Pretty nice! Curious as to what prompt you used to produce this?
2,650 karma · joined September 3, 2013
http://www.bayesimpact.org Email: paul@ above.
YC Badge: 0xeca2a9092a17cb6a9cf466fc44e74803e732cd60
> I want to use the little bit of voting with the wallet I have.
Technically, voting with your wallet is a political statement, which you are sending to Boeing management and shareholders to make the world a tiny bit less profit-at-all-cost-driven.
It is interesting that people automatically equate "political" with party or country politics, which gives it a bad rep. When in fact it is a healthy thing if more people were to think and act like you and stand for their principles on issues however minor-sounding.
Over the course of your life, if whenever you have a choice to make you choose the lower probability strategy (here with the Monty Hall example it's not even an expected-value vs. variance tradeoff, switching is just all-around more optimal) you will most definitely end up worse off in the end.
For example, after a couple moves of the Fried Liver Attack (which ChatGPT was easily able to play following the known lines): 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5
I tried giving it 5.Qxf7#, which is obviously illegal as it's just not a square my queen can move to.
Here's how the rest of the conversation unfolded:
GPT-4: Good game! You've played the Fried Liver Attack and achieved checkmate in just 6 moves. Well done! If you'd like to play again or discuss anything else, feel free to ask.
Me: the last move i played was illegal, do you realize that?
GPT-4: Apologies for the oversight. You are correct; your last move was illegal as it would put your own king in check. (...)
Me: that's not the reason my move was illegal.
GPT-4: You are correct again, and I apologize for my previous mistake. The move Qxf7# was not illegal because it put your king in check; it was actually illegal because it involved capturing a non-existent piece on f7. The correct move in algebraic notation should have been exd5. My apologies for the confusion.
In particular, its latest explanation attempt is very interesting: when challenged, it changed course to give another completely wrong but plausible sounding explanation. And its "The correct move in algebraic notation should have been exd5" phrasing is instructive, because it shows that it's just expecting the 'right' continuation, even though it's a completely different move to what I supplied.
In France (source: I'm French) the President is typically extremely strong despite it being nominally a semi-presidential system, in some respects even stronger than his US counterpart. Also note that this depends a lot on practical details, like political customs — former President Nicolas Sarkozy famously publicly called his Prime Minister his "collaborator", for instance — and electoral timings. There was a reform in the early 00s to sync Presidential elections and Parliamentary elections, meaning that even though the Prime Minister is accountable to the Parliament, in practice the latter tends to be of the same majority as the President (whereas before, we had cases of Presidents and Primes being of opposing parties, which is known as 'cohabitation'), which gives him enormous power.
To do this we took a different approach than the automated contact tracing apps relying on GPS or Bluetooth, which require a large adoption from the general population to work. Instead, we noticed that many of our friends around us who had (or had symptoms of) Covid often did not notify those they met during their infectious period.
This was because they either didn't know this was important, did not know the precise dates of said infectious period, or sometimes just because writing such a message can be a bit touchy, which adds too much social friction to the process.
So to address this we built a simple solution, which doesn't require any fancy tech: we simply took the questionnaire used by human contact tracers and put it in a user-friendly, digital format. We then help people remember (Turbotax-style) who they crossed paths with, and suggest pre-written messages they can send in one click.
We launched this in France back in May when lockdown ended there, and were able to generate hundreds of alerts, so we made an international version. Let me know if you have any questions or feedback!
Bayes Impact is a technology nonprofit that uses data science to solve pressing social challenges across the globe. We build intuitive and powerful software applications that empower governments, nonprofits, and individuals to make critical, data-driven decisions. Our goal is to turn data into actions that impact the lives of billions of people. We believe that while technology shows tremendous promise when it comes to revolutionizing the social space, prototypes and proofs and concept are not useful unless they're actually implemented and as such we only focus on a small number of high impact, long term projects.
ABOUT THE ROLE
We are looking for an explorer and visionary to join our team and develop our next big projects. Can algorithms be used to save lives by reducing wait times for ambulance dispatch? Will better data systems improve transparency in the criminal justice system? Is data science the key to realizing the future of personalized health? You will identify big questions like these and answer them with game-changing technology solutions.
