There are, however, two key limitations with these setups as shown by a case study in Figure 1.

Aiming to address this limitation, we present Easy2Hard-Bench, a consistently formatted collection of 6 benchmark datasets spanning various domains, such as mathematics and programming problems, chess puzzles, and reasoning questions.

We introduce Parsel, a framework enabling automatic implementation and validation of complex algorithms with code LLMs.

OpenAI's President Gave Millions to Trump. He Says It's for Humanity In an interview with WIRED, Greg Brockman says his political donations support OpenAI's mission--even if some employees at the company disagree. OpenAI's president and cofounder Greg Brockman doesn't consider himself political, which is surprising, because he was one of President Trump's biggest individual donors of 2025. Greg and his wife, Anna Brockman, gave $25 million to MAGA Inc--a super PAC that supports President Trump--in September of last year. The pair also gave $25 million to a bipartisan AI super PAC, Leading the Future, which says it plans to oppose politicians that jeopardize Americans' "ability to benefit from AI."

First, by using LFMs to identify desirable behaviour to imitate, we improve in task-completion rate over strong behavioural cloning baselines on three distinct language grounding environments (Touchdown, ScienceWorld, and ALFWorld). Second, imitation learning using LFMs outperform using LLMs as experts to directly predict actions, when controlling for the number of LLM output tokens.

The image pairs are mined from existing datasets.