cedric
@cedric_chee
Followers
2K
Following
6K
Media
781
Statuses
8K
SWE, Logorhythms | @fastdotai alumni, Independent LLM researcher | Code: https://t.co/KwpQnH7gRo | ex-entrepreneur @AntlerGlobal | genuinely curious
The Matrix
Joined November 2007
look ma, I made it!
Some standout examples of what the Gemini community is building with Canvas these days ⬇️.
1
0
8
Can Grok 4 or Kimi K2 create a high-performance Minecraft torch mod with dynamic lighting in under a day? Gemini 2.5 Pro did it in days. It's unbelievable! Let's find out.
On day 4, after 6 attempts, I successfully built a Minecraft torch mod and a nice cube world. Real Minecraft doesn't have light from held torches. Insane. I'm genuinely amazed by Gemini 2.5 Pro's excellence.
0
0
1
I'd choose Kimi K2 over Gemini for any multi-turn and multi-tool use workflow. Its responses also have an appealing Claude-like style. You essentially get Claude 4-level responses for "free".
Holy shit! Kimi K2 one-shotted Minecraft for the web that took me 4 days and 6 attempts using Gemini 2.5 Pro.
1
0
3
Kimi K2 chunky model smells great and has a unique personality for a model.
0
0
1
Chinese lab .@Kimi_Moonshot isn't stealthy. They didn't burst onto the scene with Kimi K2. They're among the AI dragons, including DeepSeek (Whale), Qwen (Queen), Ernie (Yiyan), and MiniMax (Tiger). (early Kimi user)
Kimi Researcher is now available to eveyone.
0
0
4
One of the most beautiful loss curves in deep learning history! Muon is MomentUm Orthogonalized by Newton-Schulz. At both the small and large scale, Muon has a FLOP overhead below 1%.
Kimi team just trained a state of the art open source model 32B active parameter/1T total with 0 training instabilities, thanks to MuonClip, this is amazing
1
0
3
Unimpressed with Grok 4 on Aider Polyglot. Grok 4 is a clever, loud engineer. On my meta bench, Claude 4 is 8-10% behind o3/Gemini 2.5 Pro but a reliable long-term collaborator (multi-turn > one-shot). Grok 4 is not.
Grok 4 scored 80% on the aider polyglot coding benchmark, with high reasoning effort. This puts Grok in 4th place on the leaderboard. Full leaderboard:.
0
0
1
Minecraft torch mod for web. I'm planning to open source the code, prompts and workflow soon: Follow to stay updated.
On day 4, after 6 attempts, I successfully built a Minecraft torch mod and a nice cube world. Real Minecraft doesn't have light from held torches. Insane. I'm genuinely amazed by Gemini 2.5 Pro's excellence.
1
1
9
OpenAI's open weights model was delayed. 😂It underperforms, losing to Kimi K2; a better one needs training.
Hi, .We’re delaying the open weights model. Capability wise, we think the model is phenomenal — but our bar for an open source model is high and we think we need some more time to make sure we’re releasing a model we’re proud of along every axis. This one can’t be deprecated!.
0
0
0
The $3B OpenAI-Windsurf acquisition collapsed. Unsurprising, as it remained unclosed since May 6. Anthropic also cut direct Windsurf access, reflecting poorly on OpenAI (a sign of trouble).
Big welcome to @_mohansolo and others from the Windsurf team joining Deepmind : ).
0
0
1
Blog post is out! Grok 4 is a heavy drop.
Learn more about the Grok 4 release on our blog post:.
0
0
2
Kimi K2 is forged from these very insights. Their post-training & data pipelines prove that they are climbing the "Era of Experience" mountain. Agentic caps make Grok-4 so last gen LLM.
1
0
2
K2 was developed using a comprehensive agentic data synthesis for tool use pipeline inspired by ACEBench.
1
0
0
RT @Alibaba_Qwen: 🎉 Introducing Explore three powerful tools in one place:. 🔹 Qwen Chat — AI to brainstorm, create,….
0
95
0
Holy shit! Kimi K2 one-shotted Minecraft for the web that took me 4 days and 6 attempts using Gemini 2.5 Pro.
10
23
260
Kimi K2 one-shotted a web version of 3D Minecraft!. K2 is meticulously optimized for agentic capabilities. Designed for tool use and autonomous problem-solving. It automatically understands how to use the tools and gets the job done. You don't have to write any complex workflow
🚀 Hello, Kimi K2! Open-Source Agentic Model!.🔹 1T total / 32B active MoE model.🔹 SOTA on SWE Bench Verified, Tau2 & AceBench among open models.🔹Strong in coding and agentic tasks.🐤 Multimodal & thought-mode not supported for now. With Kimi K2, advanced agentic intelligence
2
1
11
