mlthoughts2018

No. Absolutely not. Your comment amounts to nothing but gainsaying and is completely ridiculous.

It is not toxic to oppose wanton racism and fascism. If you are ashamed of opposing racism, then you are being toxic and you are enabling and sympathizing with racism. That’s on you.

You don’t get to just gainsay the opposition, claim it is “toxic” and then somehow have that treated like it’s on equal footing or deserves credibility in the same way that the original call outs of toxic racism or fascism themselves do.

You’re simply acting as part of the racist and fascist problem yourself at that point.

This long form Twitter stuff is so miserable. This seems like an interesting story, but I couldn’t read past the second segment.

And no, Threader doesn’t solve it or help (even taking for granted the extremely unlikely assumption that I was willing to migrate the thread to another service and navigate to that other service just to read it).

This is interesting in conjunction with the story itself, a man having a hard time adjusting to digital interfaces.

So much of it is caused by the abject poorness of digital interfaces foisted on us that we just accept and stumble through.

The problem is that what you describe is a question of pragmatics and specific data curation, which is not sexy or marketable for political fairness and bias researchers. They need to manifest false angles of the problem, connections to climate change or #metoo or whatever is the social justice outrage du jour, so that it can win news cycle attention, TED talks and political notoriety. It leads to perverse incentives where AI fairness stops being an engineering problem that should be exclusively focused on objective criteria that need to be met to solve concrete problems, and starts being a tribalistic song and dance about wokeness, cancel culture and shoe-horning subjective values into learning systems at a strategic level.

Unfortunately AI fairness is not a serious research domain yet. It’s a venue for trying to create political cottage industries around convenient social justice drama and then trying to parlay that into a career by building political moats such that if anyone questions the legitimacy of this or that aspect of “fairness research” then that skepticism alone can be used to cancel them and argue they are regressive.

Ugh, here we go with more of the sanctimonious nonsense about AI fairness research from the Times.

Google was well within rights in firing both employees (Gebru sincerely quit and was not fired, given her ultimatum, but Google would have been in the right).

I’m no Google apologist - their ads monopoly should be broken up, they are anti-union, pay harassers huge sums of money and lead the way in poorness of leetcode software culture and low accountability. Google has poor customer service and largely has no idea how to build products for people, constantly shutting things down, killing acquisitions, and using poorly attended message forums as a bureaucratic wall to steamroll any real way of seeking customer support or appealing black box takedown decisions. Google is a mess.

That doesn’t make Gebru in the right - she’s not at all.

Here are several previous comment threads going on at length about why Gebru is sincerely in the wrong here:

- https://news.ycombinator.com/item?id=25314824

- https://news.ycombinator.com/item?id=26147178

- https://news.ycombinator.com/item?id=26238365

As long as there are open-plan offices instead of private offices for each knowledge worker (even in dense urban real estate markets) then please stop talking about developer productivity or efficiency.

Companies utterly don’t care. The degree to which they don’t care is so exceedingly flagrant that they use open-plan offices.

What more can there possibly be to discuss?

I wrote up a post about some inherent, fundamental contradictions in leveling / salary band approaches to structured career development.

I think it’s under-appreciated that these systems are so inherently flawed, especially regarding meritocracy.

https://managingml.substack.com/p/a-cap-theorem-for-career-d...

Thinking it is an unjustified proportion seems exceedingly racist on your part. The equivalency between proportion of attention paid to harmed groups and size of the group, especially for groups as large as ~12%, is inherently quite an extreme racist stance to take.

I’m sure you wouldn’t agree, but that’s exactly the problem. As long as we’ve got people acting like NPR’s current amount of coverage including racial angles is somehow “unjustified” then we clearly need to terraform more and more racial angle of content into the discussion to educate that racist ignorance out of existence.

The same is true for sexism and poverty as well.

