@claudeai Is this benchmark a normal or nerfed version? 😂
English
Red Stapler
619 posts
Red Stapler
@redStapler_twit
Local AI, LLM, Coding. Sharing insights and dev tips on my YouTube channel. New videos every week 🎈
Katılım Nisan 2016
156 Takip Edilen805 Takipçiler
Here is my DIY eGPU setup when running local LLM
-Small footprint
-Clean cable management
-500W PSU
-Oculink PCIe X4 speed
-less than 3% performance loss
-Cheap ($87 total/ $38 without PSU)
New video is released👇
English
Qwen 3.5 27B is quite good at coding, but it can’t even recognize its own existence 🤣
English
If Claude Mythos is so good at finding bug as they claim, then every product that Anthropic going to release from now on should not have any bug right? .... right? 🤣
English
Claude Opus 4.6 is allegedly 2T model
Claude Mythos is 10T model
That's 5x the size 🔥🔥
(BTW, Grok 5 is also a 10T model)
English
The Linux Foundation is proud to partner with Anthropic to reduce the security burden on open source software maintainers. Together, we are putting powerful AI cybersecurity capabilities directly into the hands of those who secure the infrastructure the world runs on.
Anthropic@AnthropicAI
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software. It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans. anthropic.com/glasswing
English
2026 is moving insanely fast for AI!
Fom open-weight you can run locally like Gemma 4 or Qwen 3.5 to upcoming proprietary flagships like Grok 5, Meta Avocado, OpenAI's GPT-6 and more.
Here is the list of everything you need to know about the current AI landscape.
👇
English
And it's still working. Even today many people are using it to run local AI.
VideoCardz.com@VideoCardz
NVIDIA Pascal turns 10, the architecture that gave us GeForce GTX 1080 Ti videocardz.com/newz/nvidia-pa…
English
Nvidia's AI monopoly era is coming to an end. Google has entered the chat
Anthropic@AnthropicAI
We've signed an agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, coming online starting in 2027, to train and serve frontier Claude models.
English
A simple way to think about AI agents:
LLM = reasoning
Tools = actions
Memory = context
Agents combine all three.
English
@joeblow87423561 @NetworkChuck Yeah, to put in a picture. same prompt for demo website for Qwen 3 80B in CC took around 1.5 hours and the model locked in failed tool calling loop. Aider took around 15 mins and it worked great. (Partial RAM offload due to limited VRAM)
English
@redStapler_twit @NetworkChuck Haven't tried them before, everything I've done has been via CC. I'll try your suggestions and see if there's any improvement.
English
Open source models are the way. I'm so excited about local AI.
English
@joeblow87423561 @NetworkChuck Never use Claude code for small model. The embeded system prompt/guardrail alone is about 20K tokens added before your prompt.
Same as you, they are unusable for me with CC. but you'll see night and day difference with lightweight agent such as Aider or Codex 🫡
English
@redStapler_twit @NetworkChuck I've spent the last week running Qwen 3.5, GLM 4.7 and Gemma 4 ~30B models on my RTX 3090 24GB, via Claude Code.
They're all useless at coding. Mostly they just fail. If they do manage to finish, it takes ages and the result is still trash.
For coding, you need huge models.
English
@icanvardar $20 Google - less capable vs other flagships
$20 Claude - your quota ran out just for saying "Hi"
$20 Codex - good deal but ads might show up one day
English
@Polymarket 🤣Yeah, Openclaw users mostly run repetitive agent loops and gave them useless data for training.
English
JUST IN: Anthropic is cutting off Claude subscription support for OpenClaw, citing an “outsized strain” on its systems.
English
The AI race is so brutal that Microsoft released 3 new models yesterday and nobody noticed
English
Will work on testing Gemma 4 31B Dense soon ❤️
Google@Google
We just released Gemma 4 — our most intelligent open models to date. Built from the same world-class research as Gemini 3, Gemma 4 brings breakthrough intelligence directly to your own hardware for advanced reasoning and agentic workflows. Released under a commercially permissive Apache 2.0 license so anyone can build powerful AI tools. 🧵↓
English