Log inSign up
GooGZ AI
9,101 posts
GooGZ AI profile banner
@PaulGugAI

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
789
Following
1,242
Followers
RepliesRepliesRepostsRepostsMediaMediaArticlesArticles

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • Pinned
    @PaulGugAI
    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
    11
  • @PaulGugAI
    GooGZ AI
    @PaulGugAI
    1h
    Bonsai 27B 1-bit is touted as the small VRAM GPU king. Can you build a viable personal agent on it?
    Article cover image
    Article
    Bonsai 27B Q1_0: Can 27B 1-bit thinking save a small VRAM GPU?
    I have read this a few times on the tier lists: Under 8 GB, Bonsai 27B 1-bit is the best option. I'm often not sure what that's based on (anecdotes, stats?) but I wanted to find out for myself. If a...
  • @PaulGugAI
    GooGZ AI
    @PaulGugAI
    5h
    Plus Subscriptions Matter!
    Image
    @embw_l0x
    embw_l0x
    @embw_l0x
    5h
    Replying to @PaulGugAI
    Ouch, being on a 20x account I always assume things arent so bad, I dont even know how to use the light version in codex
    2
  • @PaulGugAI
    GooGZ AI
    @PaulGugAI
    7h
    For GPT Plus users out there, a short GPT-6 Astra (light reasoning) anecdote to consider if you are trying to work out quota burn. - I just went from 100% to 65% of my 5h limit with one prompt. - The prompt ran for 1 minute and 16 seconds. - The prompt was a request to review a
    Image
    3
  • @PaulGugAI
    GooGZ AI
    @PaulGugAI
    10h
    I keep hearing Bonsai 27B Q1 is the top-recommended model for older small-VRAM GPUs (like my RTX3080). I'm giving it a real-world trial with my personal agent 'controlled outcome' testing, to see if those benchmarks actually convert. Can it outperform a high-precision small
    Image
    00:00
    3
Advertisement
Advertisement