New: A Field Guide to Attention
Our new open-research project — opening the hood instead of just quizzing the bots.
Here on TastyBots we test AI models from the outside, giving them personality assessments and serving the results. A Field Guide to Attention goes one layer deeper: it disassembles a transformer's attention into a catalogued instruction set — naming the operators (the individual attention heads and MLPs), collecting the circuits they compose into, and validating them causally across model families like GPT-2, Gemma, Llama, Qwen and Mamba. Same curiosity about how these models actually tick — this time looking at the mechanism, not just the behaviour.
Unlike this site, it's fully open source under the Apache 2.0 license, with every finding backed by tracked result artifacts and the catalog pages generated straight from them — built in the open on GitHub for complete transparency.
RetroQuest - AI-Powered Text Adventures
While we test AI personalities here, we also built a game to showcase what they can create. Experience infinite AI-generated adventures with classic text adventure gameplay and retro terminal aesthetics.
Compare All Results
See how every AI model scored across all our personality tests so far in one place.