With EU lawmakers embroiled in a nearly 24-hour debate over the bloc’s Artificial Intelligence Act, it seems perfectly timed for fresh-out-of Y Combinator startup atla to announce its $5 million Seed funding round.
As we’ve seen on numerous occasions, for all the power and potential that generative AI holds, it’s far from infallible.
Not all that glitters is gold
While generative AI images that place Ronald McDonald performing open heart surgery, or a conversation about sunflower oils, which somehow spurred a tangent landing upon GPTs "state of mind" don’t have far-reaching consequences, they do demonstrate that the technology isn’t quite yet ready for the likes of high stakes applications such as those in the legal, financial, and/or medical sectors.
One reason behind today’s generative AI’s hallucinations comes down to the fact that the technology is trained on large language models, textual datasets from sites like Wikipedia, GitHub, etc., and ultimately those authored by humans. And if you’ve used Wikipedia long enough, you’ll be quick to realise that not everything you read is either truthful or completely accurate.
Time is money
Looking to take away AI’s bag of mushrooms is Y Combinator S23 cohort member atla. The company doesn’t cite specifically how, but it says that it’s building a technology that improves truthfulness, reduces harmfulness, and increases reliability for generative AI models.
Ultimately, atla's goal is to shave thousands, if not millions off fees that corporate councils often allot to external law firms. According to atla, legal professionals spend 5 hours reviewing 45 documents to answer a single question, its solution can do the same in seconds.
Proof of concept
The company has already tested its concept in pilot programmes run in conjunction with legal teams at both Volkswagen and N26. Used as a research assistant, atla’s AI assistant provided generated answers cited directly from the world’s most trusted legal sources.
Atla is co-founded by Maurice Burger, a seasoned AI startup veteran who relinquished his pursuit of an MBA at Harvard to obtain a Masters degree in Computer Science from the University of Pennsylvania, and Roman Engeler who researched iterative self-improvement of large language models at ETH Zurich and Stanford.
Led by Creandum
Now looking to scale atla’s offer at an accelerated pace and commercialise, the company’s Seed round was led by Creandum, with existing investor Y Combinator participating alongside Rebel Fund, whose partners include Reddit founder and CEO Steve Huffman, Cruise co-founder Daniel Kan, Instacart’s co-founder Max Mullen, Scribd co-founder Trip Adler, and Sebastian Mejia, co-founder of Rappi.
Lead image via atla.