GPT-6 Astra update
For readers interested in AI economics, Marcus's analysis of OpenAI's GPT-6 Astra model and its financial viability provides a concrete comparison to Claude and other competitors.
Gary Marcus says OpenAI's GPT-6 Astra model shows progress on formally verifiable problems but modest gains on the ECI benchmark. It's not significantly better than the latest Claude, and next major models would be expensive with limited returns. OpenAI faces competition from China, Anthropic, and Amazon/Google, and its agents have major security flaws.
short summary of GPT-6 Astra and why it doesn’t bail OpenAI out of its economic mess: Gary Marcus @GaryMarcus they have made lots of progress (partly using neurosymbolic techniques!) on problems that can be formally verified, but that progress was very expensive and doesn’t translate well outside of benchmarks, and on the ECI the gain was modest. Astra is not that much better than the latest Claude. The next major model would cost much much more, possibly also only for modest gains. They face immense competition, from China and Anthropic and eventually Amazon and Google etc cutting into already negative margins, and their agents have HUGE security problems that seem not competent to solve. OpenAI is master of the press release but their economics still don’t make much sense. And building GPT-7 at the same pace of scaling might not be economically viable. 🔗 View Quoted Tweet 💬 0 🔄 0 ❤️ 4 👀 1626 📊 1 ⚡
