Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AlphaGo's next move

deepmind.com

Post reply on HN
Print view View original

161–170 of 179 posts

Re: AlphaGo's next move

#161

Post by visarga »

espadrine wrote: post #82
gwern wrote: post #77

Earlier quoted context omitted.

> At this point, it is hard to complain. Actually, it's very easy to complain. If they released the model, people could generate arbitrarily many self-play games instead of depending on DM to release 50, could create arbitrarily many tools using the model instead of depending on DM to create and maintain a single tool, and could verify the results of training a clone based on even sketchy descriptions of the methods…

> people could generate arbitrarily many self-play games I have doubts. Their TPU design may be a large factor into making matches at this level within the time limits. And at this point, some implementation details might hook into Google-specific libraries that require the ability to spawn processes in thousands of servers, which past blog posts[0] have hinted at. [0]: https://deepmind.com/blog/decoupled-neural-netw…

They could just release the trained nets and let us re-scale the code. Even without a large MCTS it is still powerful.

Re: AlphaGo's next move

#162

Post by visarga »

tim333 wrote: post #105
temp_12345 wrote: post #75

So, a question bordering on the philosophical : What can be done to prepare for the end of human supremacy, and quite likely human civilization? For instance, as a software developer it feels almost pointless to continue improving at my craft if AI systems will surpass me within 2-4 years (even if the pessimists are right about it taking 5-10 years, that's still an awfully small timeframe). Likewise, it feels a littl…

Well, firstly AI outdoing us in software development is going to take a while, probably >10 years. Secondly you're looking mostly at the negatives but not positives of AI advancing. Some of those: We will likely use AI to enhance ourselves rather than just have it take over. Such merging may lead to the end of death. At the moment sure you can develop away then age and die - the AI thing may be jollier. Robots at som…

Also, we'll get to use amazing AI tools in our projects. That should create a lot work opportunity, even for non-AI experts.

Re: AlphaGo's next move

#163

Post by amelius »

polskibus wrote: post #127
cocktailpeanuts wrote: post #88

There was a time not too far back when people used to be considered a "genius" for their ability to memorize things well. Nowadays nobody thinks of them as geniuses. Also, people used to be considered geniuses for knowing a lot of things. Nowadays information is just a Google search away, so knowing a lot doesn't really mean as much as it used to. What matters more nowadays is your ability to learn synthesize the thi…

Google made fact recollection less important than before. However, Google itself prefers people who do well on whiteboard interview, which could be solved perfectly by some Google searches! Memorization of historical facts is one thing, but training your memory for fast retrieval of facts and associations is still highly valued. Therefore it is still important to train your memory to the extreme. You'll do something…

> Google itself prefers people who do well on whiteboard interview, which could be solved perfectly by some Google searches!

I don't see the problem if these problems can also be solved without relying exclusively on knowledge.

As an outsider, I'd say that Google's hiring strategy is pretty successful.

Re: AlphaGo's next move

#164

Post by mark_l_watson »

oska wrote: post #149
mark_l_watson wrote: post #45

The 10 AlphaGo vs. AlphaGo games are a nice gift! I have always liked playing through great games, both Chess (using the book The Golden Dozen) and Go (modern games and the ancient Shogun Castle games). I have some history with computer Go. In the late 1970s I wrote a Go playing program in UCSD Pascal that I sold for the Apple II, and also for a lot more money I sold the source code to a few people who wanted to expe…

Do you know if there's video or audio of Peter's talk online? I just did a bit of a search but couldn't find it.

https://vimeo.com/215418110

Enjoy!

Re: AlphaGo's next move

#165

Post by nathan_f77 »

doktrin wrote: post #107
nightski wrote: post #103

Earlier quoted context omitted.

I'd love to be proven wrong but that hypothetical general AI is at least decades away in my humble opinion. It's not going to happen without many more students and researchers in the machine learning/AI field. Now would be a great time to join and make contributions.

> I'd love to be proven wrong but that hypothetical general AI is at least decades away in my humble opinion. I hope so. General AI is the source of dreams and nightmares, and I think it will take us decades just to adequately prepare for it from a safety & management perspective. > Now would be a great time to join and make contributions. Sincere question : short of embarking on a PhD, is AI research something the a…

I've skimmed some papers, and my impression was that you would need to have at least a bachelors degree in mathematics. And by "skimmed some papers", I mean they may as well have been written in a foreign language.

