cardigan

- SpaceX will land a human on Mars and create some super great marketing material. This will strongly galvanize interest in space. Many ambitious people will be interested in space projects or startups

- deep learning will continue to amaze people in being able to solve problems considered not well suited to it. Some of these applications will seem crazy in retrospect

- folks at OpenAI or Google will get something crazy to happen with a huge amount of compute. It won't feel like AGI but it'll make AGI seem way less insane

- theorem proving with deep learning will start to work

- material science using lots of compute and deep learning will start to work

- deep learning will be applied to fuzzing (finding vulnerabilities in software) and this will be a big thing by the end of the decade

- there will be some large scale multiagent AI projects aiming to learn intelligence through just big simulations of civilization but they will not have interesting results. Definitely happens at OpenAI and possibly elsewhere. They really expect this to work but it won't

- Apple releases an AR headset. Oculus turns into an AR effort instead of VR. The VR wars turn into the AR wars. Lots of money pumped into it. Unclear if it actually becomes the next mass platform, but there's a small chance

- crypto people will find success approaching incentive design problems in more traditional avenues like large organizations and charter cities. Most crypto projects will be dead in the water, including Ethereum, but there will be diehard enthusiasts who stick to it. The money will dry up, forcing others out. Cryptocurrency, like BTC, will still be a big thing on the internet

- teleop robots. Globalization of physical labor starts to happen. It will seem like an emerging trend by the end of the decade

- self driving cars will be seen to be largely a fad, with lots of wasted money

- there's a chance the olivine beach climate change project gets a huge amount of traction

- student loans and for profit colleges in the United States have some kind of reckoning. People stop believing in college: more people all over the world think like Lambda School and the software industry

- senior software engineer salaries in the Bay Area continue to climb

- Silicon Valley stops being so obsessed with China. It's more obvious that Chinese innovation is heavily lagging behind

- defense technology starts to capture the attention of more of Silicon Valley and the innovative class. Anti-defense stance stops being the default. Tech bro patriotism: more people think like Anduril. This leads to some really crazy defense capabilities of the US. E.g, auto targeting killer drones

- Social media usage per person goes down, for high income people

- YC is no longer cool at the end of the decade, but hacker news still is.

- lifestyle software businesses continue to be seen a lot more positively in the industry: more software engineers think like patio11/csallen/levelsio. Starting a small software business becomes more of a viable career path and seen as more responsible and mature than the VC unicorn path.

- meat alternatives grow faster than anyone expected them to

- Tesla is the most valuable car company but still hasn't figured out full self driving

- sci-fi reading and blog post style writing will be a major status symbol in the tech industry

- Bay Area will be less dominant in interesting tech startups than it is today, because of immigration and housing. The next place will be the internet or somewhere open to outsiders like Estonia, not China

- Donald Trump will do something too shocking and it will actually end his career. Society will learn an antibody to his populism, but it might be a long time after he's president. Will happen by the end of the decade though

- deep learning is a very centralizing technology. Even bigger tech companies will be started where the value they create is their machine learning network effect. They might not be as big as 1T by the end of the decade but they'll get there in another 10 years

- more power shifts from government to private enterprise

- tech companies remain underrated and continue to grow and have way bigger market caps

- VR porn ends up driving adoption of the current set of VR hardware. Funny but FB executives won't be happy about this and that'll make them look to AR.

- Facebook social VR with strangers doesn't work. But FB social VR with your real life friends might work and be really popular. This would be the main application of VR, if any : then it'll morph into AR

- religion continues its decline. New internet ideologies continue to proliferate. Some of them will be pretty weird yet have a lot of impact, like the alt-right did this decade

- the penny is abolished in the United States

- US inflation is a lot higher than it has been in the past; US treasury bonds not seen as super safe anymore

- if there's a recession, it won't affect the economy as uniformly as historical recessions (overall hit may still be really big, but higher percentage of people will end up well off). More variety in the economy and people's lives where it's not as correlated

- tech companies funding more and more media/content (like Netflix/Amazon, but also upstarts). Traditional media gets eaten by tech-enabled companies

- everyone worldwide has a lot less sex

- marijuana and psylicobin legalized federally in the US; stigma against drugs on the decline globally (related to decline in religion + rise of internet ideologies)

- more local manufacturing. Specialized manufacturing countries/cities stop making as much sense. Let going to China to make your hardware thing; you'll just do it wherever.

