Sushaantu

I’ve been thinking about defaults - how they quietly decide outcomes - for a long time. I finally published a first draft of The Default Book. AI helped me accelerate the writing, but the ideas have been fermenting for years. Curious to hear what others think. 📖 thedefaultbook.com

@thekitze Codex loves Cardception I switched to gemini for frontend which does way better job than Codex

codex just LOVES nesting card layouts 💀

Every single post I see on my "for you" timeline has AI created replies

For a second I thought it was Santiago, Chile

@JohnnyNel_ Totally - I also like the benefit of using new and upcoming models you only get to see inside opencode

@Su doing the same thing... one workspace beats juggling a bunch of subscriptions. consolidation is so underrated

Cancelled Codex and Claude Code plans and got me Copilot Pro+ Using it for coding/chat with VS Code, OpenCode, Pi, OpenClaw and Hermes. Excellent value!

Did my first ML fine-tune this morning. Trained Llama 3.2 3B on my own bank transactions (1,200 labeled examples, 37 expense categories) using QLoRA on an RTX 5060 Ti. Result: 60% accuracy. My 300-line regex gets 100% on known merchants.

Still surprises me how many people don’t know what happened with radiologists and AI

Milei is correct - you can't negotiate with the left

Context:

The political left in Latin America creates problems for the political right to fix which requires budget cuts. Political left uses Palestinian style rejection by putting children/students on the forefront. Classic.

Openclaw has a been super useful for my workflow but pretty much every update breaks something and I have to be very cautious about tokens running out. *pointing out 2 areas where alternatives can target and make a space for themselves*

💯

OpenAI Symphony is good. Really good.

@_tenZdhon_ Remnants of when not everyone had an ID card to show their age

Just found out India has height based pricing Under 4.5 feet and you pay less Might be the new pricing strategy for consumer apps

@snowclipsed It used to be the other way round until a month ago

opus 4.6 as planner and 5.3-codex/5.4 as executor is pretty sick

@nateliason Same. Just switched to Opus in the end.

Codex 5.4 is such an amazing coder but so hilariously bad in OpenClaw. For some reason it is ~4-5x more likely to say "I'll do X" instead of actually doing it compared to Opus. Even Kimi & MiniMax do a better job here.

As long as a human is involved in decision making process - the AI race will come down to just vibes - agents with better vibes and intelligence will win. Intelligence alone might not be sufficient

Anthropic is still the goat when it comes to personality of its agents. It feels like talking to a really smart helper who gets you.

NotebookLM, like most AI tools is great when you need to condense information. It absolutely blows when you give it a small idea and ask to expand on it.