Introducing Invent a Dataset. Describe the dataset you need. Get a structured, training-ready dataset back. No existing data needed. Dataset creation used to start with collection. Now it starts with specification.

Read the full announcement:

“Most of the world doesn't have data.” 🎯 Off-the-shelf data misses African realities (fintech, agritech, climate). Turning prompt intent into training-ready instruction sets & preference pairs directly in @adaption_ai unlocks huge potential for localized AI. 👏

Can't wait to see my imagination come to life!!!

Love to see it! We deployed something similar about a year ago. It didn't have the stickiness we anticipated. Happy to connect on lessons learned.

guys can you please solve multilinguality, tried adaption on Arabic and it sucks

Surely this is only good as the underlying data they have?

that's an insane addition 🔥

Invent a Dataset by @adaption_ai is the best way to look at systems and go wow, how they keep adapting with each invention , and this very speaks volumes of far we can stretch and still be surprised there’s more we can still figure out!! All the congratulations to @sarahookr @adaption_ai family this is really huge

you could not only imagine it but create it . excited

Amazing, good.luck

Can’t believe I choose to watch from sideline 👀🥹 Time to get my lazy a*s up here and bring my idea to life

@mindofjota

Very cool, a lot of the health data companies have similar offerings to sell to life sciences. E.g. Tempus has Lens (describe the cohort you want in plain text, get a structured cohort back with visualization tools)

Specification-first dataset creation. Game changer 🔥

A great solution for accelerating AI development and experimentation.

this is very cool

@Scobleizer holy shit this makes so much sense

INSANE! Let's gooo

@sarahookr excited to try this out i’m interested to see if i get any refusals lol

This is awesome.

been waiting for this

No existing data needed, that's the wild part

The start of synthetic data era

The idea of generating training ready datasets from a simple specification is genuinely interesting. This could make dataset creation much faster for AI builders. I’d definitely like to try Invent a Dataset and see what I can build with it. 🔥

Huge milestone, congrats team 👏🏾

whoa this is cool! curious how y’all maintain parity with real world data since it’s often so unpredictable or incorporate human feedback to improve these generated datasets?

any risks like biases or overfitting?

sadly won’t work if i want to generate low resource ASR data (minority languages), extremely happy if you point out otherwise

Interesting. So generating synthetic data with your harness.

@BrianRoemmele

Better no data than synthetic datam. There is no real world correlation that's why

This is massive! One of the biggest gaps just got closed. Well done @adaption_ai team