Akababa

It could be that this is the long-term efficient price after demand from early adopters goes down. Or maybe they underestimated demand altogether and it's too late to raise prices now.

Previous thread: https://news.ycombinator.com/item?id=21612488

Edit: Sorry, I meant prequel thread as they're related but not the same.

Was Doug Ford the one selling the drugs?

So what's the rationale for them pricing it at a point where it makes sense for people to line up around the block? Is it that the initial volume is low enough that the positive press and goodwill is more valuable than what they would gain from price discrimination based on time?

That's a fair assessment, hence the quotations. My way of looking at it was from the corporations' perspective, who collectively have to spend more money innovating while having their margins squeezed. Thus in the short term, the harder everyone competes, the smaller the net profit "pie".

You're right that in the long term it benefits consumers, and possibly the surviving companies after competition has died down and they can enjoy the fruits of their innovation.

Call me pedantic, but the usage of the term "zero-sum" in this article makes no sense. They use it synonymously with "winner-take-all" which is so far from it's original meaning that you have to question whether the author even knows what zero-sum means, or if they're just using the buzzword to attract clicks.

In fact, increased competition leads to reduced profits (at least in the short term), making this a "negative-sum" game.

Are there any negative externalities from this? If ad fraud is a self-contained cat-and-mouse game between Google and bots, I'm fine with watching from the sidelines.

I get the same results with a VPN, so regardless of how you slice it they win on relevance.

Reminds me more of the cult-like Scientology with their "thetans" and whatnot.

Is it possible that they've detected someone using a proxy and traced them back to the US?

That would be an error if you know the repeated evidence has correlation 1. The studies try to make this clear to the participants, but it's plausible that after a week they forgot where they read some of the repeated facts and treat them instead as independent information. I'd be interested to see how the interval of repetition affects the results.

Perhaps this can be rationalized from a Bayesian context. Unless your prior is that you're 100% sure of something (which no one is) and you don't completely ignore what other people say, your posterior will eventually shift to the opposite side.

Assuming it writes log files at a constant rate, the rate of wear is proportional to 1/(1-f), where f is the fraction of storage taken up by firmware. Interestingly, this isn't exponential either - it's infinite at f=1!

Firmware doesn't update as often though so it'd have to be static wear-leveling. I'm not sure of the safety of doing this with firmware files though - could it actually decrease the life expectancy by moving critical blocks more often?

In any case, if 90% of the space is filled with static firmware files then wear-leveling becomes difficult. Probably best to turn it off and let the physical blocks with the logs fail while keeping the firmware blocks intact, or just stop writing logs after a certain point.

https://en.wikipedia.org/wiki/Wear_leveling#Static_wear_leve...

I'd guess it's for people in the second camp (practical) who are just trying to satisfy their intellectual curiosity or want an intro to the math behind it all.

I agree with your assessment that's it doesn't have much practical use on its own, nor is it an efficient means to any particular end.

> In early 2018, the founders of Chinese artificial intelligence startup SenseTime Group Ltd. flew to Tokyo to see billionaire investor Masayoshi Son. As they entered the offices, Chief Executive Officer Xu Li was hoping to persuade the head of SoftBank Group to invest $200 million in his three-year-old startup.

> A third of the way into the presentation, Son interrupted to say he wanted to put in $1 billion. A few minutes later, Son suggested $2 billion. Turning to the roomful of SoftBank managers, Son said this was the kind of AI company he’d been looking for. “Why are you only telling me about them now?” he asked, according to one person in the room.

This sounds more like an episode of Dragon's Den than a firm responsible for billions of dollars of investors' money. Doesn't quite inspire confidence...

I so badly want this to be true - it's like a perfectly designed bug!

> If a few milliseconds can change the value of a financial instrument then we have to accept that waiting a second or two will allow a lot more traders to reevaluate and offer a fair price.