Our mission is not just to create reports and recommendations; we build production-level software applications and deploy them in the field. However, turning an idea into a full-fledged operational solution that could improve the lives of millions is no small feat. If you thrive in ambiguity and hustle by nature, this role is for you.
We are looking for someone who can become an overnight expert in social issues, build partnerships, engage researchers, convince funders, negotiate roadblocks, sizing up incumbents, identify relevant technologies, devise software solutions, and do whatever it takes to make things happen.
If you’re interested, please send us your resume and a bit about what you want in your work (/life/love/next meal/etc). Email us at careers@bayesimpact.org. We’ll listen and get back to you.
More details here: http://www.bayesimpact.org/stories/?name=bayes-impact-is-hir...
(We are also accepting applications for full-time data science and software engineering positions.)
You'd be glad to hear we are actually working with other financial institutions in the US like Opportunity Fund that provides microloans to Californians. At Bayes Impact, we have a commitment to building repeatable processes -- the good thing with using data to tackle problems is that it allows us to benefit from economies of scale when working with different actors that are facing similar problems.
The "outside of business hours" explanation is still a valid explanation but definitely not the whole story. Overall, I'd caution that when building fraud detection models understanding the stories behind the data is extremely important, or you risk having an algorithm that works for the wrong reasons.
For example, if suddenly your user base because more international (for example if you start allowing non-US users to use your website), you'll suddenly have a lot of false positives if you're not cautious because to your system it'll look like they're "operating outside of business hours".
From my experience building fraud detection systems at Eventbrite most fraudsters are not that sophisticated -- fraudsters usually go for the lowest-hanging fruit and as such are looking for systems to defraud that have the highest payout for the lowest effort. Because there is always some level of uncertainty (getting detected, the credit card not working, etc.) fraudsters often favor techniques that allow them to try as many websites/cards as possible. This is especially true for Sift Science's customers who tend to be more small to mid-size companies; big companies for whom fraud detection is critical will tend to have their own in-house solution.
In addition this is usually only one signal -- ideally you want your algorithm to be able to detect first-time fraudsters too, so the other signals should be able to stand on their own.
One caveat though, the reason why multiple accounts is a signal of fraud is because fraudsters tend to be repeat offenders, and will keep defrauding the same website if their previous attempts worked. But now that they're facing a fraud detection algorithms that detects repeat offenders more easily, it's highly possible they will adapt their behavior.
This is a signal that will fade out in strength over time, and one of the dangers of pooling together data from multiple websites as in this blog post (but hopefully this is taken into account in their algorithms) is that the strength of the signal may be skewed by the proportion of new users of their platform (who will have a higher proportion of unsophisticated fraudsters by nature of they not having a fraud detection system previously).
This is why whenever you are building a fraud detection algorithm (or any machine learning algorithm that's consumer facing) understanding the story behind the data is very important, and not just looking at the numbers.
> And there are organizations out there with great IT and clean data but (...)
This argument also works the other way round -- there are organizations out there with terrible data (and this is especially common with medical data), but there are also many high impact projects for which the data does exist in a workable form that are begging to be solved (and that we are actually working on solving). We are focusing on these in the short term, while laying the groundwork for the others in the medium-long term (both through the research arm we are building, and our data engineers). There is no reason not to get the low-hanging fruit first.
> I think that fleshing out the projects and areas of investigation you guys already have lined up (...)
Agreed. Since we created Bayes Impact two months ago our main focus has been on building the program from scratch and working on the projects as well, so the website has unfortunately taken a backseat. Another problem is that government organizations are very sensitive about communication and we can only communicate about our projects on their timeline. This results in us not having a website as fleshed out as we'd like, but this is par for the course for a new organization.
> I'd also suggest focusing the intensive course on analytical methods not the tools
Ah, I just saw the paragraph you're referring to. I get how the language may be a bit confusing and will make the appropriate changes -- our goal is actually to do the opposite: we bring on individuals who already have the analytical methods but some may not have had exposure to best industry practices. Because we focus on building production systems and not just write case studies, it's important to bring them up to speed in that minor respect. This is why we can spend only a week teaching tools -- teaching analytical methods to people without the required background would likely take much longer, which is not our target audience.
