O1 (high) -high reasoning effort
complete
t
theunmindful
When O1 was launched, it was set on medium reasoning effort by default for API usage (with no option to change I guess). Now, the reasoning effort parameter could be set higher. O1 is still a wonderful full model and O1 (high) would be lovely.
Currently, Merlin only has O1 with original default settings I guess.
endu
complete
t
theunmindful
endu Can you please guide me where this change appears as I fail to spot it? Thanks.
endu
theunmindful: We've set the reasoning effort to be variable so it can adjust based on the complexity of your queries—this way, it can handle them efficiently and save time. If we fixed it to "high," it would take much longer for all queries. That said, do you have specific use cases where you'd prefer it to always be on "high"? We'd love to hear your thoughts! 😊
t
theunmindful
endu thanks for the update.
I will share my use case with you. I prefer O1 (with high reasoning) in tasks which requires integrative thinking using multiple knowlegde systems (psychology, astrology, philosophy, biology etc).
Imagine engaging with a wise therapist who is an expert in all these fields together. That integrative thinking is like a polygot spiritual person on steroids.
(Thinking loudly now) Going forward, mental health is going to be big with these LLMs. I am told these models (especially gpt 4.5—I second that) are great at someone wise to talk to. Merlin can also think of a dedicated product with different segments like older people, or people with limited English skills. It's like making the best of intellectual/ wise tools to masses.
For that models like O1, 4.5, O3 —which have large data/ inherent knowledge—with high reasoning effort perform better.
Alternatively, a wisdom repository with different knowledge systems including modern psychology—RAG— and a powerful small reasoning model (with high reasoning effort) and ability to write empathetic human-like responses can work wonders.
endu
theunmindful: Got it! Those are excellent use cases! Thanks for sharing your thoughts—it’s clear you’ve put a lot of careful consideration into this. 😊 The idea of combining wisdom, integrative thinking, and empathy in these models is truly exciting, and mental health applications have so much potential.
endu
theunmindful: Sure! The change is in the backend, so there’s nothing for users to tweak just yet since we haven’t created a UI for it. Maybe down the line, we’ll add it to the settings once the UI is in place. 😊 Thanks for checking!