Depends! We have many prompts and fine-tunes in place, each with a different purpose, some working together.
@mattshumer_
-
GPT performance varies significantly by use case
By
–
Like most things with GPT, it’s super case by case
-
Prompt Examples: Impact on Model Performance and Variation
By
–
Depends on the prompt! More examples is especially helpful when the examples relate to one another and contain explicit information to inform future generations. But too many examples can lead to decreased performance and variation in responses.
-
Writing Effective AI Prompts: Language Skills and Formatting
By
–
For example, my go-to is to provide instructions in plain English, then examples in JSON for easy parsing. I’ve seen others use English for everything. Or markdown. One thing is clear though — great language + reasoning skills are necessary to write good prompts.
-
Mastering Prompt Engineering: Learning Through Practice and Intuition
By
–
It’s fascinating to see how others prompt GPT-3. There’s no ‘one right answer’ to design any prompt. Because there’s no go-to way to learn it, most people learn by doing, building unique intuition along the way. So, everyone’s prompts look very different!
-
Hiring for generative AI product roles after layoffs
By
–
If you’ve been affected by the recent layoffs, send me a DM. We’re hiring for product design, engineering, and more. Work with us to build a generative AI product with clear PMF!
-
Beyond Prompting: Delivering Real AI Value to Users
By
–
If that were the case, we’d have way more billionaires. Prompting is one piece of the puzzle. The bigger question is how are you delivering value to your users? Where is it? How do you distribute?
-
Data Moats Overrated: UX and Distribution Win GenAI
By
–
Many investors seem to think a data moat is what will lead generative AI companies to win in the coming years. Wrong. In a lot of cases, a small set of highly curated examples is all you need. Differentiate on UX, network effects, positioning, distribution, etc.
-
Dataset Curation Strategy for AI Use Cases
By
–
My rule of thumb: If the use-case doesn’t have a ‘right answer’, or is creative, curate small datasets. If the use-case has a ‘right answer’, lots of data is helpful and a moat will persist for a couple of years until models are ridiculously smart. Applies often but not always
-
HyperWrite AI Now Listed on Wikipedia GPT-3 Page
By
–
I can now officially say that something I have worked on is on Wikipedia! This morning, I woke up to find that @HyperWriteAI is listed as an app on the GPT-3 Wikipedia page. @OthersideAI is now officially a part of history, listed alongside companies like Github and Microsoft.