throwawayjava

Maybe the meanings of these words have morphed over time, but in what sense is a "machine-learning model" not a back-end concern? It sure as heck isn't front-end...

(E.g., the current wiki page for "front and back end" explicitly mentions "converting the symbolic phonetic representation into actual sounds" as an example of a back end concern, and that's certainly the type of task that some people use machine learning for.)

Honest question: does "back-end" now mean "the CRUD and API client portions of a web-based CRUD app?"

This might go toward a larger question that others are asking: what exactly is darklang and what does it do?

> Not by choice they aren't.

This is why I was glad to see this in TFA:

>>> ...while increasing the importance of sharing safety, policy, and standards research."

Even for Weak AI, history tells us that these factors -- policy and industry collaboration -- tend to be significant ingredients in successful safety reforms.

> Again, this breaks down the vast majority of the grandiosity and carves out one's true motivations... Everyone in the industry wants to convince their investors the prior approach can iterate to the later but they know in their heats it can't (Shhh! don't tell anyone). So, the question for an individual is how aware and honest are they with themselves and what is their true motivation. You can put on a show and fool lots of people but you ultimately know what games you're playing and what shortfalls will result.

The rest of my post is a response to this sentiment.

> As for research, the only thing that is required is a computer, power, an internet connection.

All that's necessary for world-shattering mathematics research is a pen and paper. But still, most of the best mathematicians work hard to surround themselves by other brilliant people. Which, in practice, means taking "cushy" positions in the labs/universities/companies where brilliant people tend to congregate.

Maybe most great mathematicians don't purely maximize for income. But then, I doubt OpenAI is paying as well as the hedge funds that would love to slurp up this talent! So people working on safe AI at places like OpenAI cannot be fairly criticized. They're comfortable but clearly value working on interesting problems and are motivated by something other than (or in addition to) pure greed/comfort.

> Profit seeking. Career building. Fame and prominence aren't sinister. Instead they are common human motivation. Common enough to easily group a significant portion of the Grandiosity centered around 'AI'.

So what? None of these motivations necessarily preclude doing good science. Some of those are even strong motivators for great science! The history of science contains a diverse pantheon of personality types. Not every great scientist/mathematician was a lone genius pure in heart. In fact, most were far more pedestrian personalities.

The "pious monk of science" mythology is actively harmful toward young scientists for two reasons.

First, the ethos tends to drive students away from practical problems. Sometimes that's ok, but it's just as often harmful (from a purely scientific perspective).

Second, this mythology has significant personal cost. More young scientists must realize that it is possible to make significant contributions toward human knowledge while making good money, building a strong reputation, and having a healthy personal life. Maybe then we'd have more people doing science for a lifetime instead of flaming out after 5-10 years.

> It means the thinking that weak AI is centered on could cause one to be locked out from perceiving that of AGI.

Thanks for the clarification!

First, you could say the same thing for all AI research at the moment! Grandiosity is perhaps even more common in subcommunities of AI that aren't safety focused.

Aside from grandiosity (either opportunistic or sincere), I don't think there's any sinister motivation.

More importantly, I don't think the safety push is misplaced. Even if the current round of progress on deep (reinforcement) learning stays sufficiently "weak", the safety question for resulting systems is still extremely important. Advanced driver assist/self-driving, advanced manufacturing automation, crime prediction for everything from law enforcement to auto insurance... these are all domains where 1) modern AI algorithms are likely to be deployed in the coming decade, and 2) where some notion of safety or value alignment is an extremely important functional requirement.

> ...and has, from what I can see, nothing to do w/ AGI nor are the approaches compatible

In terms of characterizing current AI safety research as AGI safety research? Well, there is a fundamental assumption that AGI will be born out of the current hot topics in AI research (ML and especially RL). IMO that's a bit over-optimistic. But I tend to be a pessimist.

> ...principal ideology...

As an aside, I'm not sure what this means.

