Get ready to hear a lot about robot and AI ‘swarms’

GIZMODO

Startup company Fortytwo published benchmarks that showed its approach of running small AI models on personal computers has managed to outperform some of the latest models in reasoning tests, claiming better outcomes than OpenAI’s GPT-5, Google’s Gemini 2.5 Pro, Anthropic’s Claude Opus 4.1, and DeepSeek’s R1.

The company’s theory, which appears to be borne out in the results, is pretty simple: large AI models can get stuck in reasoning loops when given complex tasks. Smaller models, meanwhile, not only limit those reasoning death spirals but also provide multiple answers that can be ranked to find the best available answer.

Read more | GIZMODO

---