@DanielSmidstrup codex is better, you're more free to use the shit you want to use while claude forces you to use their ecosystem or spend 50$/1M tokens
they really only care if companies start cancelling seats. The subsidized subscription only exists to get people hooked to their ecosystem so that when they go to work they talk to stakeholders about how much they love claude models. They also want to keep max subscriptions at 20-50% used out of the 100% offered. That's why they took openclaw away. It's really too bad because pi was my favorite harness for agentic development. Claude code is actually dog shit and the developers that prompted mythos to make it should be ashamed of themselves.
@MacroBombastic@Web3Marmot A large component of these prices is based on a prediction that companies are going to 10x their revenue. If that doesn't happen then people will sell regardless of what these chips can do.
The revenue is silo'd into a few companies while those that believed they'd 10x-100x are stagnating. The real question is which company will be the first to break out of AI psychosis and realize LLMs and the associated costs will not make their company 100x more productive and 10x more profitable.
@examaddaorg@vivoplt im 90% positive that if a CEO could give himself a higher salary by replacing 8k employees with an LLM they would. Source Mark Zuck
@vivoplt I really think this timeline depends on who you are.. I'm still in a 2023/2025 perspective. I would never blindly trust these frontier models in an enterprise codebase.
I thought his reasoning was that Zig was a language meant for education. Students and those interested in learning programming languages could contribute. The cognitive load that AI PRs created caused the team to miss the PRs that had substance. Then there are projects like NetBSD that are banning AI commits because of licensing concerns. All are valid reasons.
@ThePrimeagen they were working on their 3rd todo list create-react-app then GPT 5.3 dropped and software engineering descended into the depths of hell. Now we have BridgeMind.
idk how people are imagining AI in f500s or f100s.. we're still working on tickets... we just have an artists assistant. It is not generating 100x productivity. In fact, there are people that are trying for 100x productivity but they're the ones that shouldn't be pushing code. They push without understanding, their PRs have no comments just blank approvals and their code is some of the sloppiest shit I have seen in my career.