… Cursor has also started training in-house AI models that it can cost effectively serve to customers. The startup recently launched Composer 2, an AI model based on an open-source system from the Chinese AI lab Moonshot AI, that Cursor did additional pretraining and post-training on. …
… At launch, OpenAI said the model “excels at accurately generating and debugging complex code.” Andrey Mishchenko, OpenAI's research lead for Codex, says a key reason AI models have become better at coding is because it's a verifiable task. …
… The startup says it developed a video encoder that can compress videos into an AI model’s context window, which it claims is 50X more efficient than previous computer use models. To show off its AI model’s capabilities, the startup hooked it up to a car, a live videofeed, and a computer keyboard. …
… OpenAI released its first “reasoning” model in 2024, where the chatbot needed “thinking” time. …
… Anything you share in a chat with Meta AI may be stored and used to train future AI models. “We keep training data for as long as we need it on a case-by-case basis to ensure an AI model is operating appropriately, safely, and efficiently,” reads Meta’s privacy policy about generative AI. …
… The researchers analyzed 97,895 AI-related conversations on cybercrime forums since the launch of ChatGPT in 2022 until the end of last year. …
… Like Sun, Newton, and Lorenz, Roose feels that he’s still just better at writing than an AI model. “I think the models tend to be fairly generic and depersonalized, but also, I like doing this,” he tells me. …