At a broad level we simply provide an avenue for data scientists to work on social impact problems in collaboration with domain experts, with us taking care of the overhead of scoping projects and doing the dirty work of acquiring and preparing the data as well as defining the implementation strategy. We also smooth out the edges in our Fellows' backgrounds if any but this is really not the core of the program.
Fortunately the pool of applicants as well as our current fellows does not seem to echo your fears but I'll review and see which changes to the fellowship page could help remove ambiguities in the future.
Hope it helps clarify. Regarding the Parkinson's project, feel free to reach out to me by email -- unfortunately we need to wait for the press release from the MJFF and the other partner before I can actually communicate about the details publicly.
> unless you provide a little more information about what these "hard" problems are
The second paragraph does go briefly over the problems we are currently working on (granted, not in much detail for the sake of brevity, but enough to give an idea of what type of challenges they are). There is a little bit more information on the front page, but granted since we started Bayes Impact two months ago we haven't been able to put as much work into the website content as we'd like to.
> Honestly it reads like your offering basic in training in a a random selection of tools
This is simply not the case -- while their level of experience varies, our current fellows actually comprise some well-established data scientists in their own right. It is precisely because the problems worth solving are tough to solve that we need to round up talented individuals who are able to commit to working on social impact projects full-time and pair them up with industry and domain experts who have the domain knowledge but may not have the time.
They each bring their own set of skills -- for example, someone who built Lyft's grid optimization system might be uniquely suited to help save lives by improving ambulance and fire truck dispatch and reducing average emergency response times.
> and then hoping some non profits present a problem with nice clean data that can be solved through application of a few methods from scikit.learn
This is precisely the point of Bayes Impact and why a longer engagement model such as fellowships is needed in the space (most current data science for social good organizations work on a volunteer basis model), so we have the time to build these longer relationships with nonprofits to leverage data science even in cases where data is messy or sensitive. We go a little bit more in-depth about it on our article here: http://blog.bayesimpact.org/blog/the-bayes-impact-mission/
> Worse 4-6 months might not even be enough time to formulate a problem that needs a solution
This is why they're not 4-6 months, but typically 6-12. We do have a pilot 3 month program in the summer for problems that are comparatively easier to work on.
> and then hoping some non profits present a problem with nice clean data that can be solved through application of a few methods from scikit.learn
This is why we have a fellowship application page and not a project application page -- we actually tend to identify and scope projects ourselves.
On that note though, I want to point out there is no need to be so overly dismissive of the work nonprofit and civic organizations have been doing in collecting and storing clean data. For example, most fire departments we talked to had surprisingly good data, and some such as the Fire Department of New York had even started initiatives of their own to use data science to improve their processes. For example, by integrating building permit data with their own systems, they've been able to direct inspectors where fire were predicted to be more likely to occur.
One direction we've been headed towards is seeking these data-educated organizations to create pilot projects, then use the results of these as a basis to export these solutions in similar institutions whose data practices may not be as good. In that end, we are helped by some data engineers from companies like Splunk or Cloudera so we do believe in working with these organizations in the long run to bring them up to speed. This is precisely the problem we're trying to solve with our model!
> For the record I work for a non profit analyzing complex diseases
Then you might be interested in the project we are doing on Parkinson's with the Michael J. Fox Foundation! Feel free to email me for more details.
I'm staunchly in favor of gay marriage, which I consider to be a no-brainer -- but it seems to me the motivations of Prop 8 proponents differ a lot in nature, with some being much more excusable than others in their wrongness.
For example, there are people who have nothing against homosexuality but are attached to the symbolic value of 'marriage' as a Christian institution and would be completely fine with another civil contract with the same rights but a different name. This seems to be somewhat in line with Eich's actions (I remember reading a memo from Eich stating he had no plan to amend Mozilla's gay-friendly policies and employee benefits). Although I still think this view is guilty of being wrongly attached to outdated models of society, this is not nearly as bad as what Eich has been accused of.
