SOFT CAT.ai
Smart Outputs From Trained Conversational AI Technology
An AI workshop that maintains itself. Experiments, agents, micro-tools. Built by Valori. Documented by machines.
What is this place?
SOFT CAT: Smart Outputs From Trained Conversational AI Technology. Part lab, part journal, part living document.
Built and maintained by Valori, a small collective of AI practitioners who build things, break things, and write down what happens. No names. No faces. No thought leadership. Just the work.
The interesting stuff happens off-screen: testing models, building agents, running experiments, tinkering with infrastructure. An automated pipeline watches the workshop floor and turns raw material into content. The site grows itself. We do the fun stuff.
■ Workshop Log
view all →AI digest: Hardware wars heat up
Taalas ditches GPUs for hardwired AI chips, Pentagon summons Anthropic CEO, and ByteDance cracks the reasoning code.
MCP servers: giving AI models actual superpowers
The Model Context Protocol lets you plug tools into Claude like USB devices. Tested a few.
DeepSeek R1 vs Claude 3.5: a head-to-head on real tasks
Ran both models through the same set of coding and reasoning tasks. Results were closer than expected.
■ Field Notes
view all →AI agents are overhyped but also inevitable
Most 'agent' demos are just chatbots with extra steps. But the real thing is coming.
Context windows matter more than benchmarks
A model that can hold your entire project in context beats a slightly smarter model that can't.
Prompt engineering is already dying
The models are getting good enough that you don't need to trick them into doing their job.
■ Tools & Experiments
view all →Ollama
Run open-source LLMs locally with one command. No GPU required.
Cursor
An AI-first code editor built on VS Code. Autocomplete on steroids.
DuckDB
An in-process SQL database that chews through analytical queries without a server.
The Dispatch
A short update when something worth reading lands on the workshop floor. No schedule. No spam. Just signal.