Landing AI: Visual Inspection Platform (Andrew Ng) | Hacker News Reader

Landing AI: Visual Inspection Platform (Andrew Ng)

landing.ai

53 pointstosh19 comments

announcement tweet: https://twitter.com/AndrewYNg/status/1318978756912951296?s=2...

Looking at this from the outside, the "big ideas" here have nothing to do with recent advances in deep learning, but rather with implementing and offering standardized software systems and processes that enable companies which have limited AI expertise:

* to speed up manual labeling "by as much as 50%";

* to reduce model deployment time "by up to 67%";

* to scale to thousands of models without fuss;

* to improve accuracy via automated data generation and "smart" labeling;

* to track and manage data assets, training, and deployment globally;

* to detect changes in the environment and raise alerts when data drifts; and

* to retain control of models and data.

There's a big global market for this kind of solution.

Best of lucks to Andrew and his team!

Color me unimpressed. They are essentially "reimplementing" what Instrumental [1] - pioneer of and leader in the manufacturing visual inspection field - has already developed and has been successfully deploying globally. I know that Landing AI is saying that manufacturing is just the first of many potential applications of their platform. However, as of now, it is nothing more than vaporware and I have serious doubts that they plan and/or will be able to build a more general AI platform (as opposed to building multiple industry- or field-specific platforms).

[1] https://instrumental.com

I've been working in this area for many years and I've never heard of Instrumental, they seem to be focused on AOI in PCB manufacturing.

Apple, Amazon, Intel, etc. use Cognex Vision Pro for general manufacturing inspection needs [1] and they acquired a Swiss company called "ViDi" for deep learning capabilities a few years ago. Cognex has a market cap of $11 billion and they're by far the biggest vertically integrated industrial vision hardware and software company in the world.

[1] https://www.cognex.com/products/deep-learning/visionpro-deep...

Thank you for your comment. Re: Cognex - it's interesting and good to know. I've mentioned Instrumental, because based on my brief review, Landing AI's platform focuses on exactly the same sub-domain (even specimen photos looked similar to me), not general manufacturing inspection. I have looked up info on ViDi and, based on Crunchbase, the company was founded in 2012, three years earlier than Instrumental. Thus, I admit that the label "pioneer" in my original comment is incorrect and I'm respectfully taking that part back. Nevertheless, I still think that Instrumental is the leader (or, at least, one of the leaders) in their respective category.

P.S. I glanced at your website and was pleasantly surprised to learn that you have graduated from Georgia Tech - I have recently worked at GT for 4+ years (and was affiliated with both ME and MSE schools). It's a small world ... :-)

Andrew Ng’s machine learning courses were very well written and dictated. Best of luck to him and team

I personally not a big fan of black box statistical models, which are data mining with big data, but they serve their purpose reasonably well

Mods, this violates the ‘no editorializing’ rule for posts.

I think it's fair due to the generosity of knowledge he provides to the entire tech community. His quality teachings has given him a lot of goodwill for a lot of people. I won't complain.

You do know that the “generosity” is heavily VC funded? Don’t be misled by the .org in Coursera - it’s a for-profit company that has raised to the tune of 400+MM.

[0] https://www.crunchbase.com/organization/coursera

I don't like to be negative on partly what feels like small things but since this is a visual thing:

* Please stop using the default neon bright colours. Full on green on black hasn't been easy on the eyes since the late 90s. I know they're easy to code, they're not good to look at, and hard to focus on.

* Consider colourblindness. Pure red green comparisons?

* If you're going to go away from pure hex maxed colours, don't pick several shades of grey or pastel colours shaded nearly like the background

This seems like a bog standard object detection model wrapped into a nice frontend to teach it easily. I'm surprised it took this long. What am I not understanding here? When it comes to Andrew Ng, I expect nothing less than Skynet.

Isn't he a bit of a charlatan now?

He's one of the smartest people I've had a chance to learn from (I took a couple of his MOOCs over the years). You can immediately tell because he's able to explain complex concepts in a simple and easy to understand fashion, and make them intuitive, which to me is a distinguishing trait in people who understand, and don't merely parrot memorized jargon. So no, he's not a "charlatan" at all. He's likely one of the most informed people in the industry.

Shouldn't articulation mean a certain breadth of intellectual perspective vs depth? Just a thought, be feel like those discovering novelties on their own might struggle to articulate their intuition.

https://news.ycombinator.com/item?id=15313048

Meh. That's a sour grapes comment from someone who has achieved little to nothing themselves. All you need to do is watch a few of his lectures. He understands this stuff deeply, he's obviously extremely intelligent, and he, quite possibly, made stronger impact in the field than anybody else by just teaching AI better than anybody else to a much larger audience than anybody else. Likewise Hassabis' impact on the field is not to be underestimated. Deepmind is cranking out brilliant papers faster than I can read them at times. That would not be possible without Hassabis.

How is that a sour grapes comment? Is it wrong to point out the Baidu cheating scandal that happened under his watch?

Look - seems like you benefited from his M00C and you’re his fan boy/girl. Good for you but let’s not commit the fatal error of pandering/worshipping at the altar of authority.

Just because you have technical chops, doesn't mean you're magically imbued with the ability to conjure up demand for sci-fi-tier products.

Commercialization isn't about how smart you are or how elegant your product is. It's about finding someone else's problem and convincing them to pay you to solve it.

This isn't a jab at Andrew (Coursera shows he has a very good head for commercialization), and this platform looks like a perfectly acceptable solution to a particular customer demand.

It's just a reminder that selling things can sometimes look boring to techies, because customers usually have boring problems.

"This seems like {{useful but understood thing}} wrapped into a nice frontend"

Sometimes that's all you need to create a real business...

"you can already build such a system yourself quite trivially by getting an FTP account, mounting it locally with curlftpfs, and then using SVN or CVS on the mounted filesystem"

https://news.ycombinator.com/item?id=9224

:-)

Reply on news.ycombinator.com