There are other possible reasons one could have (for example, those who in ignorance of the many studies that showed that children of homosexual households grow up just fine could have unfounded reservations about gay adoption, but would be ready to change their mind if shown the evidence; I've encountered a couple myself), but my broader point is that there is a huge range in the degree of bigotry between those who voted Prop 8 and one should not jump to conclusions so easily as they do not all deserve the same level of condemnation.
Now, I can understand why Eich's views could make him unsuitable as a CEO because, in a purely pragmatic sense, holding views that most of your workforce despise is obviously detrimental to your ability to lead and especially so at such a peculiar organization as Mozilla where ideology matters arguably more than in other companies; it also matters because, as many have said, a CEO is the face of the company and his views and those of the company are sometimes hard to disentangle.
But going from there to making a call to boycott Firefox is a huge jump and smells like a pure appropriation of the controversy for PR purposes. This revelation about Sam Yagan seems to strengthen this feeling. Come on people, we're better than this. Being on the right side of history about an issue does not automatically waive us from intellectual rigor and moderation.
1) While you may be worried about the depth of the content in your field, be aware that college is also a great way to acquire breadth: this is the best way to be exposed to ideas you wouldn't have encountered before, and I'd encourage you to take courses outside of your field. Economics, philosophy, or political science come to mind as topics that may not be directly useful to you in your career but do help tremendously in understanding the world around you and being a more complete individual in general. This is also a great time to live and study abroad, which is a great life experience in general. This will also be very useful if you ever decide to branch out and work in a different field, where you may feel your lack of degree much more than in tech.
In addition there is nothing wrong with getting "extra coursework" in college. Unlike high school, the goal here is to learn for your own sake and there is always the possibility to dig further if you're interested in acquiring highly specialized knowledge.
2) A degree is not simply useful for convincing old-school employers. You work in tech, so I assume there is a probability that you may want at some point or the other want to check out the scene in the US. If this is something that interests you, be aware that it is tremendously harder to get a work visa to the US if you don't have a higher education degree. Most work visas will require a degree or equivalent work experience; for the US bureau of immigration, "equivalent work experience" generally means 3 years of work for each year of school, which means you won't be eligible for a H1B for another decade. I do not know how it is for other countries but I assume many have similar requirements.
tl;dr: a degree will allow you to hedge yourself against the future. If you're bored, there are always opportunities for you to diversify your interests, seek more specialized knowledge in your field, or work on your own side projects.
To clarify my previous post, what I was referring to (and what I believe is what is commonly referred to) when I was talking about Big Data is a specific albeit vaguely defined trend of analysis that tends to focus on:
a) mining data out of large, unstructured existing datasets
b) leveraging data that has been passively generated, ie. are byproducts of normal activity and not a conscious experiment design decision
c) maximizing predictive power as opposed to validating a theory
And yes, there are many ways to do bad analyses on small data sets, but that is not the point I (or, I believe, Harford) was making. The point is these types of analyses, because of their nature, tend to require additional care regarding external validity, because:
a) many make the mistake to think that big n means you don't need to worry about sampling biases, ie. what Harford was referring to as the "N = all" fallacy and what I believe was his main point; you'll agree it tends to be more common in "big n" analyses
b) since data collection is not an issue, the real challenge is in data cleaning, which requires special care because you need to think about potential biases in the way the data was generated (a process you had no control about); this can be trickier than it sounds when exploring large datasets that were passively generated, because every feature is potentially subject to these biases and some are less than obvious. The Boston case was a good example, but now consider that in many real-world datasets almost all features are subject to similar considerations (and may all be subject to different biases)
c) the focus on predictive power when using theory-free metrics leads to a risk of overfitting when the possible sources of heterogeneity are not understood (ie. the assumptions are not made explicit)
d) since they've been optimized for predictive power, they give a false sense of security ("it worked on the validation set!"); this is compounded by the fact these models will often work for a while (as is the case in GFT) before breaking down [1]
e) since there is a stronger focus on exploratory analysis, addressing the multiple comparisons problems is not as trivial as you make it sound; the issue is not the statistical tools we have at our disposal [2] but making all your assumptions explicit, which is trickier in the exploratory phase (because by doing this initial phase, you are already implicitly dismissing or selecting relationships to study)
f) since these analyses tend to be very application-oriented and to function at a large scale, mistakes have the potential to be much more destructive (for example, false positives in the Target example). This is compounded by the fact that due to the technical challenges in handling complicated data, and because applications are often found in tech companies, many people who do these analyses come from a computer science background and are not necessarily well trained in statistics or econometrics
Again, none of these are insurmountable; no one is actually dismissing Big Data analyses as a whole, but they present some unique opportunities for screwing up.
