OpenAI has officially released GPT-4.5, its latest language model, in February 2025. The model focuses on enhanced emotional intelligence (EQ) and improved conversational capabilities. It is described as more empathetic and better at tailoring responses to user intent, offering a more natural and personable chat experience. GPT-4.5 is reported to be less prone to hallucinations, with improvements in factual accuracy and social awareness. However, it does not represent a major leap in reasoning capabilities compared to previous models, instead emphasizing refinements in user interaction and emotional understanding. OpenAI CEO Sam Altman noted that the model 'feels like chatting with a thoughtful person.' The model is currently available only to ChatGPT Pro users at a cost of $200 per month, with plans to expand access to Plus users in the future. Advanced Voice Mode and other features are not yet included. Observers suggest that GPT-4.5 provides a glimpse into the direction of future models, including the upcoming GPT-5.
Very interesting finding and thread from @littmath AI is excelling at professional-level math research questions far greater than expected but this is mainly through superhuman knowledge retrieval, not novel + creative mathematical reasoning https://t.co/Nf1NGAyV6A
Interesting thread from Daniel Litt on what it takes to do real research in math for an AI and a human I guess we had similar thoughts this week as it's really aligned with my AI-Einstein short essay: https://t.co/8LcCZNwl6K https://t.co/YyQksXLoOF
[CL] From Language to Cognition: How LLMs Outgrow the Human Language Network B AlKhamissi, G Tuckute, Y Tang, T Binhuraib... [EPFL & MIT] (2025) https://t.co/Z0PpQ3ouWL https://t.co/bggXP6XNzF