As you may have heard Uber is cutting 3000 additional jobs, closing 45 offices, and reevaluating many of it's bets. One thing that stood out to the Artificial Intelligence community is the statement that Uber doesn't think AI research is core part of their business. "As part of the latest changes, Uber will scale back on noncore businesses. Mr. Khosrowshahi said the company is winding down its product incubator and artificial-intelligence lab, and exploring “strategic alternatives” for Uber Works, which pairs prospective employers with gig workers." Source: https://www.wsj.com/articles/uber-cuts-3-000-more-jobs-shuts-45-offices-in-coronavirus-crunch-11589814608?redirect=amp#click=https://t.co/J9owqvEdjV Do you think this is a precursor that we're living in an era of overhyped promises about what AI can deliver? Are AI science team more expendable than engineering teams?
In general I think AI is over hyped. In Uber case the company is survival mode and needs to drastically cut costs now. So anything that doesn't make money or expected to very soon will be cut or shutdown. I heard they need to cut 1 billion in costs quarterly. Not sure you can really draw conclusions from Uber under the circumstances.
AI labs at Uber is specifically research focused with no connection to product / the business. The company needs to become profitable so research roles are just nice to have at this point.
Use evidence to falsify hypotheses, not to confirm them.
It got bad because in the past 5 years every college grad is am “ML enthusiast” and that killed the potential. Now they will run and hide - at some point they were thinking it’s the best way to earn more . (I am in ML and I see how bad the talent is)
Detecting CoVid-19 with 99.9% accuracy using AI - check out my blog post on medium to learn how I used deep learning to publish the exact same content as the rest of the 50,000 ML enthusiasts out there!
My model can detect 100%!
By the very nature of the discipline, data science and AI teams provide the extra edge in most tech businesses while SWEs are the core engineers needed to keep the product running. So they are almost always more expendable.
They don’t provide the extra edge. They’re a risky bet on there being a possible edge to get
If you're talking about the kind of R&D teams that were cut, that's true. But that's not generally true of data science/ML
- I don't think Uber could rely on its collected data/talent to utilize AI for growth (beyond some core applications such as fraud detection and demand prediction). It's not as profitable as maximizing engagement time or ad revenue, for example. - Your suggestion that this article can be a proof of AI hype is somewhat cherry picked. - There is a wide misunderstanding of what can be achieved with data, and many marketers ride the "AI hype", which pisses off dedicated researchers who now have to compete with garbage research work flooding the field and creating just noise and damage. - It's easy to build *something* with AI (e.g. image classifier - just fork from GitHub!), but requires much more to solve a *specific* data dependent problem and deploy a probabilistic solution that can reliably cover unseen cases as well. Very different from engineering, heavy maths foundations are crucial. - Startups and companies that build effective ML solutions can solve an entirely new realm of problems at scale. There is a tangible and irreplaceable value from some AI applications. - What indeed stinks is companies insisting on using AI even when unnecessary. You see miserable graduates forcefully applying decision trees to business problems that aren't properly defined or can't scale, just to have their managers' satisfactory boasting to use AI.
If you think Uber doesn't have the right data/problem to utilize AI effectively, you know nothing about AI. The talent is up for your judgement but there isn't many people who can call Uber's head of AI Lab unfit. You have demonstrated exactly what is wrong with AI, i.e fresh-faced researchers who know nothing about building AI system at scale but thinking they are the shit.
To be fair, Uber does have (or had?) a stellar AI department and decent internal data tools. Not sure why op is shitting on it
Looks like they are also winding down their self-driving research unit. This would be a setback to the RL field. :(
That's short-sighted. Self driving is certainly not essential, which is why it's being cut. I think they are just going to piggy back on some other company's research in future. Uber just doesn't have the capital to continue investing in pie-in-the-sky ideas. It's not a setback for RL; plenty of big fish out there who will continue to invest in it. In retrospect, Uber was just a small fish which didn't self-evaluate correctly.
They spent millions on their research and published some really good stuff. I hope they can make a comeback and start over again, or at least their stance doesn't cause a butterfly effect with Waymo and Tesla cutting funds for autonomous driving department. Maybe the professors returning to CMU might become a blessing.
Case study: logistics team at Uber has about 40 DS. Logistics at GrubHub has 5 DS.
2024 Presidential Election
Yesterday
628
Weak old Biden
India
Yesterday
854
Modi's hate speeches.
Tech Industry
Yesterday
400
Are we currently in a recession, and is the U.S. not acknowledging it because of the impending elections?
Personal Finance
Yesterday
1096
Engagement ring conundrum
Tech Industry
Yesterday
792
Why is everyone here so obsessed about green card?
AI is done in most areas. Its equivalent to flying cars in 1960s
...?
Boeing: we'll make you believe some day. PS: if AI is still not there, there's a whole lot of money to be made in getting it there.