I've mostly used it for reading textbooks or academic papers (and taking notes on them) and it works fine for me. It can be a little small but you can zoom in if you need
Surge AI (https://surgehq.ai) | Software Engineers (full-stack or applied ML) | Full-time | SF or Remote
Our mission at Surge is to build the human infrastructure powering the next wave of AI and LLMs. We’re building a data platform that powers AI teams at OpenAI, Anthropic, Meta, Google, and more. Reinforcement Learning with Human Feedback is the critical technique behind the new generation of AI assistants, and that human feedback comes from us. Our product has been a game-changer for the top AI teams in the world. Here are some examples of our past work:
You’d be joining a small, rapidly growing team of former engineering and ML leaders from Google, Meta, Twitter, and Airbnb. We ship quickly, deeply care about the problems we are trying to solve, and value autonomy and ownership. No previous AI experience is required, if you have the engineering skills you can learn what you need on the job.
We're looking for engineers with a few years of experience and you can work out of our office in SF or remotely. Please reach out to us at [email protected] with a resume and 2-3 sentences describing your interest. Excited to hear from you!
It's likely because GPU calculations are non-deterministic and small differences in floating point numbers could lead to different outcomes (either in the way you described or somewhere deeper in the model)
That's not how the First Amendment works. Cornell University is a private organization and it can choose not to do business with people if it wants to, provided that it's not breaking other laws by doing so. Students get suspended or expelled for speech all the time. (I would guess public universities can too - the First Amendment says "Congress shall make no law..." - not sure where its boundaries are drawn though.)
Many of my friends who are tech workers who've worked at unicorns and tech giants for 5-10 years bought their first homes in SF during the pandemic. I would guess part of the drop in rents is a shift in demand from former renters to new homeowners. Also, there was a general increase in prices across all asset classes during the pandemic that continued to drive prices up. Finally, property prices have actually gone up less in SF than they have in the US as a whole, likely because of the general effect of people moving away - compare the change from January 2020 to today in this chart https://fred.stlouisfed.org/series/MSPUS vs this chart https://www.zillow.com/san-francisco-ca/home-values/
Rents are definitely down - or at least they were 1.5 years ago when I last moved. Before the pandemic, 1 bedrooms were usually at least $3500 and often $4000+. When I looked last January, there were decent places between $2500 and $3000. I emailed some landlords saying "is there any chance you could go lower" and they offered me free rent for 1-3 months. I ended up with a pretty good deal on a large one bedroom in the Mission.
Do you have specific mathematical topics you're interested in?
I have tried reading textbooks a few times to teach myself but found it hard to stay motivated, so I found a tutor on Upwork to assign and grade homework problems and answer my questions. Along with math textbooks and YouTube videos, this has been super helpful for fleshing out my knowledge of college-level math that I never learned properly. It's also great because they can go at the pace you want and focus on topics you find challenging, interesting, or useful.
I like yoga but I don't feel that it's a super social activity. Maybe I just go to the wrong studios. Do you find that you talk a lot with other attendees? Or do you think the yoga itself is social?
Thanks for posting this! I found myself feeling quite lonely at the end of a 5-year relationship a few years ago and have spent a lot of time in the past few years trying to work through loneliness. I can't compare it to your situation - divorce with children must be really difficult - but I think male loneliness is a super important problem in our society and unfortunately it's something I had to put in a lot of work to address.
> The thing is I don't even like hanging out with people most of the time but I get into this space where it feels like a biological necessity.
Connecting with people is a biological necessity! It takes work but it's totally worth it.
I definitely agree with others that therapy is a good place to start. Also, it's okay to feel lonely. It's also okay to want to connect with people and to put work into it.
A few things that have worked for me:
- Just because older friendships have withered doesn't mean they can't be revived. Reach out to people to catch up - everyone loves feeling wanted every once in a while. Most of these won't turn into close friendships, but some might. And even if they don't, seeing old friends will make you feel more connected to more people and make life more colorful. The world is a lot more interesting when you hear more people's perspectives, I think.
- Join a club. Or join multiple clubs, even. A lot of people are looking for social outlets as pandemic restrictions recede. I like running so I joined a running club, but I'm sure you can find other clubs to join. There will probably be a lot of new members like yourself wanting to socialize - my running club is 20 years old but most of the active members joined since June 2021.
