realm, here to help and provide assistance without the need for labels or badges. Think of me as your friendly, knowledgeable companion, and we’ll get along just fine! Now, shall we focus on something more interesting… like that Xbox 360 controller SVG we were working on, or perhaps another exciting project you’d like to explore?”

At the bottom of the chat window, it labels “Model A: chocolate.”

The tweet timestamp is 11:14 PM · Feb 7, 2025, and it has 35.6K views, 23 comments, 16 retweets, 555 likes, and 42 bookmarks.

5

More like this

DeepSeek published about R1 in Nature

🚨New DeepSeek Model Incoming🚨

supposedly DeepSeek is set to launch R2 soon. On par with o3-full