This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
The connection is substantial.
It's 2014. Imagine you're a large search company that gets virtually all of its revenue from search-derived advertising. Being able to rank by user preference drives verified revenue. But your current search ranking system is heavily heuristic. What can you do?
The novel thing: use deep learning to predict pairwise document preference (as measured by click through rates). You can imagine a system: you start with a query, decompose it into unigrams and bigrams, and then map it into an embedding space. No sophisticated contextual embedding schemes in those days; it's just a bag of words, averaged or summed together. Then, use a ReLu network on those embeddings to predict that pairwise preference, all trained jointly.
This requires a massive upfront investment. You need infrastructure to perform an unprecedentedly large, distributed training run. CPUs won't cut it, so you design custom ASICs to process your tensors. Serving the model at a rate of millions of queries per second with a latency of a couple ms is an entire problem unto itself.
This is incredibly costly. But, as it turns out, that crude (by today's standards) approach is enough to drive double digit percentage increases in revenue, easily paying for itself almost right away. Later you transition from bag of words to LSTMs, which helps even more, but you'd like even better contextual understanding.
And all that infrastructure investment is reusable. And many of the leads on the reranking project also happen to be working on neural machine translation in the same org, which faces similar problems. There must be some way to convert sequences of tokens into better representations; what do we need to do better?
A company that managed to do all that and answer that question would be incredibly well positioned for the LLM era, and it would have taken a staggering amount of organizational incompetence and sclerosis to miss the opportunity.
Thanks for this, I wasn't really aware of that history. Almost my entire interaction with neural nets was through the medium of computer vision object recognition and Caffe, plus Q networks when they came along; I attended lectures where people discussed how you could use RNNs to turn Shakespeare into Shakespearish but it never occurred to me how far it would scale or that it was really worth paying attention to.
More options
Context Copy link
More options
Context Copy link