Hacker News Reader

Comment by lvncelot | Hacker News Reader

Some cynic remarks that during the first AI golden years, claims of imminent success seemed to come from a place of hopeful naïveté of a fledgling science, whereas those same claims nowadays seem to come from a place of cold calculation of a booming business.

View on HN

Is Deep Mind a “booming business”? They are achieving great things academically, but their business successes are either kept secret or mostly absent. All I know about is the Google data centre cooling scheduling, probably a big saving for Google but hardly an achievement that in its own professes their business success.

Deep Mind is cutting edge ML in general, right? Doesn't Google actively apply the lessons learned all over the place? YouTube content recommendation stands out to me in particular. Translation and automated closed captioning are also obviously ML based. I'd guess that most of the really interesting stuff would be behind the scenes and not immediately visible to end users though.

Yes, its hard to tell the exact algorithmic underpinnings of production models that Google uses but you have to assume that although they have some done some impressive strides in fields that isn't immediately profitable (AlphaGo, AlphaFold...) they also continuously push new research in things that are obviously of interest for Google and Alphabet- especially in text-to-speech, speech-to-text, information-retrieval etc.

For reference : https://deepmind.com/research

> YouTube content recommendation stands out to me in particular.

If that's "cutting edge ML", then going off my YouTube recommendations, we're back in another AI winter. If I watch one video from a channel I've not seen before, I'll get that channel recommended constantly even if it bears no resemblance to what I normally watch. On my Explore page, the first 22 videos (of which 8 are Fortnite-related!) hold no interest for me. My Home page is just channels I've watched repeatedly and/or am subscribed to. It's a mess.

How often do you use YouTube? Personally I am a very heavy user and in my experience the obsession with a new video kind you watch only lasts for a few recommendations unless you lean into it.

I would guess about two thirds of the channels I consistently watch I originally discovered through algorithm recommendations. I think it works extremely well.

I think you raise an important point. The youtube algorithm is pretty bad if you don't use youtube very much or only use to consume very popular content. Youtube's recommendations used to be terrible for me, too, but sometime last year I crossed a threshold and since then it has been recommending a lot of small, highly specific channels that nevertheless are great fits. My wife's recommendations are still utter garbage though.

> How often do you use YouTube?

Every day, averaging 2-3 hours. It's background for working and foreground for evening viewing.

That's because you fit into YT's conception of how viewers behave. For people who don't fit into "normal"-ish behaviour it has little utility.

For me, probably 90% of what I watch I'm not interested in and often I'm repelled by. This is because I mostly watch to find out what things I'm not familiar with are.

For example let's say I'm a liberal. I'm not going to watch liberal political videos because I know generally what they're going to say and I don't need my political views stroked in order to be happy. But I will watch various other political videos, no matter how extreme or not, so I can be at least a little familiar with their behaviour and views.

YT can't cope with this. To their systems I seem to be randomly picking videos with no correlation with the subject matter or other users and no reinforcing pattern. It just gives up and recommends things based on the behaviour of the general population, as if they had no data on me at all.

Is the Explore page controlled by videos I watched ? Because there isn't a single video on it i would watch. Not one.

I’m stressing the business part. YouTube is a loss-making business year after year. Deep Mind gloss doesn’t seem to change that.

If indeed it even is Deep Mind making those improvements, Google has lots of other ML groups, such as Google Brain, and these are more directly focused on Google products.

There’s no denying their academic success, or game playing etc, but as far as I can see, the data centre cooling bit is the only palpable (public) business success.

How do you know that YouTube is a loss making business?

YouTube made $6B revenue in Q1. [1] While they don’t release profit numbers, it would be pretty surprising if they were negative.

Did you mean to write DeepMind instead? If so, I don’t disagree.

[1] https://www.cnbc.com/2021/04/27/youtube-could-soon-equal-net...

Deep mind’s protein folding algorithm is probably worth a chunk of change. As far as I know, they’ve been holding onto the secret sauce rather than publishing it.

They have an applied division which applies ML to Google products. I suspect they are very valuable in $ terms just for the work listed here: https://deepmind.com/impact. Google's entire business from the start was doing research and bringing it to the masses, so this shouldn't really surprise anyone.

No not really. Deep Mind is almost all cutting edge agent-oriented reinforcement learning, hence the nature of the claim they're making. The impact on Google's business from AI has come almost exclusively from other kinds of ML, or that's at least how it appears from the outside. E.g. replacing Google Translate with neural translation doesn't seem to involve RL and certainly doesn't involve agents playing video games.

Deep Mind is best understood as the following bet: if we can train an AI that can learn from "its environment" and do the sort of things a human would do in that situation, then we have achieved AGI and from that ... business ... will follow. Hence their focus on video games as a training environment.

This sounds intuitive but is actually a very agent-centric viewpoint and most AI doesn't resemble this type of thing at all. Most AI deployed so far doesn't have anything resembling an environment, doesn't have any kind of nexus of agency and doesn't need to actively make decisions that then feed back to its own learning, only make probabilistic predictions. And in fact you often don't want an ML model to train on the outcomes of its own decisions.

Reply on news.ycombinator.com