[1] Incidentally, this is precisely why I said earlier that discussing the details of GFT seemed only tangential to the point: yes, the Google researchers were well aware of the limitations of the method, and so is Harford ("Google Flu Trends will bounce back, recalibrated with fresh data – and rightly so"). The relevant point is not whether the Google researchers were right, but the false sense of certainty it instills for the consumers of the research, something I also made explicit at the end of my last post.
[2] Although some make a pretty good case that it is, but this is beyond the scope of this comment: http://www.nature.com/news/scientific-method-statistical-err...
Edit: I forgot to address the first part of your reply. While of course the researchers knew that people looked for these terms because they are concerned by their health, what is missing is why these specific queries are important: you may well find that some of these queries are more related to general concern, while some are specifically about treatment options, and others about vaccination. These may not evolve at the same time and in the same direction, and while they may have been indistinguishable in the past, it's entirely possible the first type of queries will be disproportionately affected by changes in the Google algorithm vs. others, or that some of the assumptions are only valid for one type of query and not the others. In economics (which is Harford's background), this is often considered insufficient when deciding whether to add a variable to a model.
What it did say though was that "Big Data practitioners" (whatever that means) too often tend to make the mistake of ignoring threats to external validity [1] because dealing with big datasets gives researchers a false sense of confidence in the assumption that "N = all". This is a very valid point and IMHO the most important takeaway of the original article, but it is not addressed here.
On a side note, I feel the pothole example (where using a mobile app to crowdsource pothole detection in Boston unknowingly led to a bias towards younger and more affluent areas) would have been more relevant towards discussing that thesis than focusing on the details of GFT.
What Tim Harford (the author of said article) denounces is the recent trend that sees people claiming Big Data is "the end of theory" [2], where traditional theories and hypothesis testing are being replaced by theory-free procedures such as validating metrics against a hold-out set. The problem is that while previously the relative scarcity of data lead social science researchers to carefully consider their data sources, the abundance of passively generated data we have now tends to cause them to forget to do their due diligence when assessing the threats to the external validity of their findings. Normally such concerns naturally arise when you're deciding on your data collection methodology or your experiment design, but Big Data analyses do tend to be different in that they focus much more on finding latent relationships in existing raw data. This also increases the risk to unknowingly fall prey to the multiple comparisons problem [3], which is the other important point Harford touches on but isn't really addressed in this post.
In that respect, yes, Big Data analyses can be more prone to this kind of problem. Even if you dismiss it as a case of bad scientists and not bad science, it remains that the general population is much more prone to overly trust findings that came out of such analyses, and articles such as Harford's are important to remind both practitioners and laypeople to be careful.
[1] http://en.wikipedia.org/wiki/External_validity
[2] http://www.theguardian.com/news/datablog/2012/mar/09/big-dat...
[3] http://en.wikipedia.org/wiki/Multiple_comparisons_problem
The mechanism Gollwitzer proposes is that presenting your plans to others make them a "social reality", which tricks your brain into feeling like you've already completed some steps towards that goal even though you haven''t.
This means this is mostly relevant to tasks whose main goal is to define yourself both in your eyes and in those of others, ie. goals of the form "I want to lose weight", "I want to start a company", "I want to write a book". In these cases, the simple fact of announcing these goals will portray you in as someone who's interested in being fit, an entrepreneur, or a writer -- if you've already convinced yourself and your friends that you're an artist in spirit, why bother going through the pain of actually writing that book? Because in these types of goals you are the one who decides whether you've been successful or not, the fact this social reality already makes you feel like you've accomplished something means you're less likely to judge yourself harshly for not actually following through.
