Platform / reputation lock-in is going to be a substantial factor, here, especially as AI grows in prominence and people start to emotionally or tribally ‘identify’ with brands. While I have many complaints about OpenAI, canning 4o and the marketing approach it represented was, in retrospect, a significant sacrifice in pursuit of the common good.
I’m not a heavy user of AI coding, but I’d expect that Codex and Gemini would do okay on the software engineering / RSE tests that Claude’s been put through, based on my experience testing them against hard engineering problems and their benchmark performance. A substantial share of Anthropic’s ‘vibes’ advantage right now comes from the fact that they’ve been more effective in building the kind of infrastructure that people want for these kinds of tasks, rather than anything directly tied to their LLM’s abilities. For example, I set up Claude for autoresearchthe other day to test it out, and doing so was a very quick, very seamless experience with lots of online references.
Platform / reputation lock-in is going to be a substantial factor, here, especially as AI grows in prominence and people start to emotionally or tribally ‘identify’ with brands. While I have many complaints about OpenAI, canning 4o and the marketing approach it represented was, in retrospect, a significant sacrifice in pursuit of the common good.
I’m not a heavy user of AI coding, but I’d expect that Codex and Gemini would do okay on the software engineering / RSE tests that Claude’s been put through, based on my experience testing them against hard engineering problems and their benchmark performance. A substantial share of Anthropic’s ‘vibes’ advantage right now comes from the fact that they’ve been more effective in building the kind of infrastructure that people want for these kinds of tasks, rather than anything directly tied to their LLM’s abilities. For example, I set up Claude for autoresearch the other day to test it out, and doing so was a very quick, very seamless experience with lots of online references.