Prioritizing safety results in a different vantage point on AI/ML/RL. Ensuring safety includes, as a sub-task, really understanding the mathematical foundations of new algorithms and techniques. In some sense, safety research is one way of motivating basic science on AI.

Managed well, a research program on safe AI is a "waste of resources" only in the same way that any basic science is a "waste of resources".

Developing a network of peers IS a good thing!

> Isn’t it intensely anti-meritocratic

I guess that depends on how yo define "merit".

The ability to build professional networks is actually a valuable skill, not just for the individual but also for the business. And the network itself is valuable.

I can think of three of reasons why a business would, on merit alone, prefer someone with demonstrated ability to build their professional network, or who is plugged into a particular professional network:

1. When it comes to hiring, "merit" does not mean "the best person"! "Merit" means "someone who can do the job and who we can convince to take the job given recruiting budget X and max comp package Y." If locating a candidate among the chafe costs more than "X", then that candidate ISN'T a meritorious candidate. No matter who good they would be at the job!

2. Businesses often recruit to access to an entire network/pipeline, not just a single engineer. This is a common pattern at lower-paying shops especially. Recruiting your Nth engineer from State U is easy because you already have recruit N-1 to give you the names of five people who would probably do well. Now, in any given year, the business could recruit zero people from State U. But then they have to rebuild their State U pipeline in subsequent years.

In other words, your network is part of the value you provide to the company! Someone with no professional network is strictly less valuable than someone with a professional network. And in some regional job markets, "Alumni of State U" is a valuable recruiting network.

3. Someone with a strong professional network has signaled that they have a modicum of skill at "winning hearts and minds". Depending on the shop, engineers who are good at building and maintaining reputation over the long term bring way more value than just the LoCs they commit during their period of employment.

4. It's also worth pointing out that someone you recruit via one of these networks might have a niche combination of skills that would be otherwise hard to discover. E.g., "oh, we're hiring a new dev to work with a new client in industry XYZ. I have an old friend who double majored in CS and ABC, which gives him the background needed to talk with folks in industry XYZ. Should I give him a call?". Again, the definition of "merit" is "we can find you with budget X and you're willing to do the work for comp package Y". NOT "you're the ideal candidate if we had X+Z recruiting budget to cast a wide net and find you"

All that said, hiring college buddies can be anti-meritocratic, for sure. Life isn't fair. Just one of the reasons colleges require freshman read Shakespeare :)

CCs offer 2 tracks.

Track 1: college prep. Students take a year or two of courses and then transfer to a 4 year institution.

For this track, the "trustworthy signal of competence" is by far the most important aspect of the product. So for CCs offering this track, your question is a bit ill-formed. You're basically asking "aside from the most important thing, what else is important?"

Track 2: Terminal associates degrees for students who will enter a trade immediately out of CC.

For this track, at least sometimes, MOOCs don't work as well because you need hands-on experience. Often, you also need connections to local institutions (for setting up volunteer hours and for job placement). IMO MOOCs shouldn't even try to focus on this track, at least for certain trades like nursing.

> Are the value gaps narrowing?

For track 2, it depends on the trade, but mostly no.

For track 1, MOOCs need to figure out the assessment problem and provide compelling evidence to higher ed that completing a MOOC is a strong signal of future success.

Finally, it's worth pointing out that MOOCs work well for some types of students (I wish they had been an option for me), but do not work at all for other types of students.

It's also worth pointing out that Pell Grants for CC tuition often constitute less public money than you were consuming during K-12.

Many affluent high schools have an abundance of AP/IB courses. Many high schools also have additional college-credit courses (where the high school acts as a sort of proxy for a local college, often a community college). At a "good" high school, students can combine all of these options to have not just a year, but even junior or senior standing by the time they start college.

There are three problems with this approach:

1. Not all high schools are affluent / have their crap together enough to offer these options.

2. students need to be prepared. Enough of them that you can fill a course. This can be a barrier in less affluent areas.