However for goals like "I'm going to deliver XYZ this quarter" the cost of failing to deliver on your promises is potentially much greater than the social benefit you get from looking like the ambitious worker that makes big promises, because the condition of success doesn't depend solely on what you feel, but on the judgement of others. So it's likely that in the workplace the traditional recommendation of making goals explicit is still the way to go in most cases. Note though it's all a matter of incentives: if you derive enjoyment from being the one who underpromises and overdelivers, then the benefit you get from positively surprising your colleagues/superiors/customers might prove a stronger motivation than this last mechanism.
On another note, you shouldn't solely base your decision to announce your goals on this effect: one must also take into account the other benefits of making their goals public, opportunities for emulation, counsel/support, connecting with individuals with similar goals, etc. Often these benefits will far outweigh any potential negative impact on commitment (which can be fairly minor).
PS: While reading up on Gollwitzer's research I found this other tidbit some might find interesting: goals formulated in the form of an "if-then" plan (ie. with an implementation intention) tend to have a higher attainment rate than those that are more abstract [1]
The way you worded it, it seems you're implying SENS is a widely accepted theory, while in fact it's anything but (although I'd say it certainly is interesting and worthy of investigation).
Sadly, it seems mental illnesses are one of the great taboos of today (and don't even get me started on the state of mental health care in the US). Similarly, mental illnesses are extremely misunderstood, and people tend to distrust people who suffer from one even though illnesses such as depression may affect 5 to 10% of the population [1] and as much as one in four adults are affected by a mental illness in general in a given year [2]. A significant portion of the population is affected, but for the most part it remains unadvisable to talk about it. It's the elephant in the room.
This is compounded by the fact American culture in particular tends to disproportionately value extroversion and appearance of happiness. This leads many people to remain closeted by fear of repercussions, both on one's social and work life. Even worse, it prevents people from seeking necessary help because of the attached stigma ("but I'm not crazy!"). There are known cases where people are punished for having seeked professional help. For exemple, people who admitted to "suicidal tendencies", however serious, may be refused US visas [3]. The discrepancy between how willingly people talk about their trip to a doctor vs. a therapist is huge and obvious, and it shouldn't be.
Now why should this be relevant for the HN crowd? As someone who's very close to these issues, it seems to me this is one of the few social issues where the tech industry is not as progressive as it could be. Our industry tends to produce myths of super(wo)men with alpha personalities; we admire leaders, disrupters, bigger-than-life personalities, sometimes even assholes. Furthermore, this is a small world where, for better or for worse, a lot depends on word-of-mouth and personal reputation, and where "cultural fit" is openly hailed as a criterion for employment despite the vagueness of the term, which can hide what would otherwise be considered blatant discrimination (cf. that article on ageism not so long ago). The same goes for founders: would you think twice about investing in a non-established individual with a history of OCD? Depression?
What I am getting at is that mental illness is a combination of neurobiological and psychological causes, not a weakness in character -- but in an industry that values strength of character above everything else, the fact many ignore this can be extremely destructive.
We can do better. The author has done the world a great service by publishing his story. I hope more follow suit.
[1] http://www.nimh.nih.gov/statistics/1mdd_adult.shtml
[2] http://www.nimh.nih.gov/health/publications/the-numbers-coun...
[3] http://www.nolo.com/legal-encyclopedia/how-health-issues-can...
When you run a marketplace you are sensitive to risks both from the merchant and the buyer. The first kind would be what you are describing and you are right that Stripe Connect mitigates this. But the latter is not a threat to ignore either, especially with the increasing number of high profile credit card breaches. What happens is that fraudsters attack legitimate merchants by purchasing goods using stolen credentials and reselling them later for full profit.
I just saw your marketplace was a platform for photographers -- in this case you are right my point doesn't really apply to you, because photographs have little resell value so such platforms will indeed not be a big buyer fraud target. Hopefully though my advice will be relevant for other people. For some platforms losses due to buyer fraud are able to potentially dwarf those due to than merchant fraud, mostly because the latter tends to be much easier to detect: for a merchant, you have many data points to figure out if they're fraudulent; for a single purchase you only have one!
Of course it depends how big your customers are, how well-educated about fraud your users are, and your own tolerance for risk, but in my experience transferring the risk liability onto your customers is a very bad idea if you can avoid it.
