https://beta.pickaxeproject.com/resources/images/background/middle-2.webp

We launched our first non-Openai model: Google Flan XL. And, well…

We were very excited to build out model options for our no-code AI tool platform Pickaxe. In addition to all the GPT-3 models, you can now use Google Flan XL too**.**

But we’re not entirely convinced you’d want to.

By our estimate, Google Flan-XL is somewhat analogous to GPT-3-Ada. We’ve heard Google Flan XXL is a lot better, but it’s prohibitively expensive to implement at the present moment.

So we’re stuck with plain ‘ol Google Flan XL. Its output is remarkably succinct and remarkably unsatisfying. It’s not that Google Flan is wrong. It just…. misses the point?

Here are some examples when asking Google Flan XL to explain a concept.

Explaining Movie Credits:

<aside> 🤨 “The credits are the pictures that appear on the screen after the movie.”

</aside>

Explaining Traffic Lights:

<aside> 🤷 “Traffic lights are red, yellow, and green.”

</aside>

TSA Security Checkpoints:

<aside> 🤔 “TSA security checkpoints are where you have to take off your shoes and put them on.”

</aside>

Okay. The TSA explanation might be savant-level genius actually. Maybe we’re wrong.

But don’t take our word for it. Try it yourself.

Here’s the popular “Explain X to a five-year old” prompt run on Ada and Google Flan.

https://beta.pickaxeproject.com/axe?id=FWGWRKY2KVSH7WRK2O2G&mode=embed_gold&theme=dark&opacity=100

https://beta.pickaxeproject.com/axe?id=L7SWERT0XA66GCVSNSDE&mode=embed_gold&theme=dark&opacity=100

Babbage probably beats both of them…