3. Students have to know about this option, usually starting in middle school so that they can begin working ahead early (e.g., finish algebra in middle school).

> ...and college

The pattern here seems to be picking the college vs. picking the more advanced courses at the college. So, you don't take "more advanced couses at regional State campus"; instead, you go to (your field's equivalent of) MIT/Stanford/Berkeley/CMU. Or at least the state flagship. C.f., algorithms at an elite university vs. algorithms at the branch campus of a middling state system. The differences in both depth (rigor) and bredth are, in many cases, quite extraordinary.

> No child left behind and other programs hinder education by applying one size fits all policies.

Do you have any data to justify this assertion? AP offerings have increased since NCLB was enacted.

I don't disagree with you that NCLB is a net harm, but I don't think it's accurate to place the blame there for this particular problem. (Also, whenever NCLB comes up, it's worth pointing out that states bear a lot of responsibility -- they have quite a bit of leeway in how NCLB is implemented.)

Not to mention there's a LOT of useful stuff on the internet that's not hosting on sites with Facebook trackers.

>>I recently received a "terms of service" update from Comcast, with the notification that they can now "monitor and record anything going through the network. Including, but not limited to: audio recording, video recording, ..."

I mean, look, if they're not going to do it, then why did they lobby so hard and successfully to do it?

Frankly, I have a had time imagining that this won't eventually happen. And sooner rather than later.

That's exactly the opposite of how corporations work. Especially corporations in monopoly/duopoly markets.

In fact, US ISPs lobbied heavily for the right to do exactly that, without user permission, and got their way.

> Thirdly ISPs in most countries are required by law to support the police with surveillance information. That is done in transparent way (you can read the laws and ask your politicians).

But that's not what this is about.

Today. But in terms of risk moving forward, I think ISPs are way worse. Two reasons:

1) I really can choose not to use Google/Facebook. There exist very solid, privacy-respecting alternatives for every service these two companies offer. It's a matter of consumer choice.

This is not the case for ISPs. If my (one) local "high-speed" ISP demands that I install a new root cert so they can MITM all my traffic, my choices are to a) capitulate, or b) find a way to live with very low-speed DSL/dial-up.

Monopoly power backed up by a vast network of cables has way more staying power than monopoly power backed up by social network effects (FB) or superior software offerings (Gmail).

2) You might argue that history demonstrates we don't have to worry about ISPs demanding to MITM customer traffic. But Historically, ISPs weren't incentivized to snoop because of regulatory barriers that prevented collection/use of data for advertising purposes. In the case of US ISPs, changes in regulator landscape suggest that past behavior doesn't guarantee future behavior.

> ...So (b) is slightly less trivial

I'd argue not. In both cases, a fully capable and prepared person has to jump through a couple of relatively trivial hoops. And when those hoops aren't possible to jump through it's a weird case. We can split that hair, but it's silly.

a) didn't bother applying but would've been shoe-ins, or else

b) knew very early (freshman year) they were research-bound and optimized for a non-industry objective function (but could've skated into an industry job of their choice given a shift in undergraduate career focus). E.g., couldn't pass a coding interview and no industry internships but have one or more top-tier publication in a hot subfield.

But (b) is kind of stupid to think about. It's like saying a successful lawyer would not make captain in the military. This may or may not be the case, but either way, who cares?

1. You really think Berkeley's top-ranked PhD program is recruiting people who couldn't find jobs? No. Not only can 100% of successful top-tier PhD applicants find jobs, 100% of them are strong candidates for the top echelon on entry-level jobs.

If you disagree, go look up the people in Berkeley's CS PhD program. Point out a single person you think didn't turn down mid-100s job offers to attend Berkeley.

Getting into one of these top-5-to-10 PhD programs is no small thing...

2. MOOCs aren't PhD programs and in nearly all cases aren't designed to feed into PhD programs.

3. Finally, at least in CS, at least for the moment, educators are in extremely high demand. And again, at least for CS, that demand isn't being manufactured by the academy.