None of this makes it better. “We give candidates all the time they need” (which you suggest is something insanely short notice, like 1-2 weeks) is not an answer. What about people who cannot give you time outside of 1-2 hour blocks within business hours of a weekday, as in they can never work on your assignment apart from that. You’re just openly admitting that you won’t assess that whole cohort of humans. I urge you to reconsider. You are determining a process that you happen to fancy, maybe because it makes you feel like you’re a visionary of interview process or you’re going against the trend, but you’ve got blinders on about your commitment to this and what it means for people who just happen to have obligations.

The standard whiteboard hazing trivia is bad enough, and 4+ hour long marathon on-sites that require weeks of study and taking vacation days just to interview are ridiculous. But your method is not much better. You’re still perpetuating this myth that somehow, unlike virtually every other profession on the planet, software engineering requires intense examination of on-the-spot demonstrated skill, rather than simple conversation probing past experience and domain knowledge.

We just do interviewing completely ass backwards in software. Part to assert gatekeeper superiority, part to judge fealty. It’s just bananas.

Sorry but no way. You send me a programming task you expect me to do in my personal time, and you’ve just revealed you practice extreme ageism, discriminate against parents or people with elder care responsibilities, and many more negative and disrespectful signs. It shows excessive entitlement on your side, indicating a poor working environment. Out of consideration for finding a healthy, reasonable employer, I would reject you immediately, and so should everyone. It’s not a matter of some candidates liking your process and not others - the process you describe solely perpetuates destructive and exploitative interview behaviors for everyone, even people free of outside responsibilities who can have the luxury to make room for a multi-hour take home exam within a month or two of being assigned.

> “You can brag about owning such-and-such an image, but only as long as enough other people desire images from that particular set.”

I think that’s the whole point. An enduring aspect of human culture is that we praise and confer social status on owners or stewards of rare curiosities. You aren’t buying NFTs for tangible value, you’re buying a speculative investment in social status. And people feel very confident that the market cap and liquidity for buying and trading this type of social status will go way up, and that unique, verifiable digital ownership allows the pay-for-status desires to take place untethered to physical goods.

? it seems like you are expecting bitcoin to achieve the same ease of use as a payment processor like Visa, but that is not and never has been the goal or point of bitcoin. “Ease of use” is a relative term. Bitcoin is deeply impractical compared with fiat currencies backed by strong governments. Who cares? It is deeply practical when compared to fiat currencies for unstable or severely corrupt governments.

It’s ok. The type of HN users who downvote comments like this do not offer anything. The downvote literally means nothing apart from their tribalistic “bitcoin bad” unfounded assertions. It’s very easy to not care.

Meanwhile, people in Venezuela who diverted some assets into bitcoin can actually transact for food or medicine at prices less volatile than the local fiat currency, even including high transaction fees and verification times. Things they can’t do with precious metals held by custodians that transact in the fiat currency, let alone the fiat itself.

But by all means, let’s hear some more specious “Bitcoin mining uses X amount of (largely renewable) energy and I personally happen to think X is too large for what I perceive to be no benefits” diatribes.

But bitcoin is just a scam of course.

People have been trying this for a really long time and there are just a ton of business considerations that make it impractical.

1. People buy photography, even stock photography for ads or business uses, with some intention to affiliate with the photographer / artist or some notion of an artsy style or aesthetic. Autogenerated art starts right off the bat at a disadvantage for being “commodity” in nature, even compared to repetitive inventory of sites like Shutterstock. Maybe you can get past this for certain niche areas where the real photos are already exceedingly commodity, like backgrounds, office photos, landscapes. But even then, the status of an artist counts for a lot.

2. It’s not actually that cheap to operate generative image at scale. You have to ensure that pre-generated content is of sufficient quality, and covers sufficient subject matter, compositional and aesthetic variety. If content is generated on the fly, you’ll be dealing with pretty high throughput on a very resource intensive model.

3. Competition can replicate your image model pretty easily, so your differentiator comes back to branding and a sense of “not commodity” quality, as well as all accompanying services and support, which is where all your operating costs come from anyway.

