mljoe

Based on conversations with people who have worked with Masayoshi, I don't think he does this for PR. It's a very genuine sentiment.

Ditto. It almost felt like Google got its nerd cred from being featured on Slashdot multiple times. BitCoin too (waaaay before it was "popular"), and probably Linux itself.

I can trace almost everything that helped me professionally and even many of my tech opinions to something I read on Slashdot. Really the site that influenced me the most.

Sure, there is no other way for computer science to go. AI always been the last and most important frontier. There may be more AI winters, there may be more AI summers. Ultimately people will continue working on this problem until it is solved or computer science goes away.

A student and faculty congress. There is precedent for this, many such bodies already have authority over millions of dollars.

I think it's worth trying. A 100% faculty based university, with student employees and contractors doing the more low level administrative work. The faculty would take on the high level admin work, e.g. the IT or CS departments managing computer systems in partnership with the students.

I don't understand how what you described is ADHD and not average human behavior. If people were not easily distracted by Twitter and Facebook, they wouldn't have a functioning business. There are entire business models that rely on the truism that a huge fraction of the population will procrastinate on a trivial task that will clearly and unambiguously benefit them. Mail in rebates for instance. When does this become ADHD?

The recording industry was very hardline with DRM, until one day Apple decided they won't support DRM anymore in the iTunes Store. Thus no music is purchased with DRM from any of the major e-tailers.

If the choice is no DRM or literally not making any money on the Internet, content creators will choose no DRM every time. Technologists have all the leverage here. If the web simply didn't support DRM, do you think Netflix, Hulu, and NBCUniversal would pack their bags and lose on billions of dollars because of some theoretical increases in piracy? Of course not. They'd just offer their content with no DRM.

There are many file sharing, illicit streaming and torrent websites that have the entire HBO catalog available. They are pretty easy to find with simple Google searches. Piracy is not esoteric, it's pervasive even today. There is absolutely no technical obstacle to watching any HBO or major studio's content online for free. And that's arguably a lot easier then figuring out how to copy a huge file for non-technical users. DRM can't have any major effect here, people who want to pirate will pirate, DRM be damned.

The only thing DRM does is introduce barriers to legitimate interoperability and innovation. You get situations like "does this operating system or device support Netflix?" which is unheard if the web was truly an open system.

I wouldn't say that. If it between having no DRM and having no web presence, I think broadcasters would pick no DRM. DRM really doesn't do very much to protect against piracy. It's just a way for C-level execs to show they are doing something to protect their IP. It's a kind of security theater.

These libraries leave most of the actual day to day work for ETL. ETL happens to be highly data and problem dependent, so it can't be easily automated or reused. For this reason I think the best thing to be a good applied ML person is a solid programming background. You should have a working knowledge of statistics and linear algebra, but the most useful skill really is being able to write good code. It's different for research of course.

IgG is usually implicated in fighting off pathogens like viruses and bacteria. It doesn't typically cause an allergic response. IgE is usually implicated in fighting off pathogens like parasitic worms or mosquito bites. It's the antibody that regularly get confused and binds to peanut proteins and pollen. There is speculation on why the IgE mediated immune response gets confused. One hypothesis is IgE evolved much more recently and "needs more work".

The immune response to IgE binding is intense. Allergic reaction is an IgE response to a harmless antigen. However, IgE can be downregulated by repeated exposure to the antigen in question. If the body uses IgE too much (eg. from repeated exposure), it starts to replace IgE antibodies with the more subtle IgG antibodies.

The immune system is very fascinating really. IgE might be the key to effective cancer immunotherapy. Imagine if we could program your immune system to fight a cancer as intensely as it fights a pollen grain or peanut in an allergic individual. There is a lot of research in this area these days.

It's hard for people to get their head around this but you are exactly right. A company has value tied into their market value and accumulated assets even if they pledge to never pay dividends. Berkshire Hathaway is a notable example here. Stock trading is a way to make this value liquid before any actual dissolution event.

There has never been anything quite like the patent situation around HEVC. Using HEVC legally requires signing four different license agreements at the minumum[1]. The license agreements with Technicolor and Velos Media are actually secret. This includes their fees. The other two licenses (MPEG LA and HEVC Advance) are public. They are far more expensive then H.264 taken individually, even at current promotional rates.

As a result the Alliance for Open Media (AOM) was formed by major hardware and software companies. AOM is developing an alternative to HEVC. AOM has almost all hardware manufacturers of any significance as members.

[1]: Other essential patent holders who have not officially licensed their patents include AT&T, Microsoft, Motorola, Nokia, and Cisco. They may elect to license their patents unilaterally, which adds additional uncertainty to HEVC.

