Writing crud websites with spring/.net ms isn't really that difficult. The work is largely done, you just have to fill in a few small details (your domain models, your db credentials, your endpoint paths, your validation rules, your css, your logging configuration and a few templates html pages) and you are done. The sites have very low load, they have almost 0 concurrent writes for the same db entity, and the front-end is largely a simple form.
Operations are handled by a different team than the initial development team - they don't even need to know how to deploy the thing. It's code by numbers, and in that situation, it pays to have someone that doesn't deviate from the number system and the low need for creativity drives the value of the work down and makes the code into a commodity product. Value is low, competition is high, margins are low. Certification is just a sign that you are open for business in that market.
I suspect that a large fraction of the software in use is written at least in part using that method. Now you can debate whether that software actually works or not, but I think that its value is not net negative.
A little bit of both. The body shop outsources the risk of putting butts in chairs, and removing them when you’re done.
It’s crap work for crap pay. In the 70s and 80s when this stuff was new, it made sense to “grow” and train operators and more blue collar folks driven by runbooks and tight frameworks. Nowadays you just hire low skill folks “off the shelf”.
> Side question - won't it be more profitable to sell some and keep some for potential future profits, so the overall risk is lower but you still profit potentially?
Selling all your stock comp immediately is more about diversification of risk.
If rsus are a nontrivial aspect of your comp, then you're already extremely exposed to risk in your company's stock price stock price (if it tanks take a big pay cut and they might even lose their job).
> if it was machine generated, did it take creativity?... but did the output that is being copyrighted?
In the US, this question has been settled since at least the 1990s (e.g., in the context of videogames). The output of algorithms is, in general, copyrightable, although there are some rather common-sense exceptions.
The question isn't whether you can copyright the output of an algorithm. The more salient question, in my mind, is whether the output of ML algorithms belongs to the owner's algorithm or to the owner of the training set.
> It's just that the owner of the training set -- not the owner of the algorithm -- is the one with the valid claim to copyright.
Not all that different from a pop song made by editing together licensed samples, no?
In that case, the song is certainly a derivative work of the samples, and so the producer of the song needs to get derivative-works-allowed licensing from the samples’ authors (which is what you must necessarily get when buying samples from a sample library, for them to be of any use at all.) The produced song is then its own work with its own copyright. Sometimes, larger samples (like reused vocal performances) require payment in, essentially, “equity”—a percentage of the song’s royalties are transferred as royalties to the sample. But in most cases, the sample is purchased for a flat fee, and there is no ongoing relationship between the revenue of the song and the revenue of the sample.
Is anything different if you replace “song” with “news article” and “samples” with “training set”?
Copyright isn't a natural right, and giving rights to computational algorithms isn't a normal legal act - how does it benefit society to do that? Is the deal good for the populous as a whole?
We can have the output for the cost of the energy, or we can perpetually (AIs never die!) pay tax to a wealthy capitalist and have the same output; why is the latter better?
One possible legal theory: because the algorithm was trained on a text corpus upon which the algorithm's owner has no legal claim.
In this particular case, I don't think that theory would hold much water.
However, consider, e.g., a model that produces encyclopedia entries and is trained on a half dozen existing encyclopedias. IMO, if that model is using techniques similar to SoTA and isn't producing utter garbage, then the owner of that model should have a very difficult time claiming that the output of their model is anything more than a sophisticated round-about way of copy/pasting from existing encyclopedias.
But still, in that case, the output is still covered by copyright. It's just that the owner of the training set -- not the owner of the algorithm -- is the one with the valid claim to copyright.
>> One possible legal theory: because the algorithm was trained on a text corpus upon which the algorithm's owner has no legal claim.
The same can be said about human writers: they learn to write based on thousands of "training examples" - the articles and books they read thorough their life.
Or rather, Who knows? Maybe. But certainly, at least today, a SoTA model generating a quality encyclopedia certainly is not doing what human writers do, and is certainly effectively copy/pasting.
Maybe in 50 years -- or 10 years with a major breakthrough on the level of general relativity -- that statement might be true. but it's certainly not true of today's deep NLP systems.
A better example is the "copy and paste" news articles that saturate feeds everyday.
The exact same set of facts, that were obviously reported originally by a single individual, then rearranged, reworded, and republished by 100's of "reporters"/"bloggers", (sometimes) with an attribute of origin.
That would be a problem, but would be a data licensing issue, which is distinct. It's more analogous to "Blurred Lines" infringing on "Got To Give It Up" or w/e.
It's helpful to remember that the Bay Area is extremely expensive, and even more helpful to remember that some people have large families.
I could pretty comfortably raise a family of 6 on $100K in a midwestern city, or a family of 6 on $80K in the rural midwest/south. Without sufficient retirement savings, mind you, and my safety net would be non-exisent. I'd need at least another $30K-$50K on those numbers to build a strong safety net (remember, it's a safety net for 2 parents + 4 kids, not for one person...).
Doing the same in SFBA would probably require at least $250K. Maybe more.
> Surely you don't believe less than 1% of people in the US can live comfortably and have a safety net, do you?
The number is larger than 1%, but probably still smaller than you think.
The average American definitely doesn't have a sufficient safety net. IME, the average American raising more than 1 or 2 kids almost certainly doesn't have a sufficient safety net.
So, no, you don't need to be in the top 1% of earners in the US to have a safety net. But if you have a large family, you probably have to be in the top 10% to live comfortably and have a safety net and retirement.
> Doing the same in SFBA would probably require at least $250K. Maybe more.
