Could Colin Kaepernick's new AI startup Lumi Stories use ethical AI models? 🗣️
Awareness of biases is good. We also need broad awareness of ethics in sourcing content for foundational AI models & platforms. Lumi could be a shining example. (Audio; 3:41)
This past week (on July 24), civil rights activist and former NFL quarterback Colin Kaepernick announced a $4m round of VC funding for his new AI startup, Lumi Stories1 (lumistories.ai). Kaep became well known for his 2016 attempts at respectful protest in support of Black Lives Matter; thousands of articles, most of which he didn’t control, were written about him. The press release 2 notes that he felt he lost control of his story on why & how he protested, and he ran into gatekeepers and obstacles while trying to get his story out. This inspired his mission with Lumi for helping everyone control and share their own stories. Lumi’s stated goal is to “enable you to create, publish, and monetize your stories all in one platform.”
The TechCrunch article on the announcement 3 highlights that Lumi will use multiple foundational models and that Kaep is aware of risks of biases in AI (which is good):
Kaepernick’s business will use a few different foundational AI models, but he’s keen to address the biases that have plagued LLMs for years. “AI models as they exist today, they’re going to have the biases of the past built into them. We’ve seen that, we know that,” says Kaepernick. “When we think about narratives and how people are represented, AI is going to accelerate things exponentially. That can be both exponentially good and bad. Our hope here is that we build something to create that better path forward for storytellers and creators.”
However, aside from mentioning biases, none of the announcement articles or Kaep’s LinkedIn posts 4 say anything about the ethics of the foundational models they intend to use, or show awareness of theft of content used to build many of the best-known platforms. This Fast Company article 5 makes 2 relevant points:
“The company is also working on ways to compensate users if their work is used to train future AI models.
Right now, Lumi is using a mix of specially trained third-party models and in-house technology.”
The Lumi Stories website also does not have information on the LLMs or AI technologies they will use. I contacted Lumi’s Engineering lead, Stefan Dasbach 6, via LinkedIn on July 28 to inquire about the LLMs that Lumi will use. He has not yet responded.
My hope is that startup founders like Kaep who value equity and ethicality will be open to ensuring that their new startups use ethically sourced and trained AI models.
Ethical foundation models ARE out there, e.g. KL3M 7, which is “the first language model to obtain the Fairly Trained L Certification” 8 from Fairly Trained 9, the startup established by earlier this year.
As I’ve posted before, I believe that companies which resist QDPs (questionable data practices) 10 deserve extra recognition and consideration. It would be amazing for Lumi Stories to set a shining example of ethical AI usage for others who claim it can’t be done.
I’ll post a further update if I hear back from Lumi about the AI models they’re using. In the meantime, I’ve signed up for their waitlist. You can sign up too, here (using this referral link will help me move up in the waitlist to an earlier ‘wave’).
https://techcrunch.com/2024/07/24/colin-kaepernick-launches-ai-startup-help-creators-storytelling/
Includes pricing info: “Creators can choose from three monthly subscription tiers — $20, $40, and $75 — with more expensive tiers unlocking advanced features for creators.”
https://www.linkedin.com/in/stefan-dasbach/, Lumi engineering leader
May 22, 2024 article where Ed Newton-Rex defines QDPs (questionable data practice companies)
This sounds like an interesting startup. Hopefully, he expands beyond Manga. Thanks for sharing. I signed up for beta access using your referral link!
As of today, Sept. 24, there are still no indications that Lumi is going to take the high road on using ethically trained models. Super disappointing.