ThinkingCap cuts AI reasoning tokens by up to 58% with basically no accuracy loss.
We break down how BottleCap trained it and ran it ourselves against the base Qwen3.6-27B model to see how good it is.
Better Stack
@BetterStackHQ
Tracing reimagined with eBPF and OpenTelemetry.
Instrument your clusters without any code changes.
Get an AI SRE and a robust MCP server.
Joined January 2021
- Decimen sends files using nothing but a screen flashing QR codes and a camera reading them back. No WiFi, no Bluetooth, no network at all. We tested it and got 3 KB/s instead of the claimed 128 KB/s. We dug into the source code to find out why.
- Postgres 19 runs graph queries natively, so the five table joins you write to answer one question is now optional.
- html-in-canvas is basically shaders for HTML, and Canvas UI is and awesome library that makes it easy to actually try.
- Turbo Fieldfare runs a 26B model on a Mac in 2GB of RAM at 23 tokens per second. Most of the model never leaves the SSD.