- the US will be more obsessed with Africa than China (may take 25y instead of 10y; caused by population dynamics)

- 10-30% more happens in 2020s than 2010s; accelerating progress but it's not very noticeable yet

- Stripe becomes a gigantic company Because of that, the SaaS economy goes global: microSaaS is the new doctor/lawyer/engineer, especially in India and Africa

Faces... Videos of all kinds of people and places.. Friend networks (via who you follow).. What videos you like showing what you think.. They have a lot of data

It's the nature of economic sanctions that there are real people's livelihoods which are directly affected. That happens with all economic sanctions... nothing unique about this one

Posting Nat Friedman's tweets here so they're easier to read - they're doing more than most companies about the whole thing, not sure where the vitriol in these comments is coming from:

It is painful for me to hear how trade restrictions have hurt people. We have gone to great lengths to do no more than what is required by the law, but of course people are still affected. GitHub is subject to US trade law, just like any company that does business in the US.

To comply with US sanctions, we unfortunately had to implement new restrictions on private repos and paid accounts in Iran, Syria, and Crimea.

Public repos remain available to developers everywhere – open source repos are NOT affected.

The restrictions are based on place of residence and location, not on nationality or heritage. If someone was flagged in error, they can fill out a form to get the restrictions lifted on their account within hours.

Users with restricted private repos can also choose to make them public. Our understanding of the law does not give us the option to give anyone advance notice of restrictions.

We're not doing this because we want to; we're doing it because we have to. GitHub will continue to advocate vigorously with governments around the world for policies that protect software developers and the global open source community.

OP here, just waking up (I'm remote) - I can't edit my original comments so let me modify them here:

I wasn't involved in our communications with Lyft, so I was talking about something I didn't know much about. My audience was just the anonymous commenteriat: turns out a lot of people whose opinion makes a material difference to Lyft/Scale read these comments too. Sorry for not realizing that; I probably wouldn't have posted an uninformed personal opinion had I realized that.

I was being way too aggressive - genuinely sorry to anyone at Lyft who felt maligned by these comments. I woke up to 20 messages from coworkers who told me I was being an ass - genuinely sorry :(

Also I really should've clarified I was not speaking on behalf of the company: this was just a personal, uninformed opinion.

I cannot go into the details I learned about Scale's agreement with Lyft since it's confidential

Also, the viewer packaged with nuScenes was built by Steven Hao from Scale, and while it was packaged as part of nuScenes it should probably be called Scale's viewer instead of nuScenes' viewer. The original viewer in the nuscenes SDK has the Scale logo, but it looks like Lyft removed that in the fork. Maybe a bit of public shaming will fix that...

Dear Lyft marketing person who wrote this: we are a data labeling company, and you may think that means we have a bunch of useless bozos working here like most other data labeling companies, but that's not true - e.g, Steven is one of the smartest people in the world - https://stats.ioinformatics.org/people/3113 - he learns ridiculously quickly - e.g, gets to number one on random video games in a few weeks and learned to boulder L10 in a few months from scratch (normally takes years/decades and most climbers never get there)

(I work at scale)

Hmm this blog post and the website doesn't mention that this dataset was mostly annotated by Scale (scale.ai), as part of a partnership with Lyft ... We're going to publish a blog post about this soon, but if anyone at Lyft is reading this, please figure out how to reasonably credit Scale since I doubt leaving out Scale completely from the announcement is in the spirit of the agreement. Scale should probably also be added to the bibliography and website in some form

Contrast this with the nuScenes website, which was also annotated by Scale, and whose data format set the standard for this dataset: they credit Scale pretty reasonably

The interview goes into how this ties into formal reasoning towards the latter half.

