wilde

I suspect part of it is that he’d need to go and chase down permission for a bunch of the anecdotes since YC office hours have an expectation of privacy. But maybe there’s a version of this essay with a tic more detail you’d prefer?

I mean the companies advertising the hardest are often the ones who can’t hire and retain due to underpay, bad culture, etc. Unless you’re willing to spend a bunch of money on a comp benchmarking dataset, I think the best self-published one is levels.fyi. Those values are below median but not uncommon based on that data.

How is killing not a specific crime? Like usually the “incontrovertible” is the hard part.

Yes. https://www.law.cornell.edu/uscode/text/18/373

> Fundamentally, SQL lacks meaningful abstraction, and there's no sane way to package very common operations.

Tables are the unit of abstraction.

You’re on the right track with your instinct that you should be able to pull repeated work into a reusable unit. In many DB systems you can just register these sub queries as views and treat them as real tables. In the ETL world that I’m typically working in, you go one step further and just write pipelines to make useful intermediate tables.

Reuse tables, not subqueries.

> The effect on input->output behavior from a single dataset example feels obvious, whereas the effect on input->output behavior from a single line of source code may depend on me understanding the other zillion lines...

Sure, but you have no idea how much weight any given sample contributes to the model. Maybe the behavior is included. Maybe not. The weight assigned could even vary drastically between runs with different random seeds!

I probably picked a bad paper with CycleGAN. My main point is that the mapping from data to behavior is non-trivial and there are lots of examples of this in the lit. Medical models that realize that the output variable is encoded in the text in the upper right of the image rather than attempting to analyze the image, etc.

I really enjoyed this article! The author walks us through design of an ML system and typical problems in a fun and understandable way.

The conclusion is pretty wrong though.

> Regardless of how complicated your program’s behavior is, if you write it as a neural network, the program remains interpretable. To know what your neural network actually does, just read the dataset.

This is not true at all? That’s like saying “all binaries are open source, just read the machine code”. Most datasets contain a reasonable amount of pollution (even the author ran into this). And if you let the model train on its own output it’s pretty easy for it to cheat: https://techcrunch.com/2018/12/31/this-clever-ai-hid-data-fr...

Moreover current ML techniques are pretty sample inefficient. So your dataset is likely to be much, much larger than the equivalent program. Right now we haven’t developed much tooling to help you map a sample bad input or behavior back to training data that might be relevant. So “just reading the data” seems like lots more work than debugging a modern program.

I do think we’ll get better at debugging this stuff in time, but I don’t think it’s currently true that ML systems are simpler or more interpretable than a corresponding imperative program.

I used to work with a lot of ex-Pivotal folks. Their workflow was something like:

1. Talk through the requirements and write down some key bullets

2. Whiteboard some design options. Make sure you have at least 2 interesting ones.

3. Prototype them out. Decide on your favorite.

4. Delete the prototype.

5. Start writing tests now that you know where you want to be, and TDD from there.

The key point is that TDD has to start from a relatively defined sense of what you want. If you don’t have that it’s totally fine explore. Just don’t ship that code.

Gotta love the two ads that block the user from seeing the free tool lol

I like this piece, but it basically measured which app do you use more. If they wanted to really drill in, they should have had folks compare performance on new accounts. I still suspect TikTok would win? But you’d get cleaner results.

What stops me from using the first over-18 pin I can find? It only relates to age so there’s hardly any incentive to keep it secret.

Are all non-citizens banned from services now? As a tourist in your country, how do I access services?

Like I said, lots of problems. :)

CA literally just passed the bill you want.

That said, the problem is that sites don’t know for sure who is over and who is under 13. And more robust age proofs have lots of problems: https://www.techdirt.com/2022/06/29/california-legislators-s...

Even if they did, many parents strongly disagree about what content is age-appropriate. When I was growing up my mom was shocked that other 10yo were allowed to view James Bond films.

So while you might be able to set some baselines, parents still need to do a lot of the legwork to make sure kids are ingesting media that’s aligned with their values. This was true on older media and it’s still true today.

They need their employees to get with the culture and commit more fraud. https://www.bloomberg.com/opinion/articles/2021-01-13/fraud-...

Good riddance. AMP pages loaded noticeably more slowly for me and were typically broken when they finally painted.

Yeah, the OP article is trash but the analysis has valid concerns. It seems to boil down to:

There’s lots of very vague language in the bill that we’re going to need to wait for the agency to actually define in regulation. The article authors would rather the law more clearly define things to keep the agency from overreaching.

This used to be a selling feature of cable. In the not too distant future, all of streaming will have required ads too.

It’s great to hear them admit that there is a problem. For years we’d get “our internal quality measures say we’re doing great. You’re imagining things. Fuck off to Bing and see how you like it.”

I suspect they have an inventory problem, not a ranking problem. Why would any real humans publish publicly on the internet at this point?

If you’re posting for fun you have to deal with moderation and stolen content.

If you’re posting for pay Google just takes your content and shoves it onto the SERP, skipping your ads.

Bill Gates had it right. If you want a sustainable ecosystem, you need to make sure the other players are making more money than you in aggregate. As far as I can tell, that’s only true on the internet for eCommerce and so that’s all that’s left.

What do you want to build? I wanted to train ML models for Kaggle so I picked up Python (even though I disagree with pretty much all the design choices of the language lol)

Considering tech outside of an application context feels weird to me. Like yeah it’s hard to choose with no requirements or constraints.

That said, Rust is probably here to stay. Linux, V8 and some other high profile systems are adding first class support.

Does robots.txt not fulfill this requirement? It’s the same problem as web search. Google builds a derivative work (a mapping from queries to pages) by scraping the web with opt-out consent.

“Our observations in §4.1 to answer Q1 revealed that all SFAs exhibited political biases in the months leading upto the 2020 US elections. Gmail leaned towards the left (Democrats) whereas Outlook and Yahoo leaned towards the right (Republicans). Gmail marked 59.3% more emails from the right candidates as spam com- pared to the left candidates, whereas Outlook and Yahoo marked 20.4% and 14.2% more emails from left candidates as spam com- pared to the right candidates, respectively.”

https://arxiv.org/pdf/2203.16743.pdf

You get what you reward. Large corps reward “increase in scope”. This is hard for people on other teams to evaluate fairly, so most companies end up with “scope = number of people involved”. That’s vaguely equivalent to deploying an O(n^2) algo into prod. Looks good at small sizes but breaks down as everyone talks to everyone else to add collaborators to pad scope.

Pig butchering mostly. “I am going to play golf with my business partners tomorrow. Will you have time to join us Dave?”

https://www.rrstar.com/story/business/2022/04/30/new-pig-but...

Why are you looking for principles when most reactions are driven by emotion? Many of crypto’s loudest representatives are smug assholes. People love seeing arrogant folks taken down a peg (deservedly or no).

With blue bubbles?

This doesn’t work though. I receive enough iMessage spam specifically through Apple ids that I wish I could disable the ability to message me unless you use a phone number.

Is it? You trade retrofit costs later for maintenance costs and feature development friction every day until scale is needed.

The business model, specifically, is a joke. Did they talk to anyone before launch about it?

Dumb question: could some of these attacks be mitigated (or at least made more costly) with transaction size caps? I feel like there’s a monthly drip of stories like this.

Tokyo. It requires building way, way more units in more compact ways than Americans would currently be comfortable with.

Pick your poison. This or https://twitter.com/mattzollerseitz/status/15419943565210132...