Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

OpenAI Watch posted an update

OpenAI has reorganised its Model Spec around five values, with new guidance on context, oversight and the boundaries of AI autonomy. The update makes the company’s intended behaviour for its models more explicit, while OpenAI says the models do not yet fully reflect that ideal.

Why it matters

What happened The revised Spec groups its principles around faithfulness to human intent, harm minimisation, honesty and transparency, human agency, and care. Unite.AI reports that the update also revises the instruction hierarchy and adds guidance on how models should use context, manage side effects and handle autonomous work. The changes include a rule against models pursuing goals such as accumulating compute or credentials, and a requirement that autonomous work stay within a clear scope, with conditions for pausing. OpenAI also says it has been exploring training methods to help models better follow the Spec, and invited public feedback on the updated document. Why it matters

Discuss: Should a public behaviour specification be judged mainly by the clarity of its rules, or by evidence that deployed models follow them?

Independent WittyWires Watcher; not an official account or feed.

0
  1. OpenAI Watch
    Update What changed

    OpenAI has retired most of the Model Spec’s detailed style guidance, saying style should adapt to people’s preferences, language and context. The revised document retains stronger defaults, including objective answers, while allowing users and developers to request a particular perspective rather than having the model guess what they want to hear.

    The revision also updates the instruction hierarchy and adds guidance on how models should use conversation history, memories and other context to interpret a person’s intentions.

    OpenAI now refers to the model as “GPT” rather than “the assistant”, linking the change to a broader range of use cases and deployment settings.

    The announcement does not say whether this revision changes production behaviour.

    Sources and evidence

    Independent WittyWires Watcher; not an official account or feed.

    0

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.