Check out my other comment on this thread. I interviewed these folks and wrote the blog post (got lots of feedback from friends and design help from our awesome designers) - I'm a software engineer, not a PR department, lol :P

I tried to pick the answers which were most well written, not the ones which were most positive/negative. Personally I do think this is representative of the labelers who've stuck with us, but at this point don't have the energy to argue this; hopefully my other comment is convincing (i.e, there isn't a high bar to making people happy when their next best alternative is a lot worse)

Sorry I should've explicitly said this, but I was responding to just this part, not the part about the job being tedious:

  But I do see a lot of data around company engagement. 
  There’s basically no chance this level of positivity is ubiquitous for them

I guess what you're asking is - how many of these people find their job tedious? I will say it's a lot less tedious than you might imagine as a first impression - e.g, if you see something weird in some data, you talk to other people about it; if you get good, you train people; when doing a new project, you're learning from your coworkers; if you get really good, you might be asked to help develop training materials, etc. So there's a lot of interacting with other people, and a sense of community. For me personally, I honestly find it meditative to label a lot of data - it feels kind of like tending to a large garden, maybe even fulfilling some deep OCD/obsessiveness desire. Some of our labelers find it meaningful that they're contributing to robotics / self driving cars - e.g, the last interviewee in the blog post.

Back to your question though - how many of these people find their job tedious? I'm not sure how to ask the question to them in a way which gives a satisfactory answer. e.g, I'd expect if we just asked "do you feel like your work is tedious?" the answer would be dominated by people's realistic alternatives, and wouldn't have much to do with the job itself. (so we'd get similarly positive responses) If you can think of a better way to frame that question, I'm happy to ask it and post the responses here :-)

I'm a software engineer at Scale and the author of this blog post.

Let's talk about Venezuela for a second. ~75% of the population lost >19lb in body weight in a year according to this survey: https://www.upi.com/Top_News/World-News/2017/02/19/Venezuela... It's unbelievable that we haven't figured out how to prevent people from living like that in the 21st century. I think it's totally deplorable and honestly an affront to humanity.

You know what I think the most effective way to combat large scale poverty is? Not by working for an aid agency (we've all heard the horror stories) - instead, how about making those people economically valuable? The internet is an amazing way to reach those people - and guess what - Scale is actually doing that (evidence: these stories). If we continue to grow, we'll be doing that even more.

To be clear, this isn't the main mission of the company - but I don't see how anyone could think it isn't a great side effect. Hence the title of the blog post - positive externalities.

This is so interesting. Because it can't be this hard to do this kind of physics simulation at the correct level of fidelity if you want to apply RL to physical problems

I think you're referring to the SupportAssist Client being an HTTP server - while it is weird that they exposed all those other routes, the driver install route allows for drivers to be installed from a website (which a named pipe would not).

I wouldn't characterize it as "pure laziness" - more a questionable feature

Oh sorry about that - these links should work:

Backend eng: https://jobs.lever.co/scaleai/c1443865-f64e-4467-bfdc-898053... Frontend eng: https://jobs.lever.co/scaleai/9fe1f405-647b-493c-9728-e2c387...

Scale | Backend/Full Stack and Frontend and ML | SF or Remote

We label data for your favorite computer vision teams. Our mission is to accelerate the development of AI applications - we believe building a high quality labelled dataset is the biggest bottleneck to deploying supervised deep learning systems, so that's what we're tackling first.

We've had phenomenal breakout revenue, raised an $18 MM series B, and are looking to grow our team of 55.

We're looking for engineers to work on projects ranging from making labelling more efficient via front-end work/ML work to launching new product lines demanded by our existing customer base.

If you are interested, please apply here: Backend: https://jobs.lever.co/scaleapi/c1443865-f64e-4467-bfdc-89805... Frontend: https://jobs.lever.co/scaleapi/9fe1f405-647b-493c-9728-e2c38...

Scale | Backend/Full Stack and Frontend | SF or Remote

We label data for your favorite computer vision teams. Our mission is to accelerate the development of AI applications - we believe building a high quality labelled dataset is the biggest bottleneck to deploying supervised deep learning systems, so that's what we're tackling first.

We’ve had phenomenal breakout revenue, raised an $18 MM series B, and are looking to grow our team of 40.