I am sure generative inventory will become a bigger trend in stock photography, but I doubt it will be much of a differentiator. If you run a stock photography business with more than a few million images already, you would be better served building ML solutions for search, discovery, keyword or caption annotation, abusive content detection, automated aesthetic enhancements or assisted editing tools & style transfer. There won’t be a “holy grail” of generated inventory. Most customers just won’t care.

It’s a good example of how impressive exotic ML solutions might seem like they surely have to have consumer applications, but where it interacts with business concerns it just doesn’t matter. Monetizing ML solutions is really hard - much harder than creating the ML solutions in the first place.

> “ Why would this be an area to discuss what guidelines other kinds of businesses should follow?”

Because what guidelines to consider and why, in comparison with what ICANN considered, is highly relevant to the thread.

Why does ICANN being “private enterprise” relate to the question of when it’s appropriate, in general, to consider the stated items? Whatever logic you’re applying here doesn’t make sense, and the certainty with which you think others should have to constrain topics of conversation to meet conditions of your confused logic absolutely comes off as entitled and whiny.

The former American national soccer team manager, Bruce Arena, once was famously quoted describing a scrappy player, Clint Dempsey, with the phrase, “He tried shit.” Meaning - it was sometimes weird, didn’t always work, but he went for it and took risks.

You can think what you want about this deal, but Dorsey is similar. It reminds me of Amazon in a lot of ways. They will absolutely try anything, no matter how weird. Some start-up delivers tacos by drones? The next week Amazon has a whole new AWS managed service to deliver tacos by drones, and shuts it down the next year.

If there is any business out there right now that has some potential to be the Amazon of consumer finance, it is Cash App. Obviously the chances any business could do that are low. It’s by no means a clear path to success, but you at least have to admire Square for trying weird shit and being willing to do anything to see what works.

It’s bizarre anyone could have a reason to find that odd, since there could scarcely be any topic more directly and irrefutably related to this thread (regardless of one’s take on whether ICANN acted inappropriately).

You’re essentially saying, “I have a different system of logic from the common one, and I want my system of logic to be considered the right one. So I’ll just treat it as ‘irrelevant’ as a baseless gainsaying rebuttal to anyone else’s line of thinking.” How very whiny.

This article is all over the place. It takes an obviously negative stance right from the start by complaining that Coinbase has frequent outages and poor customer support (that hasn’t been my experience as a long time happy Coinbase customer), then it switches completely to the author complaining about how they didn’t research the fees, and got surprised by high fees that earnestly reflect the cost of operating the business. This is somehow better than the way brokerages fund zero fee retail investing transactions, but also somehow vaguely dishonest and predatory at the same time?

It’s a very weak, meandering article with no significant point. I usually expect better from Bloomberg on financial topics, but I guess it’s just the usual trendy anti-cryptocurrency drivel.

The question is asking when is it wise to (privately) do any of those particular actions.

Also, most social status systems developed organically from networks of trust, common experiences and mutually held values at a micro level like within a family, town, school, religious group, etc.

Comparing that organic system of social appraisal with anything like a mass scale surveillance social credit system, or international communications censoring organization like ICANN, is deeply unreasonable. It is clearly not the same thing, and centralized control of reputation by a government arguably poses huge totalitarian risks with zero offsetting benefits for citizens.

Why do you believe ICANN’s multi-government structure is somehow robust against any one government or small consortium of governments having too much power?

You’re missing the point. “Fling spaghetti code at the wall” will produce many failures and a few winners - the winners will not serve as evidence that poor spaghetti code is a perfectly viable business strategy.

Most of the companies you’re describing fold and fail miserably after just a few years. Only very, very, very, very, very few of them achieve any kind of significant funding or market success.

I wonder why?

Because they just wrote a big article about how they created a culture, developed special vocab (“slush”), and intentionally plan on a multi-year basis all around a code freeze that is not just a day, not just a weekend, but lasts for weeks and affects all systems (which either means no systems are in a state of proving their resiliency, or leadership doesn’t trust engineers, or both), and they are proud of this.