- You might consider finding roommates. Living alone is nice in some ways, but having roommates gives you built-in social interaction every day (as long as they're people you can tolerate). IMO more than one roommate is best.
- Make friends at work. If people don't eat lunch together, try to invite people to lunch. If they don't hang out after work, try proposing going to a bar one day. This might be higher effort than other ideas here, but I usually find that having casual work conversations fills at least a little of my social needs each days. (Obviously this doesn't apply if you work remotely.)
- Be vulnerable. You might be worried about others judging you for feeling lonely, but I don't think they will. You can express it to them in less direct terms, too - maybe "I feel like I've lost touch with my friends and I wanted to see how you're doing" or "I've been trying to rekindle friendships since the pandemic/the divorce." I had a conversation with a high school friend just a few weeks ago just like this, in fact. I think being vulnerable about things in my life (including but not limited to loneliness) helps deepen the relationships I have and makes my social interactions fill more of my emotional needs too.
For me, a mix of all these things have helped. I still feel lonely sometimes, but I feel a lot better than a few years ago. It's a lot of work to build friendships and a social structure, but it's well worth it for your happiness (and it'll make your new friends' lives better too).
You can just do that by adding large hidden layers or more hidden layers, up to a point. But eventually the signal that's coming from the input data is so diluted through all the neurons and layers that your models stop performing better as you add more neurons. Many of the advances in NNs come from structuring the neurons in particular ways - for example, in computer vision, convolutional neural networks. These are kind of a small neural network that looks at each part of the image and gives an output - basically, shrinking and summarizing the image - which is then fed into another small neural network that shrinks & summarizes some more, and so on until you have only one result, like "is this a cat" or something like that. Transformers, which are mostly used for natural language processing, have small neural network layers that let the network figure out which other words in a text are relevant for understanding a given word (or, token, really). These help simplify the problem for the NN so it doesn't have to sort through all the data it has at once, which is a bottleneck on scaling NNs.
Not sure that explains it like you're 5, but hopefully it addresses your question
It's kind of funny watching SGD slowly get closer and closer to you, and then overshoot. I was surprised that it's so much slower than me at learning this task. But then again, my brain is probably devoting a lot more computing power to solving this problem than it is so it's not really a fair fight
> This gets to the other mistake Diehl makes in that article, which, ironically, echoes a similar mistake made by many crypto absolutists: there is no reason why the Metaverse, or any web application for that matter, will be built on the blockchain. Why would you use the world’s slowest database when a centralized one is far more scalable and performant? It is not as if WhatsApp or Signal are built on top of the plain old telephone service; they simply leverage the fact that phone numbers are unique and thus suitable as identifiers. This is the type of role blockchains will fill: provide uniqueness and portability where necessary, in a way that makes it possible to not just live your life entirely online, but as many lives simultaneously as your might wish, locked in nowhere.
Halfway through this paragraph I thought Ben was arguing that phone numbers are a centralized service that shows that blockchain isn't necessary. Instead, he's arguing that blockchain will create "unique portable identities." This is an old vision that many companies have been pushing for a long time - "log in with Facebook" "sign in with Apple" etc. Today, we've pretty much settled on phone numbers and email addresses as a unique, portable identifier (plus accounts from big tech companies). You can even easily create new email addresses if you want multiple identities. Why do we need the blockchain to solve this problem either?
Another example of the same thing: he says that it doesn't make sense to use LLCs, which are built for the physical world, in the digital world. He says that the digital world can use DAOs. But, of course, "Stratechery" isn't a DAO - it's an LLC. Meta is an LLC. LLCs are already non-physical, even if they're not purely digital. That's precisely why LLCs were invented - to decouple capital from physical ownership. Why do we need DAOs when we already have LLCs? Is it just that you don't need a lawyer to make a DAO?
Ben spends a lot of time in this article arguing that blockchain can be better than centralized services for some purposes - but then fails to argue what those purposes are, in my opinion.
To me, conflating the metaverse/VR - which I think is overhyped but has some obvious value - with blockchain - which is a lot harder for me to understand - is a mistake.
