Despite recent advances in AI, the development of systems capable of executing complex, multi-step reasoning tasks involving multiple tools remains a significant challenge. Current benchmarks fall short in capturing the real-world complexity of tool-use reasoning, where verifying the correctness of not only the final answer but also the intermediate steps is important for evaluation, development, and identifying failures during inference time. To bridge this gap, we introduce ToolComp, a comprehensive benchmark designed to evaluate multi-step tool-use reasoning. ToolComp is developed through a collaboration between models and human annotators, featuring human-edited/verified prompts, final answers, and process supervision labels, allowing for the evaluation of both final outcomes and intermediate reasoning. Evaluation across six different model families demonstrates the challenging nature of our dataset, with the majority of models achieving less than 50% accuracy. Additionally, we generate synthetic training data to compare the performance of outcome-supervised reward models (ORMs) with process-supervised reward models (PRMs) to assess their ability to improve complex tool-use reasoning as evaluated by ToolComp. Our results show that PRMs generalize significantly better than ORMs, achieving a 19% and 11% improvement in rank@1 accuracy for ranking base and fine-tuned model trajectories, respectively. These findings highlight the critical role of process supervision in both the evaluation and training of AI models, paving the way for more robust and capable systems in complex, multi-step tool-use tasks.

One upshot of the current glut of streaming platforms is a flood of programming to fill them: something for every attention span, something to plug every potential gap of viewer inactivity that might render a certain streaming service irrelevant while some other service pulls ahead. And so stories get told and retold. The romantic comedies begin to feel the same. The dating reality shows rely (often successfully, it must be said) on the same dramatic tricks. Another consequence of this, for better or worse, is that the stories being told are pulling from more immediate memory.

Though you couldn't tell from the picture, these particular headphones incorporated a miniature fakir's bed of soft plastic spikes above each ear, pressing gently into the skull and delivering pulses of electric current to the brain. Made by a Silicon Valley startup called Halo Neuroscience, the headphones promise to "accelerate gains in strength, explosiveness, and dexterity" through a proprietary technique called neuropriming. "Thanks to @HaloNeuro for letting me and my teammates try these out!" McAdoo tweeted. On Thursday night, McAdoo and his teammates will seek the eighty-ninth and final win of their record-breaking season, as they defend their National Basketball Association title in Game 6 of the final series against LeBron James's Cleveland Cavaliers. The headphones' apparent results, in other words, have been impressive.