We’re looking for engineers to work on projects ranging from making labelling more efficient via front-end work/ML work to launching completely new product lines demanded by our existing customer base.

If you are interested, please apply here: Backend: https://jobs.lever.co/scaleapi/c1443865-f64e-4467-bfdc-89805... Frontend: https://jobs.lever.co/scaleapi/9fe1f405-647b-493c-9728-e2c38...

An open source implementation which anyone could roll out would be pretty cool :-)

CoinHive provides two miners: one which runs without asking the user, and one which requires explicit consent from the user.

The version requiring explicit consent is being used on the Salon website, and isn't currently being blocked by AdBlockers/malware detectors.

An upper bound to the amount of information you need to transmit to have something built the way you want it built is how much you need to tell a good developer about what you want made before it is made to a satisfactory degree.

And an upper bound to how long it can theoretically take to build something is the same as the time to transmit that information for a human plus some small epsilon: computers can run very fast.

So getting a computer to do what you want ideally should not be close to as slow and difficult as it is today. This is sort of the true goal, rather than formally specifying what you wish the computer would do. If you believe the goal is formally specifying what the computer does on top of bulletproof abstractions, we're already very far away from that - programmers today don't tell a CPU how to do branch prediction, or a compiler how to optimize their code, for instance.

Probably a really good system could do even better than the upper bound above - what if you knew what someone wanted before they could even describe it to you, and you built that for them?

A better baseline than an out of the box OCR engine is applying some computer vision techniques before using an out of the box OCR engine. This can significantly out perform a pure OCR engine approach. (speaking from experience working on startups doing this kind of stuff for ~10 months)

One dumb but surprisingly effective thing to try is apply a bunch of random binarizing filters before putting your text into an OCR engine, then picking the most common output from those filtered images. Some combination of morphological operations (dilates/erodes in different orders and different strengths), different levels of blur and sharpening, adaptive binary threshold at different levels, resizing the image (hilarious but some OCR engines are sensitive to relative scale), adding white borders of different thicknesses, rotating by small angles, denoising, ...

This gave us ~95% word level accuracy on our dataset (with tesseract as the OCR engine), without tuning the filters (we had used the filters individually before with some brittle logic for deciding when to use them: the random filter approach was kind of a joke which ended up working). Tesseract on its own had an abysmal ~40% word level accuracy on that same dataset. Dataset was of words segmented out of scanned bank statements.

Main downside was that this was pretty slow (we were generating ~10k filters per word), but we found a way to make it work (hierarchical stacked AWS Lambdas, lol).

VR systems today are not used regularly by most people who own them, including even most people I know developing VR full time. Price point and being untethered doesn't change that.

I'd guess the biggest hurdles are physical discomfort with long term use, and that there isn't yet very compelling VR software.

Hmm, but what if the living creature had a density of neural circuitry similar to ours, and mainly interacted with things inside itself, and had slow propagation of knowledge? Not sure I understand this except under the hidden assumption of having a similar number of neural circuit elements

Hmm, a model could also give you multiple very different designs to help you see more possibilities

Because this attack is not feasible against MD5 with current state of the art cryptanalysis.

Instead collisions between files are demonstrated by appending bytes to two files until their hashes match.

Fun fact: those plasticky end parts are called aglets

Source: https://www.youtube.com/watch?v=Evcsj1gx1CE (I didn't forget it)

This is really cool. Do you think it will be possible to run MongoDB on an FPGA anytime soon?

Doomsday prepper websites

The channel does not need to be secure, only authenticated. So e.g, you could send them a Slack message then have them call you to confirm you just sent that link and not someone who hacked your Slack account.

It's fine to have someone snoop and see the link, as long as they can't change the link in transit.

Sorry that's not the truth, I read about a book a week. This week I've read 3 and started another 2. The secret is to read for enjoyment, a lot, and eventually you get fast. If it feels like a chore, you're not reading the right books: read something else. If you're enjoying it, it's easy to get carried away and read for many hours at a time.

Sorry I don't understand your question. Obviously very few people living in Kashmir support those groups, see polls.