Finbarr.ca | @finbarrtimbers on Twitter | https://www.artfintel.com/
I love receiving emails. If you’re on HN, I would enjoy talking to you. Yes, you.
I'm an AI researcher and (small-time) angel investor.
I’ll happily talk to you about your startup, strategize about how to pitch VCs, or give advice on how to get a job in tech. You are not wasting my time.
I had this happen at a different FAANG. I was sent a letter with a 25% raise when I got promoted. I then had a meeting with my assigned HR person & manager suddenly added to my calendar where they told me that I was actually getting a 6% raise.
It was still competitive with market rates, so shit happens. Sucks to be on the bad side of this.
Are L1 vs L2 vs L3 chargers radically different in cost? Or is it that you want to provision charging speed proportional to the power being supplied from solar at the business?
I went to Outside Lands in 2018 and it was well run except for the crowds- lots of crushing, to the point where it was borderline dangerous to be near the front.
What I see happen though is people get downleveled. It’s hard to get senior at FAANG without big tech experience, let alone staff+. So if you spend your career somewhere else and try to switch to FAANG after, say, 10 years, you’ll end up getting down leveled. Now, you’re probably making more money, so maybe this doesn’t matter.
I had a similar, but less dramatic experience. An interviewer took me to lunch and told me that I shouldn’t work at Amazon, as she hated it and was planning to quit.
100%. In big tech, as you get more senior, the job becomes miserable. I wouldn’t want to be a Director, or even a Senior Staff engineer. At a certain point you sacrifice your sanity to be in that many meetings.
you’re not making the right comparison. The right comparison would be within the same org. Is there someone on your team that generates $1M per week? Then they’re 10x you. Or does everyone else generate $10k per week? Then you’re 10x them.
Right, but they _just_ decriminalized drugs, so we haven’t seen the impact. What you’re describing in Vancouver is the result of the existing policies which criminalized drugs.
That’s just not true. The Fed will take into consideration all of the consequences of their actions. They’re not going to do whatever they want and put the country in historical depression.
This is really obvious if you ride bicycles & do your own maintenance. You can quickly identify weird sounds/patterns with problems, because it’s such a simple machine.
I think an end-to-end RL approach will eventually work- but _eventually_ could be in a really long time. It’s also a question of scale: even if Comma’s approach is fundamentally better, how much better is it? If Tesla has 1000x more cars, and their approach is 10x worse, they’ll still improve 100x faster than Comma.
Right, but that’s not a counterexample. The bitter lesson suggests that, eventually, it’ll be difficult to outperform a learning system manually. It doesn’t say that this is always true. DeepBlue _was_ better than all other chess players at the time. But now, AlphaZero is better.
I believe the same is true for neural network architecture search: at some point, learning systems will be better than all humans. Maybe that’s not true today, but I wouldn’t bet on that _always_ being false.
There's always a tradeoff between making it easy for legitimate customers to try out the product, and making it hard to abuse. If you bias for preventing fraud, you have less customers, and vice versa. This was written about a bunch in the context of PayPal in Jimmy Soni's excellent history of PayPal [0], as this is basically why companies like PayPal/Stripe exist: banks optimize for reducing fraud, and not for making it easy to become a customer. So they have very little fraud, but correspondingly few customers (compared to Stripe/PayPal).
I agree with the conclusion. This is totally unsurprising to me as a ML engineer. If you put garbage data into the model, you get garbage predictions. That doesn’t strike me as particularly novel. The same is true for cooking, after all.
However- this has been truly shocking to all of the non-technical stakeholders I’ve worked with. They take the stance that any large amount of data can be used to do ML on, presumably because they don’t know too much about what doing ML is like.
So I’m convinced the author is right, and I’m also convinced that there will be many attempts to use ML on EMRs.
I’m also quite interested in hearing about this. I used to only hear good things about Stripe, but over the last year, there’s been quite a few negative posts. I’d be interested in more details.