Sabrina Halper Show · 3 min · Oct 15, 2026

Expert data fills gaps the internet was never built to cover

Founder & CEO, Micro1

Founded micro1, pivoting it from AI recruiting to expert training data, growing it past $1B ARR.

Sabrina HalperHost · Host, Sabrina Halper Show
Has worked as an investor at HOF Capital and founded SHS Media.

About this segment

Internet corpora include low-quality and incorrect material, while experts had little reason to publish structured work. Paid pipelines now create data the public web lacks.

From “AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari”, Oct 15, 2026. Watch the full episode

“There's never been an incentive on, you know, the internet for, like, hundreds of lawyers to go and, like, create these really complex data sets”

Ali Ansariat 21:49 · Sabrina Halper Show21:49

Takeaways

Internet-scale text includes incorrect answers and low-quality posts.
Experts rarely had incentives to publish structured work.
Paid expert pipelines produce data public web corpora lack.

Watch 3:05 from 19:40

In this episode

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

19:45
and i guess maybe just like to start with one obvious question is when llms were first created they were mostly trained on the internet do you think that we like have reached everything the internet has to offer. Everything from maybe even like the past couple of years and the future, it's all about like expert training data that you cannot find on the internet. Yeah, the internet is, yeah, it's completely trained on and it is, if you think about like what does the internet include, there's a lot of really bad data.
20:19
Like a large, like a material percentage of internet data is Reddit people like screaming and like cussing and like just saying random stuff, right? And that's not a trivial amount of data that's included in this large corpus. And that's just one example. But there's a lot of that. There's bad blogs. There's answers to questions that are incorrect. So the way to think about this is if we today delivered a data set or environment to a customer that represents anything like the internet quality, it would be the worst data set of all time.
20:57
like you you you'd be you'd be laughed at so so that that just shows that like the internet is um you know sort of a good baseline to train on to get to give models some some very basic abilities um and this is something i thought about by the way because it's like someone on one of the episodes i did said the internet is like it's like our children they inherit every bias we have and like they're just the results of the average of humanity when if you can you would ideally want these models to be like the best of humanity. So it's not going to come from there. Yeah. And it's also like there's no incentive. Like if you think about the current, you know, one example of like a data set that we would create is we would create an environment that has, you know, hundreds of lawyers that are like top lawyers that are working on verifiers to
21:50
create rewards for writing good legal memos, for example. There's never been an incentive on, you know, the internet for, like, hundreds of lawyers to go and, like, create these really complex data sets to be, like, available to the public on the internet. So it's not just a matter of, like, you know, the experts are not necessarily always on the internet, but it's also, like, they're not going to spend too much time creating artifacts for the internet, right? Because there's never been an incentive to do so. But now there is. There's a huge incentive. So because of that, There's tens of thousands of people, you know, just at MicroOne, and broadly, there's hundreds of thousands of people that are working in the space that are best in cost-centered domains that are heavily incentivized to create trained data for models.
22:40
So that's where we're going.