This also helped me realize that I'm not super excited about getting into machine learning research, because it's just way over my head. I can play around with TensorFlow, and I enjoy writing all of the code connects to a black box, but machine learning seems like an entirely different field to software engineering.

Re: AlphaGo's next move

#166

Post by cocktailpeanuts »

yladiz wrote: post #157
cocktailpeanuts wrote: post #151

Earlier quoted context omitted.

"First world problems"

It is not a "first world problem" to want to search for accurate and unbiased information.

I guess you didn't get the metaphor "first world problem".

Go back a couple of decades and you'll find central authorities and organizations with agenda to manipulate the public could easily do so because the rest 99.99% didn't have access to information.

People who wanted truth had absolutely 0 way of attaining the truth back then. Now you're complaining about how it's "hard" to search for accurate and unbiased information. It's not even THAT hard, you just need to google a bit more, sure it's not easy for everyone, but at least there's a way.

Re: AlphaGo's next move

#167

Post by YeGoblynQueenne »

jacobolus wrote: post #129
adharmad wrote: post #116

Earlier quoted context omitted.

The ability to use knowledge does not come out of thin air. It is unlikely to find someone who knows how to program C++ extremely well, or solve problems in Quantum Field theory, but does not remember most of the language constructs or mathematical equations. Continuous practice involves putting in the hours at practicing an art or a science, which by itself builds muscle memory about the language syntax/equations et…

“The best geologist is he who has seen the most rocks.” – H. H. Read

Well, in that case:

"All science is either physics or stamp collecting" -Ernest Rutherford.

Re: AlphaGo's next move

#168

Post by YeGoblynQueenne »

cocktailpeanuts wrote: post #88

There was a time not too far back when people used to be considered a "genius" for their ability to memorize things well. Nowadays nobody thinks of them as geniuses. Also, people used to be considered geniuses for knowing a lot of things. Nowadays information is just a Google search away, so knowing a lot doesn't really mean as much as it used to. What matters more nowadays is your ability to learn synthesize the thi…

>> Basically the "memory" part of human brains have become commoditized without us even realizing.

So, explain this to me like I'm 5. If Google search means you don't need to remember any knowledge any more, why is it that possession of an English-Greek dictionary does not render one capable of speaking both languages fluently?

You can look up all the words you want. Assume you have a grammar of each language at hand, also. Do you think you would be able to speak fulently or understand a fluent speaker of a language you don't know?

Re: AlphaGo's next move

#169

Post by YeGoblynQueenne »

cocktailpeanuts wrote: post #130
skybrian wrote: post #108

Earlier quoted context omitted.

Memory still matters when it matters. For example, when learning a new language. In general, good real-time performance is going to require practice and memorization. It's the trivial stuff you rarely need that's less valued. There are some other things that I think are less valued these days: - Informed speculation counts for less when you could do a search instead. Maybe a good thing? - Cleverness counts less when…

> For example, when learning a new language That's a good point, because I think this "ability to speak multiple languages" will become commoditized too through technology. You already see pieces of technology that enable you to communicate in realtime (although clunky and not accessible enough at the moment) I agree that "realtime" aspect would be the last wall that will stand to distinguish humans from machines, th…

See my post above. We have very good translation dictionaries, that list all the meanings of words in two languages you might ever need. Why isn't that multilingual ability being commoditized? And why is it still not possible for most people to speak many languages as well as their native one?

Re: AlphaGo's next move

#170

Post by polskibus »

amelius wrote: post #163
polskibus wrote: post #127

Earlier quoted context omitted.

Google made fact recollection less important than before. However, Google itself prefers people who do well on whiteboard interview, which could be solved perfectly by some Google searches! Memorization of historical facts is one thing, but training your memory for fast retrieval of facts and associations is still highly valued. Therefore it is still important to train your memory to the extreme. You'll do something…

> Google itself prefers people who do well on whiteboard interview, which could be solved perfectly by some Google searches! I don't see the problem if these problems can also be solved without relying exclusively on knowledge. As an outsider, I'd say that Google's hiring strategy is pretty successful.

Good luck passing the interview without thorough CS education, of which a lot is achieved through memorization and memory training. Your interviewer would never be patient enough to wait for you to figure things out in a couple of months.
Post reply on HN