Hacker News Reader

Comment by ColinHayhurst | Hacker News Reader

This would appear to be the research and paper behind the MUM annuncement at Google I/O. https://blog.google/products/search/introducing-mum/

It's impressive research work. However, we can expect implemetation on Google will, intentionally or not, lead to more of keeping us on their property.

Google Search will become even less so a way of navigating and discovering on the open web. Is their dream, and was it always about AI? One Brain to Rule them All.

View on HN

I don't see anything impressive in this paper. They basically say what they want to do, no results.

They are not even original with the idea for this type of paper.

In September last year Aran Komatsuzaki posted: "Current Limitations of Language Models: What You Need is Retrieval" https://arxiv.org/abs/2009.06857

I have also a prediction market from August last year with this question:

Will a language model with at least 100B parameters trained to do external information retrieval exist before 2023?

Seems like people gave it 61% in March this year. I was at 75%.

https://www.metaculus.com/questions/4876/will-a-language-mod...

Good points. Thanks for the links.

>> I don't see anything impressive in this paper. They basically say what they want to do, no results.

These are often called "position papers". There's no reason why they shouldn't be written, and published.

Yes, indeed. My point is that there is nothing impressive about it. It is just a good idea.

The way I see it is that it doesn't have to be "impressive", as long as it's interesting and fosters debate. Science is a debate- not a circus with acrobats and trained horses. We can discuss problems and how to solve them without having to overhype everything all the time. In fact, for me that's the best way to do it. Even more so in machine learning where hype is over-abundant.

Reply on news.ycombinator.com