Keep pulling the thread on Mark Chen, Jakub Pachocki.
OpenAI's primary research goal is to create an "automated researcher" capable of discovering new ideas.
OpenAI's primary goal with the GPT-5 launch was to make advanced reasoning capabilities widely accessible.
A surprising capability of GPT-5 before its release was its significant advancement in solving problems in hard sciences like physics and mathematics.
GPT-5 Pro is capable of automating scientific tasks that would typically take a graduate student months to complete.
Mark Chen predicts that OpenAI's models will see substantially greater capability improvements over the next year compared to recent advances.
Jakub Pachocki of OpenAI does not believe the AI field will become data-constrained soon and expects compute to remain the primary bottleneck for progress.
Jakub Pachocki predicts that robotics will become a major focus for OpenAI in the near future, making physical constraints like energy consumption a key consideration.
OpenAI's future evaluations and milestones for its models will focus on demonstrating economically relevant discoveries.
Prior to GPT-5, OpenAI had an "O series" of models designed for long, deliberative thinking to produce high-quality answers, distinct from the instant-response GPT series.
OpenAI's current models, which are approaching mastery of high school level competitions, can reason autonomously for a duration of one to five hours.
OpenAI's research is focused on extending its models' autonomous reasoning horizon by improving long-term planning and memory retention.
OpenAI's focus on fundamental research and innovating at the frontier, rather than copying competitors, is a key factor in retaining top talent.