Log inSign up
GooGZ AI profile banner
user avatar

GooGZ AI

@PaulGugAI
Insights / commentary on AI experience and research | 30 years in tech & business | AI agent tinkerer | Hermes Agent + OpenClaw | Ex-dev turned AI dad.
Australia
Joined March 2026
756
Following
1,125
Followers
RepliesArticlesMedia
  • Pinned
    user avatar
    GooGZ AI
    @PaulGugAI
    Mar 30
    In management, a bad hire doesn't just give you zero productivity.. they give you *negative* productivity by dragging the rest of the team down to fix their mess. Agentic AI is no different. If you deploy "underperforming" agents, you aren't just losing time; you're creating a
  • user avatar
    GooGZ AI
    @PaulGugAI
    11h
    Would love to get my 3080 to 20GB!
    user avatar
    David Hendrickson
    @TeksEdge
    Aug 21
    🧟 A modded 2018 RTX 2080 Ti is running Qwen3.8-27B at nearly 48 tok/s. Yes, a 2080 Ti. 😂 The trick is that this one has been modded from 👉 🎮 11GB → 22GB VRAM And it's running: 🧠 Qwen3.8-27B 📦 Q4_K GGUF 👉 16.8GB 🔥 Fully GPU-offloaded 🦙 llama.cpp ⚡ Built-in MTP
    ai-muninn.com/en/blog/chat-template-kv-mtp-free-speedup
  • user avatar
    GooGZ AI
    @PaulGugAI
    20h
    OMG - the difference in 9B capability in a few months is stark. I have speed and workflow related tests which derive from my most typical agentic oriented use cases, and enjoy seeing how local models progress in the ways that are relevant to me. Testing on my modest RTX3080
    Image
  • user avatar
    GooGZ AI
    @PaulGugAI
    23h
    I enjoy his streams a lot, but Matthew can sometimes land like a ‘mini Alex Finn’ with the back and forth on his position around the labs. Given how fast things are changing it’s entirely plausible, but also there is also motivation to be like that because it farms attention. It