Hacker News Reader

Comment by version_five | Hacker News Reader

This is correct, but there's also some meta comment to be made that research has an "instragram problem". What he's talking about isn't really even AI so much as deep learning: there are lots of other branches that now get virtually no play relative to deep learning. And then there are all sorts of adjacent CS, image processing (when was the last time you saw someone talk about wavelets), language analysis, and other research avenues, probably often better suited to many problems, that get buried under the hype.

This is why academic research is so important. All the corporate money is flowing to deep learning, grant agencies should be sustaining the other more fundamental areas.

View on HN

Academic research has the same problem. When I studied neuroscience, flashy fMRI studies received substantially more funding than fundamental research. Without understanding how small neural nets work, it's difficult to construct bottom-up theories of the brain.

> flashy fMRI studies received substantially more funding than fundamental research

This is particularly infuriating since fMRI studies are not statistically robust and should be held in deep distrust.

The same happens everywhere in academia unfortunately. The current big thing is building “Apps” for your project even when no app is needed.

This is the bane of my existence. I work in a lab that uses as close to industry standard coding practices that we can do with our current skeleton crew numbers (unit tests, CI, version control, unit tests, code quality standards and norms). We lost half our senior student programmers in the last two years, replacing them with first years. You know what happened? Nothing. Our development continued as normal. I’m leaving in the next six months, and my contributions to our lab’s projects will continue to be maintained. I will receive no emails at 2 am if somebody finds a bug.

In the average academic lab, the GUI they just published had one maintainer, and it’s a grad student that’s been gone for a year already. Good luck getting an email about support, even though it doesn’t work out of the box and the code is unreadable.

> This is why academic research is so important.

The problem is academia just doesn't pay a livable salary these days. When I'm getting hit with thousands of dollars of medical bills on a regular basis, despite having insurance, having a corporate salary cushion helps for peace of mind. A lot.

The goal of academic funding is to help the military and also/later the economy.

To think otherwise is naive. The whole point of research is to have an advantage. Unless that money has no strings attached, you’re beholden to someone. Every historical scientist had a major donor whether it was some kingdom or nation.

Call me naive then. While I agree that a fair amount of academic research funding is defense/military related, it’s certainly not all of it. If said research is funded by organizations like DARPA, then yes, absolutely. There are other sources of funding out there though that isn’t defense backed: NSF, NIH funding comes to mind among others, including corporate backed academic research. When i was in academia, I worked on research funded by grants from NSF, Intel and Microsoft to name a few.

Im just saying that the end goal of your research is return on investment.

So many folks get no funding, even if their field of work is important to society at large.

Intel and whoever else funded you because they see some long term value in your work.

I wish research was done with “purer” intentions but hey.. I mean we are having this conversation based on a technology invented by the gov/military, Arpa. On my drive home I used GPS. And at home I’m using a microwave to make some popcorn lol.

>"What he's talking about isn't really even AI so much as deep learning: there are lots of other branches that now get virtually no play relative to deep learning."

Can you say what are some of non-Deep Learning branches in AI that aren't getting play relative Machine Learning/Deep Learning?

Andrew Ng in the linked message said himself: "Maybe you’re training XGBoost on a structured dataset and wondering if you’re missing out on ChatGPT. You may well be onto something even if XGBoost isn’t in the news."

Boosting generally is very competitive with (or better than) neural networks on small to midsized datasets.

Bayesian techniques are always interesting.

Personally I think there are great opportunities linking knowledge engineering approaches (which are very manual-labor heavy) with neural network approaches so we can bridge the taxonomy/unstructured data divide.

Expert Systems /s

Why does it matter? We can focus on othery approaches after deep learning stops yielding so much progress...

> grant agencies should be …

or enlightened corporations - ahm … never mind.

Reply on news.ycombinator.com