P

CTO at Testamy. Building Manor. Software, agents, and operator tooling

[P @peter_olom]

IMHO, Gemma 4 is google's best AI model/product. I'm more excited when then update it compared to Gemini. Its really useful for automations.

[Google Gemma @googlegemma]

We’re rolling out some big improvements to Gemma 4, fueled by incredible community feedback and contributions!

Here is a breakdown of what’s being fixed and updated in this release: 🧵👇

[P @peter_olom]

@davis7 You and I mostly share same opinion. For me though I'm renting on hetzner because the maths makes more sense from Nigeria. Much faster internet and no power bill. Also running my own custom built oss agent environment with GLM 5.2 + Gemma 4 handling vision

[P @peter_olom]

@forgebitz apt 👌

[P @peter_olom]

@thsottiaux Till date no gpt5.6 for me

[P @peter_olom]

@thsottiaux This is by far my worse codex experience ever and I built stuff with o3 in codex cli and would would take those days over now in a hearts beat

[P @peter_olom]

@thsottiaux Feedback for you... Codex and Sol are so slow I can't even get anything done. I have been sitting here for over an hour for a simple `remove an inner section header` request. Its unusable. I'm a long time pro 20x user for gawd's sake

[P @peter_olom]

@robinebers Great video. Aside Fable's hero, I'll take everything else from Sol

[P @peter_olom]

@embirico @OpenAI Help! Requests are super slow

[P @peter_olom]

@Kagemochi1 Hashirama?? This post was definitely written to trigger me 😅

[P @peter_olom]

@nishffx Scratch that. High and medium are dumb too. GPT 5.5 is just DUMB

[P @peter_olom]

@nishffx Thank you! Happy to see someone with reasonable visibility on X say this. GPT 5.5 right now tbh is comparable to GPT 4.1. I do not kid. I find myself reaching for GLM 5.2 now. Xhigh is dumb, too dumb so I just use high and medium now.

[P @peter_olom]

@simpsoka I simply just say to the model do this work in a worktree so you don't disturb my ongoing work. I find that experience better.

[P @peter_olom]

@MatthewBerman Through your recent tweets I'm glad to now see you as a true comrade of AGI for all. Keep sticking it to the wanna be AI gatekeepers ✌🏾

[P @peter_olom]

@0xSero @davis7 @ollama please please set up this chain in the cloud. GLM 5.2 needs sight

[P @peter_olom]

@thsottiaux Improve creativity. I know it’s a very capable model but 5.5 is now lazy and cannot be creative to save its life. Getting 5.5 to build things with just a little bit of creativity and sensibility feels like trying to milk a chicken

[P @peter_olom]

GPT 5.5 has been real dumb lately. I see clearly now what I want from modern day SoTA. 1) Execute well defined task as handed to you. 2) When asked for reasoning and creative input, show some gaddamn creativity and not be a useless tool!

[P @peter_olom]

GLM 5.2 is ridiculously good at frontend design

[P @peter_olom]

@danielbergholz I normally dont care for benchmarks. It’s a good model

[P @peter_olom]

I have been using GLM 5.2 via OpenCode today and I can say it’s good. Finally I have a model that can compliment GPT 5.5 on UI for me. Excited!

[P @peter_olom]

I regretted it.

[P @peter_olom]

Google choking the AI industry today. Nice one. I just hope the 3.5 family can actually do agentic work and not choke like their predecessors. Upgrading my sub to play with it, hope I don't regret it.