That said, addressing this and other concerns isn't rocket science. The city could easily publish a detailed guide to filling potholes (what materials are ok, what materials are not, what process to use, etc).

(The signal-noise ratio on scholar-celebrities is so out of whack that simply ignoring them isn't a terrible heuristic, unless you're more interested in communication theory and marketing strategies than the subjects they're purportedly discussing.)

Heuristic: parents pass on their teenage music preferences to kids but don't pass on their own parent's teenage music preferences to their kids?

Many aspects of "scholarship" become inevitable when "tinkering over a long timeframe". You build up a knowledge base about the problem and/or methodology you're tinkering with, you eventually form a community around your tinkering, etc.

Lots and lots of tinkerers gratify themselves without contributing any modicum of knowledge to the world. But the most successful tinkerers are almost always effectively scholars in the sense of this post. In the sense that they're familiar with a body of knowledge/the people producing that knowledge. And they contribute genuinely new ways of doing things, or new ideas of things to do, or excellent executions on existing ideas that achieve the aspirations of previously articulated ideas. E.g., Linus is certainly a successful OS scholar in the sense of this post.

> In the case of very successful articles, you can fit all the readers in a (small) bus

This is off by at least an order of magnitude for any even remotely reasonable definition of "successful".

> Researcher B reads what researcher A reports and expresses an opinion, just like any honest reader could

First, the author explicitly kinda-sorta agrees with you here (...Third, it's not really the publication...), so it's pretty clear that publication venue != scholarship.

As an aside, peer review != validation!!! This is something that every young Ph.D. should explicitly learn early on.

Pittsburgh is a blueprint for the rest of the country in this regard. It sucked to be Pittsburgh in 1970-1990. Today, Pittsburgh is almost unrecognizable from the Steel City, at least in economic terms, and the county's unemployment rate is right around the national average.

> helping Pittsburgh, and almost no one else

The help to Pittsburgh would be pretty marginal; US Steel isn't even a top 5 employer in the region, and a lot of those jobs are suits that scale logarithmically with output. US Steel employs about the same number of people as Carnegie Mellon, which isn't even a particularly large university among R1 institutions.

The US is a great place to be if you are in prime working age with in-demand skills, no health problems, and no kids.

This gets complicated because of teacher's unions, but I think it's the best bet at getting high quality folks in CS classrooms.

Others will mention HS math teachers. Having spent enough time in US k12 classrooms to have an opinion on this topic, I think this can work out really well, but I'm not at all convinced it's a panacea. Mostly because many, many HS math teachers have learned how to teach specific math courses, but don't actually have a firm understanding of mathematical thinking in general. Training them to become CS teachers will fare about as well as teaching history teachers to become CS teachers (again, it can be great but lots of variance).

I get that the whole blog post is an advertisement, but infotainment systems should be physically isolated from control systems.

Furthermore, just using a microkernel doesn't actually provide any guarantees. It's a sound architecture choice, but that's all.

There are advanced control techniques that prevent replay, sensor spoofing, and other attacks specifically for control systems. But it doesn't look like this particular product implements those.

Correctly using a security-oriented microkernel only mitigates certain (really stupid) classes of attacks. These sorts of application domains deserve serious, in-house engineering effort aimed at very well understood risk/attacker models that are specific to the particular control problem.

Control systems need traditional pentesting/security types so that you're not broadcasting root credentials in plain text (...jesus). But a hardened system also needs input from controls/sensing people who understand security problems associated with particular application domains.

DHS does not use social media posts alone to place someone on watch lists, but they are incorporated into predictive assessments. Given two people who are otherwise equivalent, the one with public stated controvsial opinions IS more likely to end up on a no-fly list. By the govt's own admission.

A court that doesn't publish felony records is not a secret court. In fact, non-secret courts have been around for centuries even though easily accessible felony records are an extremely recent phenomenon (post-internet).