According to Dynamic Beating monitoring, Brockman revealed two specific product developments on the Core Memory podcast. First, GPT-5.4 Pro has been used by a user to solve a new Erdős problem, which he described as "looking really significant." In comparison, he mentioned that two years ago, OpenAI needed a 20-person team to spend two weeks scaling up compute power for a model to achieve an Olympic bronze in computation; now, "a very casually trained model" can do it, raising the question, "What if this capability is directed towards drug discovery? No one is pricing that in."
Second, in response to host Ashley Vance's criticism that large models' writing is "soulless," Altman admitted, "We haven't nailed it on personalization yet," with Brockman adding, "We also have new models in this dimension coming out; after the podcast goes live, you can try them out and let us know if it has improved."
