Full disclaimer, I am actually working in an enterprise AI startup. Joined a couple of years ago, I was super excited about building AutoML solution, that would automate all the boring stuff and discover amazing models. I had a background in ML and optimization, so that looked super attractive to me. We raised three rounds, since I joined. My stock options grant has increased 5x in value, but it's all paper money, of course. No IPO in sight. We made big engineering strides in developing our platform, especially on the side of infra, ability to process data and get model end-to-end. We got to the stage, where even non-technical people can upload their garbage dataset, get some sort of a model (if they can figure out on their own how to cast their business problem into something suitable on our platform), which can be deployed and served. Though I feel AutoML aspect just doesn't work. Maybe in some trivial cases it does, but we're still in perpetual consulting stage. If a customer has a little bit of data science knowledge, we have to manually tweak models and/or their data processing for them to get to the acceptable for them performance, because there would usually be something domain specific. Also, what happened quite a few times is that we educated a customer on data preparation, and then they were able to just use some open-source models to get good enough performance. On the other hand, it's very easy to shoot oneself in the foot, if you don't have any data science expertise. I'm tired of constantly getting emails from our CEO about taking a quick look at why a model is not working for some customer and seeing some garbage in the inputs. I'm not sure where the field is going. If I were a data scientist myself, I would not trust results coming from such a platform. It seems like our main bread and butter are customers that have no data science expertise/team, but they have data and they are curious about what can be extracted from it. Though, as they learn more, not sure how we can keep them long term. C3.ai stock is also depressing. Thinking, if it's time to move on. Also, seems like Amazon Sagemaker is pretty good, is there a point then? Thoughts? TC: $430k, where $200k is paper money.
Where do you work right now?
The purpose of AutoML is not to replace Scientist. It is to automate some repetitive laborious basic industry process. It is the “Auto” part matters, rather than the “ML” part. We expect the system automatically gives us some baseline results to avoid human labor cost and we don’t expect it to be highly accurate.
That's not how it's being sold in these companies.
Tech Industry
4d
45412
What happens when most of your team is Indian?
Tech Industry
15h
1214
Metamate- cringe max posts by this wannabe influencer
Ask Blinders
15h
670
Why no one cares about the lives lost in Gaza, Israel and busy in their own lives?
Tech Industry
Yesterday
841
Google unanimous strong hire
Health & Wellness
Yesterday
1336
Low Testosterone
I don't believe in AutoML and for the cases where it does work, AWS/GCP have you covered. But there's a lot of VC money in these startups