shipilovya

Last week OpenAI released HealthBench, the most comprehensive set of evals for health to date. The top 3 scoring models all spiked on different things:

- GPT-4.1 is best when you need a straight answer - o3 is best for complex cases - Grok is best at clarifying important info (“truthseeking”)

Made this prototype mostly to understand HealthBench deeper. I will probably use it in the future products I make.

These 41 samples came from 24 unique products

Interesting, do you feel like your / your founder’s qualities and relationship had any impact on the success of failure of your startup? Would love to hear how things went.

Yep! I discovered it unexpectedly and found it works really well, at least so far. In hindsight it's obvious it would have worked in my case.

I sent it to OpenAI because I felt comfortable doing it with my entries, but my guess is that the same can be done with an open model.

I want to see if it’s possible to have it extract more structured information about my beliefs, values, and thought patterns, and then reference it to non-intrusively comment on my writing.

Let me know if you’re interested in this, I saw your post on journaling and found it thoughtful.