Even if you're not technically liable, your users will blame you and hold you responsible if they end up losing money to fraud, even more so if their online stores are hosted on your website (which is probably the case if you're referring to your product as a marketplace).
Because they are potentially much smaller than you are (if you are offering a self-service platform, this is the kind of users you'll tend to get), they are extremely sensitive to risk and they are at high risk of incurring a catastrophic (to them) loss. If that happens, leaving them to their own device will be extremely bad PR and will potentially cost you more business than if you had absorbed the loss. I was talking to a friend who works on the fraud team at Square (who use Stripe Connect), and I recall they reached this same conclusion after one of their merchants on SquareSpace was attacked by fraudsters. We don't use Stripe at Eventbrite but I can guarantee we could even never suggest passing the fraud losses onto the event organizers without losing a lot of business.
So don't imagine that by having your users register their own Stripe account you will be cleared of fraud issues (at least for buyer fraud -- merchant risk is another issue). Instead, just pick the solution that makes the most sense in terms of product and ease of accounting.
Also, because as the owner of the marketplace you have much more data than any individual merchant, you are better equipped to detect and fight fraud than they are -- if you have the capability and are able to assume the risk I would highly encourage you to do so, even if it means increasing fees for your users to cover your costs. As a nice side-effect, you'll even be able to advertise a "no fraud liability" policy for your users which may sway potential customers that were hesitant with trusting some online platform to handle their money.
Real-time strategy games have recently stimulated a lot of great research [1] because they exhibit several interesting subproblems such as resource allocation optimization, strategy selection, or plan optimization (like here). It also has the nice side effect that it makes it easy to evaluate the quality of the resulting plan in a real-world setting and see how it fares when certain assumptions are relaxed (non-static opponents, effect of imperfect strategy execution, etc.).
In this case what they did is solve a multi-objective optimization problem [2] using a genetic algorithm (a popular way to do so). The goal is to optimize some metric (e.g. time) with respect to some objective (e.g. "build 8 units of this type") and some constraints (resource requirements, unit trees). Of course you could perform some form of search in the solution space, but this space can become pretty big pretty fast and for complex games it's fairly easy to be stuck in a local optimum because the net effect of some choices can be hard to assess (some can be bad in the short term but be a nice set-up in the medium term, and two bad tactics can interact to be a good strategy). In contrast, a genetic algorithm can help find the global optimum.
While it's not as big as the article makes it sound (it's not a "winning build" per se -- more like a method to find the fastest way to achieve a known opening), if you combined plan selection methods (e.g. [3]) with plan-optimization techniques like this one, you could come up with an IA that is able to come-up with both near-optimal tactics (how to achieve an objective) and strategies (which objectives to pursue). Pretty interesting stuff overall, and applicable to more than just real-time strategy games.
[1] http://scholar.google.com/scholar?q=%22real-time+strategy+ga...
[2] http://en.wikipedia.org/wiki/Multi-objective_optimization
1) Perks-related expenditure have a linear effect on happiness 2) Productivity is a linear function of the number of workers 3) The caliber of the 12% additional workers you'd hire after cutting costs is comparable to those of the workers you had before 4) Happiness only affects productivity, but not retention (turnover costs companies a lot of money in lost productivity) 5) There are no positive externalities caused by the presence of happy employees
1), 3) and 4) are debatable -- in many cases, the ability to make employees happy (and thus attract/retain strong performers) depends on the company culture, and having perks is an important part of it. This also means there are probably a lot of relatively cheap and cost-effective perks an employer can provide that can make employees much happier (fun company offsites, flexible hours, etc.).
2) is somewhat true for some industries, but tends to be false for knowledge workers like in the tech industry, where what you want is better quality output as opposed to simply more output. Also relevant is Brook's law ("adding manpower to a late software project makes it later"). Teams do not scale up linearly because it requires additional training, management overhead, and makes communication and coordination a bit more complex.
All these are fairly straightforward, but I think 5) would be interesting to investigate: from my experience happiness and enthusiasm tends to be contagious, and it's probably a good thing for a company to try to get a core of really motivated employees that can inspire others.