That is a whole mess of giant neon red flags that something is wrong here regarding engineering for resilience and trusting your deployment system to catch errors.

Additionally, for a freeze lasting weeks, unfreezing and resuming merges from some backlog of frozen changes would likely introduce far greater risk to revenue or to customers than managing a trustworthy deployment workflow during high traffic events. I would even question the fundamental premise that this is safer or beneficial to customers at all. It seems much, much more likely to be a “cover your ass” method to prevent any possible blowback from outages due to unremediated tech debt that the weak leadership is not capable of protecting as a worthwhile priority for quarterly goals.

The machine learning team is responsible for all search and discovery features that users engage with when searching our inventory, some of the highest traffic, uptime demanding services I’ve seen anywhere in my 15 year career. We also run all of the image and language processing that occurs in real time in the image editor post upload, along with a HA queue system that aggregates requests to route them to shared GPUs on the backend. In our case, many machine learning systems are on the critical path for customers, and we have a significant platform team within machine learning specifically to ensure we can support this, meet uptime demands, design auto-scaling solutions in our on-prem data centers, execute failover and disaster recovery (and drills to simulate these) and more. And so does every other team.

Yes, I work in a high traffic (Alexa rank above 400) ecommerce site specializing in stock photography, music and digital assets. We see a 10-15x spike around Black Friday, around Christmas to New Years, both in direct consumer sales, inventory upload, new sign ups (customers and sellers) and in ad inventory sales.

We don’t do code freezes except in very special, isolated cases of known failure risk. When we do need an isolated code freeze for one system, that is a “all hands on deck, this had better get fixed” kind of moment. If the larger system of most of our core services needed an extended code freeze to be safe, that would result in probably (justifiably) firing senior engineering leaders.

If you have a CI/CD system and you can’t trust normal automated deploys at any time, that is a huge problem. If this happened one year and you needed the code freeze, so be it - that’s just responsible risk assessment. If you intentionally plan the system to work this way every year that is egregiously bad engineering leadership.

I run a machine learning team in a high traffic ecommerce company (specializing in stock photography and digital assets). We see an even bigger seasonal spike than Etsy.

The need for code freezes around high traffic events is usually a sign of very poor engineering culture, where patching extreme tech debt is much higher priority than technical excellence or product improvements that help the customers.

A code semi-freeze for weeks is incredibly alarming. How much unremediated tech debt must be causing that?

That is a massive red flag that Etsy engineering culture must be something like the SRE tail wagging the product engineering dog. I would bet that you only get rewarded at Etsy for being a visible firefighter, stopping failures on nights or weekends, jumping in front of outage bullets, and that if you take the time to design systems robust enough and safe enough that they don’t need code freezes in the first place, you are never going to get recognized or rewarded.

It’s baffling that not only did they have to resort to a long code freeze - but they even approved a public blog post bragging about it.

Boy, that tells you a lot about the culture. Not going to be seeking jobs there any time soon.

To put in context, I work for an ecommerce company that routinely sees 10-15x traffic around Black Friday -> Cyber Monday, and around Christmas to New Year. We don’t do any code freezes for the load, almost all of which is handled through auto-scaling in our on-prem data centers.

The author mentions extended examples of Bitcoin operating exactly as intended as both a medium of exchange or value transfer that circumvents a failing fiat currency or geopolitical issues connected to economic instability, and as a store of value to decouple possible future purchase power needs from a bad / volatile fiat currency.

Then in the immediate next paragraph they assert, unmotivated and unsupported, that this makes Bitcoin a Ponzi scheme..

It baffles me the mental gymnastics people go through to criticize and dismiss cryptocurrency. Some people have just made up their minds against it, and even when they personally are writing paragraphs of examples of its legitimate, intended use, the confirmation bias that it can only mean “Bitcoin bad” is still so strong.