mks40

From reading this (and following this vaguely), I took the following assertions:

i) MM had virtually nothing to do with tripling of stock value, that was due to ownership in Alibaba/Yahoo Japan

ii) She did not turn Yahoo around as a business and made failed acquisitions.

Yet, the article wants me to believe

iii) Nobody could have done any better in this position. She achieved the best possible outcome

If her net contribution to Yahoo as a business was 0, it seems pretty unreasonable to assert that NOBODY could have made any better acquisitions (e.g. buy Instagram, not tumble), or strategic initiatives (why focus on search?).

Big thanks to Derek for his stackoverflow responses, have saved me so much time, especially considering how uninteresting that support work might be in general compared to designing and implementing new systems.

A question to Soylent customers:

Do you use it as the occasional convenient meal replacement or how far do you go into replacing all real food?

I suspect there is both, but what I am getting at is how much Soylent's long-term success depends on people seeing eating as a nuisance that should be optimized away versus something that should be savoured and enjoyed.

To me this is in the context of the larger question of personal utility maximisation. In the grand scheme of things, we have just started being able to really monitor and improve all aspects of our lives (in terms of time spent, convenience), and there is the question of how far we (most people/potential customers) ultimately want to go. It has become clear that there is the potential to optimise away friction/time spent in almost all human habits, but it is not yet clear if we really want to keep going down that route.

Will we keep optimizing things like meals just because we can until there are (conceivably) nutrient implants that make eating unnecessary, or will we sort of revert and see that maximising utility of every interaction does not lead to overall greater satisfaction?

In one world, Soylent could eventually dominate, in the other, it will remain a niche product because eating and food is too important too most, also culturally speaking.

Upon further consideration, I suspect this was a very intentional move and not just a growth effect, because they could just as well have kept moderation strict.

People do like to answer questions and be acknowledged for their know-how, but what people love is to talk about themselves and have their experiences validated.

Sure, Snapchat, Instagram, Facebook give a way to have your social existence acknowledged, but Quora offers anyone to have their individual life experiences validated, no matter if they have the lifestyle or looks typically associated with social media fame. That is a very powerful attractor but unfortunately brings out the result described above - users beginning to talk incessantly about themselves as a topic, the more one answers, the more one has the chance to convert to a topic oneself and have even more explicit opportunity to tell one's story.

Combine this with low traffic in topics of maybe more serious interest and Quora will suggest any popular content ('topics you might like). This is how one ends up having these stories in your feed without ever expressing interest in them.

You are not wrong. Of course, this is personal perspective, but in 2014-15 there was still a fairly academic tone to many questions, as in, you could ask actual science and math questions and get a knowledgable academic to answer them.

Today, if I open my feed, most of it is questions on personal experiences (from just now): 'What is the craziest thing you ever did when you were a teenager?', 'What surprised you most about attending graduate school in the US?, 'What is the most brutal death?'. I never specified interest in any of these topics.

What is worse to me is that I do not see any way to disable topics quickly so I have to perpetually mute high-impact posters who have attracted a large enough audience to be asked about their personal lives and seemingly enjoy answering the same things about themselves over and over. Like any web forum, the majority of replies comes from a relatively small amount of posters who keep retelling their personal story about their admission to MIT/their high IQ.

I suppose Quora is paying the price of growth and I realise my interests are not aligned with Quora's in attracting a large audience. It just means I am not personally interested in writing any content for it any more and I think many early users feel the same.

Anyone have the link to the original document? Could not find it (I understand the original version was pulled due to the index, but other versions).

Great, thanks!

This sounds great! Any particular reason only the MIT license is supported and not Apache 2.0? I am aware there are some differences with regard to patent rights, just genuinely curious, because there might be ambitious projects with the goal to incubate with Apache.

I have contributed to dl4j though ;)

I did not mean not criticise dl4j at all, I was simply pointing out an example of a feature I know I was missing at a point, I think we are actually agreeing. It does not always make sense to start something from scratch even though it's fun and a great learning experience. The ramp-up to something really useful in deep learning is simply very high. Further, few people can be an expert on the whole stack and I have no problem admitting to myself that even if I spent 2 years writing something from scratch, many parts would simply not be as good as something I could copy from an existing open source library. That's why contributing to open source also makes more sense to me - you get to work on a part that you can be good at.

Also should point out that when I was having problems with custom loss functions a year ago you guys were extremely helpful on Gitter in discussing issues.

I feel at this point this is not a sensible thing to do any more unfortunately. I totally get the impulse though. For instance, there is still nothing great on the JVM for deep learning with symbolic differentiation (deeplearning4j does not have this, correct me if this has changed).

On the other hand, I realize that between writing native interfaces, symbolic differentiation (e.g. writing a port of autograd), network optimisers, custom layers, parameter servers, multi-GPU scheduling and so forth, I'd spend years before getting to do what I wanted to implement in the first place.

Working on a deep reinforcement learning library that can be used in practical applications and not just simulations. The idea is that there might be many developers/ml enthusiasts interested in deep reinforcement learning, but existing research code is often tightly coupled with simulations like OpenAI Gym, somewhat brittle and requires a lot of know-how to adjust for a new problem. The goal is to have a library that allows to create and configure different deep RL agents with just a few lines, so they are easy to play around with.

Development is going slowly because there is a lot of research output that is difficult to integrate into one consistent architecture (also a weekend project), but working prototype with example usage is here:

https://github.com/reinforceio/tensorforce