Fundamentally this is related to the induction fallacy. The doge meme image actually illustrates it pretty well.

https://en.wikipedia.org/wiki/Problem_of_induction

Another closely related thing is the No Free Lunch Theorem.

https://en.wikipedia.org/wiki/No_free_lunch_in_search_and_op...

These are concepts I believe are very important to internalize if you work with machine learning. Fundamentally we are making predictions (ie. guessing) on the nature of entirely unknown information. So there is a certain inherent impossibility to the task in the general sense. It shouldn't always work.

I feel like CA has been for awhile an "almost acceptable" place to incorporate for startups. Maybe this will finally push it over the edge. Unseating Delaware would be good in the long term. They have taken advantage of their position in a number of ways. Like charging relatively unique corporate taxes (eg. a tax assessed on # of shares).

To be fair, China is a AI powerhouse in its own right. China is investing a lot of its money domestically on this problem as well. Lots of legitimate papers come from purely Chinese teams. ResNet was developed by a purely Chinese team. A lot of very important open source machine learning software out there is written almost entirely by Chinese nationals.

It's important to note this because shutting down Silicon Valley/US access to Chinese VCs and companies won't necessarily just harm them. It would harm us as well. It does need to be looked at though, just carefully.

If an HMO got stupid amounts of marketshare it would probably fix the system naturally. It would also improve stuff like allowing HMOs to be able to better take advantage of medical data and provide better care through understanding the patient. The problem is there is too many HMOs, which makes HMOs in general less convenient.

Does this often happen? Saving money on lunch might delay a company's failure for a week at best. My perspective is most startups will maintain all benefits until the bitter end explicitly to avoid giving their employees this kind of impression. Hints like this might cause critical employees to jump ship during a financial crunch time.

>And if you work in Congress you often get an accelerated schedule so one year there counts for several years toward your pension.

That's true, but the only real people who benefit is the Member and maybe their CoS. Congress staffers make joke salaries in general. The reason people take those jobs is either a sense of duty or as the requisite training program to become a lobbyist.

In DC/NOVA area $60k a year qualifies you for rent subsidy and other welfare benefits. That said $60k is not what federal employees in the DC area start at, nor is $140k what they normally end at. It's more around $80k - $155k ($180k-$190k if you hit management).

The pension has been slowly getting worse and worse though. With the newest version of FERS we are at 1% per year at service annuity partially financed with a 4.8% employee paycheck deduction.

They won't necessarily look for pipe bending jobs either, they'll try to get any job they can get. So entirely unrelated and non-automated fields can become devalued due to the increased supply of labor.

It's very black and white right now (no knowledge or very constrained knowledge). The same problem exists with just plain old parameter learning, with DL models having so many free parameters it can make training more computationally expensive then it needs to be. For instance, I want to train a reinforcement learner to do some task in the real world (eg. a robot). It would be nice to be able to define a prior that puts very low to no probability on actions that are not physically possible. This would constrain the search space considerably. But it's not really clear how to do this with neural nets.

Often these metalearning systems have to pick from a library of already understood building blocks to build novel structures from. This limits the flexibility of their hyperparameter space. It's a bit like having a robot building a building but give it premade nails and wooden planks. It will never build a skyscraper. We have to figure out a way to also allow the system to come up with its own building materials so to speak.

Yes, I know it seems silly now, but this was a very real scare that CS would be a worthless degree in the future. I was dissuaded from majoring in CS during the height of this, but ultimately changed my major back to CS when I realized that I was actually really good at it.

Right $100 billion dumped all at once into VC would basically destroy the whole ecosystem. The saying is VC can't work when there is too much gasoline (capital) and not enough fire (talent/innovation). Despite the headline successes, most VC firms fail.

It seems super sketchy. I mean the valuations based on current P/E aren't even terrible except for the fact that gaming especially doesn't seem an enduring business. That E is probably going to go down as people move on to other games.

It's illegal, at least in the United States. Why? Because venture capital is considered too risky to be marketed as a public security. There are VC ETFs sold on the Hong Kong market.

That makes sense and all, but the definition of accredited investor is tied to net worth and not knowledge. Which means that maybe you can't scam poor people, but there are plenty of gullible rich people with net worth over the $2 million in liquid assets or whatever it is these days.

Maybe treat being an accredited investor like a driver license and not tie it to your net worth?

I agree with you 100%, but MOOCs aren't necessarily the only option for educational innovation. The idea of Minerva (students still live together, but the university moves around the world) sounds interesting to me, even though it is not at all less expensive. These sort of things were never technically possible until well after the university system became entrenched. Another option is internships or entrepreneurship, what the Thiel Fellowship encourages. Chris Olah is a employee of special note at Google Brain. He never finished college, and he had the smart idea to basically reinvent scientific publishing.