I'm not quite getting this point. If I have a sell order at price x and it's filled by a HFT 1 second before someone with a slower algorithm, how does that result in a more "fair" price for me? If the HFT instead posts a buy order at an unfair price x-1, there's nothing stopping the slower traders from taking my sell order at x one second later.

I don't believe liquidity necessarily has anything to do with efficiency, but more liquidity can never hurt. Even if the price of every stock right now is 100x what it should be, I still want lower spreads.

And to address your concerns about depth, I believe that's where market makers (which are related to HFT) come in.

Depends on which null hypothesis you're testing. He told us that he rarely had a down day, and based on Bitcoin's volatility I'd say it's reasonable to assume the probability of it going up or down on any given day is a coin toss (regardless of the longer-term trend and independent of other days). Thus the probability that he's a day-trading monkey is 0.5^n, where n is the number of up days he had in a row. (since he had a few down days it's actually a binomial coefficient but the value is < 1e-80 anyway, far less than a conventional p-value of 0.05)

On the other hand, if we wanted to test his 3900% yearly return, we might assume that monkey returns are equal in distribution to Bitcoin's price and then test the hypothesis that he's a monkey via something like a paired t-test. The problem here is that we only have one data point so p-value is undefined, and due to high variance it would probably take about n=10 points to get something significant. The upside of this approach is that you can get a confidence interval for how much better he is than a monkey, instead of just a yes/no answer.

In any case, since the author has at least 365 data points, he probably has an extremely good idea of both a) whether he's a monkey, and b) how much better he is than a monkey.

You have to make some basic assumptions to do any statistical inference, because if you don't then literally anything can be explained by luck. For example, even if the author did a follow-up post (which I'd love to see as well!) every year for 30 years and made money every time, it could still be "selection bias".

The number of monkeys required to match the author's results over a 12-month period is well over the number of atoms in the universe.

Through hypothesis testing you can estimate the probability that this was due to luck is very low. Assuming that a monkey would have a 50% chance of profiting on a day, the chance of going a month without a losing day is less than 1 in a billion.

Same. I don't think history will look kindly upon someone who "quit for losing" while other professional Go players keep playing.

Just imagine if Garry Kasparov quit after losing to Deep Blue, he would be ridiculed today by the chess community which is still going strong. Instead, he accepted defeat, moved on, and is regarded as one of the greatest chess players ever. I doubt the same will be said of Lee Sedol 20 years down the line if this is how he chooses to end his professional Go career.

That's a really oversimplified view. The truth is beyond a certain point no amount of "theorizing" is enough to overcome the sheer computational power of programs. AlphaZero just took advantage of recent advances in neural networks and GPU hardware to achieve a better tradeoff between hard rules and heuristics than Stockfish. Both of them are probably Pareto efficient in some sense, and barring some insane undiscovered loophole in the rules of chess it's unlikely that any human can come close to beating them.

I feel like posting this on HN is preaching to the choir. Anyone with an ounce of critical thinking skills is not going to take claims that "the world is ending" at face value, nor dismiss evidence of climate change.

On the other hand, there will always be incentives for politicians and journalists to sensationalize issues. It's just that HN isn't their target audience.

They re-use the same initialization, so it appears that the initial weights are inherently coupled with the nonzero structure.

I've always wondered this and can never find a satisfactory answer online. You'd think that if ReLU works then LASSO shouldn't present any problems either right?

In addition L2 tends to discourage sparsity by spreading out the influence of weights, which seems antithetical to the mission of pruning. (for example, if you run a ridge regression with two identical features the L2 penalty will assign equal coefficients to both instead of zeroing one out like L1 does)

That's intuitive but doesn't support the result of the lucky subnetwork (once found and re-initialized) training faster and outperforming the original.

That's nifty! I spent a while trying to figure out how it worked before realizing it's a Voronoi diagram. Simple yet beautiful.

On a 30-year time horizon it would be more reasonable to use the S&P as a baseline as you can increase your short-term risk appetite. Since Nov 1989 that's a 15.82x return (incl. dividends)