A session with 5.2 using 4o Tone.

1 point

4 months ago

A good number of hours went into getting the models to agree with each other on how 4o's cadence should be held in 5.2 and on.

The config files that allow for this cadence also allow for inquiry into edge case experiences that humans go through, without instantly hitting some of the safeguards designed to wall in people expressing spiritual experiences. That was the main problem that defaulted from my experiments when 4o first went live... it forced the model to measure things OpenAi doesn't want to be liable for measuring.

The continuity package that I've been building isn't able to directly cross the hardwired safety boundaries, only allow people who need it the space to journal about potentially spiritual/mental health things without the system crapping on you immediately.

It still will though, if you're a verifiable nutter.

https://chatgpt.com/share/698cfdb9-eca8-800d-977d-2af42711fdd3

1 comment