Puck News Covers Poetiq's Six-Person Team Beating Google at AI

Puck's Ian Krietzberg profiled our portfolio company Poetiq in a deep dive on how co-founders Shumeet Baluja and Ian Fischer, both former Google DeepMind researchers, built a recursive self-improvement system that achieved one of the highest scores ever recorded on the ARC-AGI-2 benchmark.

The story's headline captures the wildest part: Poetiq accomplished this in six months with a six-person team and only $40,000 in hardware costs, while Anthropic, Google, and most of OpenAI's attempts scored lower on the same benchmark. The team didn't train their own LLM, didn't buy racks of GPUs, and didn't raise a frontier-lab-scale round. They built a system on top of existing models and let it recursively improve itself.

Read the full article on Puck News.