Inside The Startup Building Uncensored AI (Abliteration AI) | EP 2345

Jason Calacanis (0:00)

All right, everybody, welcome back to This Week in Startups. I'm your host, Jason Calacanis, with me today.

Very lucky to have Devon Thomas with us. Now, Devon, you and I met because you came to Launch Festival, just like a little event I do for founders. That's the origin story here. Tell us how we met, and then we'll talk about what you're building.

Devon Thomas (0:21)

Yeah, yeah. So when I first came up with the idea, I was kind of researching some pre-seed funds, and I followed you on X already, and I think you tweeted actually, you were like, oh, I'm excited about what I'm going to be doing for this project of mine, Founder University, and I was just like, okay, this sounds like you're excited about it, so let me see if I can sign up and be a part of it. And I just reached out, and then soon as I kind of engaged with you all, then they were like, oh, we have something coming up called LaunchFest, if you're interested in doing that as well. And so yeah, and then you were there. It was cool. It was a great experience.

SPEAKER_3 (0:53)

This Week in Startups is brought to you by Odoo, the all-in-one business platform. Your first app is free. Get started today at odoo.com/twist. Northwest Registered Agent. Got a new business idea? Northwest Registered Agent helps you bring it to life. Get a free domain, email, phone number, and more with no purchase required. Learn more at northwestregisteredagent.com/twistdomain.

Every, for all of your incorporation, banking, payroll, benefits, accounting, taxes, or other back office administration needs, visit every.io.

Jason Calacanis (1:29)

Amazing, yes. And this is like one of my key tenets is say yes.

So I always tell my team, like just like if you meet a founder, just say yes to meeting, just say yes to having launch festival, letting a bunch of founders come together. You never know what could come out of it. And you've got this great starter, Obliteration. Maybe you could explain what you're building, why it's important.

Devon Thomas (1:51)

Started Obliteration like at the end of last year, I was actually, I built like a home cluster, a CPU cluster, and actually I originally started kind of serving the site from my home cluster at home. But what makes it important is there's a whole bunch of professionals and companies out there that are essentially underserved or unserved by the frontier labs, because whether they're in cyber or they're doing biology research, or they need synthetic data or defense, all these use cases that essentially the frontier labs are essentially shunning. And so our kind of tenet is to try to provide the best models that will allow these use cases to be unblocked. And so we've had a lot of traction, a lot of press all around that, because maybe some people feel like these use cases shouldn't be unblocked, but those professionals seem to be super happy about it. So it's been going well.

Jason Calacanis (2:39)

All right. And before we go to a demo here and talk about this really important issue, I just want to take a minute, put my Plaud pin on. Hey, if your work depends on conversations, interviews, meetings, calls, check out Plaud at Plaud.Twist. Use the code TWIST for 10% off. Great partner of ours here. And so let's talk about it, Devon. The frontier models have full access to the latest and greatest large language models. And then if you're in cyber defense and you are, let's say, hugging face, and OpenAI wants to do a series of attacks, whether intended or not, or they're testing, or white hat, let's put it all aside, they want to test cyber.

They pick hugging face as a target. Hugging face does not have access to OpenAI's latest model. They keep that for themselves. So hugging face, if they wanted to mount a defense against the latest and greatest frontier models, they don't have the same access to the tools, correct?

Devon Thomas (3:42)

Exactly, yep. And even if they have access to those special cyber programs, still, many of them, our customers are still complaining about refusals. And these are big corporations. It's not just startups or individual researchers. It's still big corporations that have access to cyber programs are still having the issues with refusals, just trying to do their job.

Jason Calacanis (4:02)

Okay, so you are taking open-source models, I'm assuming GLM or DeepSea, and you're allowing people to run them in their native form without restriction, correct?

Devon Thomas (4:16)

Yep, yep. So, yep, the process of obliteration is to kind of remove those guardrails. And I always say, like, our unique, I guess, innovation or idea on the space is to allow customers to set their own guardrails. So we have a feature where it's kind of like, I always call it an inversion of control, right? We separate those two layers, right? We have the unrestricted model is one layer, right? Which is like all the labs and open-source models have the refusals backed in. But then we have a software layer where you can kind of define your own policy for your company. So you could have more stringent guardrails than were native, or you could have less in certain areas and stronger in certain areas. And like I said, it's kind of the idea is to give them back control over what they can and can't do.