· 3 min read
KL3M is the first Fairly Trained LLM.
Announcing the certification of KL3M as the first Fairly Trained LLM.
Doing the right thing
Large language models have the potential to transform society for the better. But this potential will never be realized unless we as a society can truly trust the technology.
The seeds of this trust lie in the birth of these models - most specifically, in their training data. And as both research and the proliferation of litigation have shown, current large language models have borne bitter fruit from their tainted seed. Content creators and rightsholders around the world have asked a simple question:
Why should we trust companies whose original act was to steal our work?
Most model companies have decided that the answer to this question is to move quickly, gain market share, ignore ethical and regulatory guidance, and intimidate or outlawyer anyone who stands in their way.
We decided, on the other hand, that it might be better to pursue a different strategy.
What if we just did the right thing from the start?
For us, this was about more than just saying we did something the right way. Trust is about proving your commitment. And that’s why we are proud to announce that KL3M is the world’s first LLM to obtain the the Fairly Trained L Certification, an independent third-party assessment of model training standards.
What is Fairly Trained?
As Wired wrote this January, Fairly Trained is a nonprofit dedicated to auditing and certifying generative AI models. Founded by Ed Newton-Rex, former executive at Stability AI and TikTok, and backed by organizations like the American Association of Publishers and UMG, Fairly Trained has launched the first certification program of its kind to help consumers and businesses identify AI models that have been trained in a way that respects the rights of content creators.
What does it mean for KL3M?
KL3M is the first family of large language model to be certified by Fairly Trained. This means that we have successfully demonstrated our compliance with the requirements for this certification, and that customers can trust that KL3M has been trained in a way that respects the rights of content creators and rightsholders around the world.
More practically speaking, this doesn’t change anything. We will continue to provide customers with representations and warranties that reflect our commitment to legal compliance. We will continue to train our models on content that is cleanly licensed or in the public domain. And we will continue to invest in scaling our models and helping customers fine-tune and extend from this foundation.
While we are proud that KL3M is the first model to demonstrate this commitment, we sincerely hope that we are not the last. We believe that the future of AI must be built on a foundation of trust, and we are committed to working with organizations who share this vision.
Get in touch
To learn more, contact us at hello@273ventures.com or visit kl3m.ai.