Why? Each child might consume, generously, an extra 5k of food and 2.5k of clothes and other inputs per year. This is an increment of $30k on top of housing. For housing, you could add one or at most two bedrooms for four children. This does substantially raise housing costs, but there are still places in Sunnyvale for $1M or just above that that have three bedrooms. They may not be the nicest houses, but they would do.
So, before you have kids, save your downpayment. On a 250k salary, this would take one to three years if you're thoughtful about it. Then buy the house and start popping out babies. Times will be tight if you need to pay for childcare, but that's a defined and short period of the childrens' lives. You're not going to be taking lavish vacations, but if you wanted those, you wouldn't have four kids anyway.
Or, be involved with your school board, and raise up the schools you can afford to send your kids to?
Ego seems to be at play here, a bit. No one wants to admit that they're in the same boat as the people sending their kids to "those schools," but they are. The numbers don't lie. Looking on the bright side: admissions look at class rank. An outstanding student at a mediocre school has better odds than a mediocre student at a "good" school.
There are plenty of houses listed near $1M on Zillow that have acceptable school districts. They aren't the best in the Bay, but I'm not a buyer on the premise that you need a 10/10 school for your child to have a good opportunity at success and happiness.
Yes, mathematicians mean something different and specific when using the word bias. The average non-expert is not misusing the word. They are using the word to express a different -- and far more popular -- meaning.
Neither is wrong, but insisting that a naming clash carries any substantive significance on an underlying issue is just silly. Similarly, insisting that nonmathematicians should stop using a certain word unless they use it how mathematicians use it is a tad ridiculous.
Of anything, it's more reasonable for mathematicians to change their language. After all, their intended meaning is far less commonly understood.
> IMO it attracts people that can't come with SOTA advances in real problems and its their "easier, vague target" to hit and finish their PhDs while getting published in top journals.
I'm also pretty wary of interpretability/explainability research in AI. Work on robustness and safety tends to be a bit better (those communities at least mathematically characterize their goals and contributions, and propose reasonable benchmarks).
But I'm also skeptical of a lot of modern deep learning research in general.
In particular, your critique goes both directions.
If I had a penny for every dissertation in the past few years that boiled down to "I built an absurdly over-fit/wrongly-fit model in domain D and claimed it beats SoTA in that domain. Unfortunately, I never took a course about D and ignored or wildly misused that domain's competitions/benchmarks. No one in that community took my amazing work seriously, so I submitted to NeurIPS/AAAI/ICML/IJCAI/... instead. On the Nth resubmission I got some reviewers who don't know anything about D but lose their minds over anything with the word deep (conv, residual, variational, adversarial, ... depending on the year) in the title. So, now I have a PhD in 'AI for D' but everyone doing research in D rolls their eyes at my work."
> Those same people will likely at some point call for a strict regulation of AI...
The most effectual calls for regulation of the software industry will not come from technologists. The call will come from politicians in the vein of, e.g., Josh Hawley or Elizabeth Warren. Those politicians have very specific goals and motivations which do not align with those of researchers doing interpretability/explainability research. If the tech industry is regulated, it's extremely unlikely that those regulations will be based upon proposals from STEM PhDs. At least in the USA.
> faking results of their interpretable models
Jumping from "this work is probably not valuable" to "this entire research community are a bunch of fraudsters" is a pretty big jump. Do you have any evidence of this happening?
> If I had a penny for every dissertation in the past few years that boiled down to...
This is very, very accurate. On the other hand, I oftentimes see field-specific papers from field experts with little ML experience using very basic and unnecessary ML techniques, which are then blown out of the water when serious DL researchers give the problem a shot.
One field that comes to mind where I have really noticed this problem is genomics.
I think it's a real open question whether the interpretable models are actually worse, or merely worse in competition/benchmark problem sets. The more deep models I build, the more I'm convinced that behind every inscrutable parameter hides a certain amount of overfitting, with maybe a few notable exceptions. E.g., can you build a decision tree that's not obviously overfit but is also susceptible to adversarial perturbations?
It’s also a real open question whether any of the interpretable models are actually interpretable, or even if they are in any well-defined sense more interpretable than “black box” alternatives.
In practice the answer is a massive “no” so far. Some of the least interpretable models I’ve had the misfortune to deal with in practice are misspecified linear regression models, especially when non-linearities in the true covariate relationships causes linear models to give wildly misleading statistical significance outputs and classical model fitting leads to estimating coefficients of the wrong sign.
Real interpretability is not a property of the mechanism of the model, but rather consistent understanding of the data generating process.
Unfortunately, people like to conflate the mechanism of the model for some notion of “explainability” because it’s politically convenient and susceptible to arguments from authority (if you control the subjective standards of “explainability”).
If your model does not adequately predict the data generating process, then your model absolutely does not explain it or articulate its inner working.
> If your model does not adequately predict the data generating process, then your model absolutely does not explain it or articulate its inner working.
That's a very dynamicist viewpoint. I don't necessarily disagree.
However, in what sense to the prototypical deep learning models predict the data generating process?
I tend to agree that a lot of work with "interpretable" in the title is horseshit and misses the forest for the trees.
I agree that there's probably a good question about to what extent looking at benchmark sets is biasing our judgment.
But it's also unclear to me how to get a decision tree to perform as well on image recognition tasks the same way that a CNN does. (Of course, as you mention, the CNN will likely face adversarial examples.)
The distinction between ML and programming is mostly propaganda, in the sense that it's not flat out wrong but is mostly used to win money/power. It's not actually a helpful way of understanding... anything.
This is exactly how once-great tech companies die. This is how governments lose the faith of their citizenry.