I agree with you that it's definitely about who you know - if you go to Stanford, you can get better jobs out of college compared to someone with similar skills who didn't. But I worked at a FAANG (or whatever it's called now) company and many of the senior engineers we hired were from small/less prestigious companies. These were people with ~10 years of experience who had become great software engineers but had never considered themselves in a class to work in big tech. Then, when given a chance, they got an offer and often accepted it because it paid so much better than their old job. If there weren't such high demand for software engineers, the company would just have hired from other big tech companies since it's safer, but they were forced to try to find people in unexpected places.
On the other hand, these people were still negatively impacted by their backgrounds: they hadn't had the chance to join fancy tech companies out of college, which meant they missed out on several years of career opportunities. They were more likely to be hired at a Senior Software Engineer level, for example, when someone similar from a big tech company might be hired at Staff+.
As much as I love hip hop and electronica, and as difficult as it is to make excellent music in those genres, it really is very easy to make okay music. Have you ever used FL Studio (formerly known as FruityLoops)? I remember downloading it freshman year of high school and being able to make a catchy-but-derivative hip-hop beat in minutes. That's not enough to make a hit song, but I think luck is a bigger factor than effort for a lot of underground hits.
Right - I think generally having a variety of energy sources is also good so that we can have a smoother power generation curve during the day rather than the big peaks that wind & solar generate. Eventually, we'll (hopefully) have enough battery or other energy storage capacity to smooth out bumps in energy production so it can be used throughout the day. But even then it'll be useful to have wind power generating some power when, say, it's nighttime.
I don't think this article successfully describes "why" SF is dysfunctional. I agree that the government is bureaucratic and slow, and the school system is terrible, and our government is corrupt. That's the "what." I think some of the "why"s mentioned in the article are accurate. It's true that SF residents tend to be transient and only homeowners vote in large numbers. And SF political actors see themselves through a national prism (because SF politicians tend to go on to higher office - like the Senate, Governorship, or the Vice Presidency) that encourages them to focus on relatively less important policies like banning vaping.
But I think saying no mayor hasn't been re-elected in 20 years isn't remotely unique to SF. Guiliani was re-elected, and so was Bloomberg, and so was the widely despised de Blasio. Politicians generally get re-elected everywhere. Further, it's misleading to say that Democrats always win. SF elections are often hotly contested between the dominant political factions of "moderate" and "progressive" (although those labels are always changing meaning), even if it's Democrat vs Democrat. Some candidates are law and order, and some are anti-capitalist. There is plenty of political competition in SF - which brings me to my next point.
I think it's wishful thinking to say that a "Bloomberg-type figure" would solve all these problems. The mayor doesn't have the power to solve these problems. In fact, that's one of the (many) root causes of SF's problems - in order to make progress on all the problems listed in the article, you'd need to pass ballot propositions, change most of the Board of Supervisors, elect a different school board, elect a new mayor, and change some state laws like CEQA too. The real problem is that power is too diffuse, which encourages corruption by encouraging everyone to demand concessions before any reform can be made. It also means that candidates have to run promising things they can't deliver, or things they can deliver that don't solve the city's biggest problems. That, in turn, trains the public to expect even less of their representatives.
I think another factor is that NYC's government controls much more of its metro area than SF's government does. If the Bay Area had a governmental body that covered the whole region, they'd be able to pass broader laws that, for example, funded transit across city borders. Today, most laws like that have to be passed at the state level instead, and the state is split between many metro areas so it doesn't bother to address Bay Area-specific problems.
Will we ever run out of CAPTCHAs that computer vision can't solve? When reCAPTCHA was started, it used difficult-to-read letters to digitize books, but that's long gone at this point. It seems like Google's been feeding us all pictures of street features to identify for years. Someday adversarial models will catch up and be able to pass these CAPTCHAs (or maybe all CAPTCHAs) right? What then?
> Why would a company pay more than they need to? I would pay 4x the price for grapes if required. I don't as I would prefer to spend less on grapes.
I think it's because they could still use more engineers to do what they want to do, and they believe that great talent is way, way better than good talent. They want pretty much everyone to think "working at FAANG is the best financial option for me" and to apply, so they get their pick of the talent pool.