epberry

First, the negatives. This space is enormously difficult. You have low margin, low technology clients with physical infrastructure. You have to get approval from operations, IT, and legal. Once you're up and running the data becomes stale fast and you have to produce insights like move this shelf here. Then you have to wait for them to do that, if they ever do, and demonstrate that the client is now seeing a ROI from that change. And prove it was you and not the new ad campaign they just ran or the better staff they have in place now that is making them slightly more money. I've been working in this area for years, also with a computer vision research background, also read Prayas's post several years ago and dismissed it, also started with specialty retail and focused on ease of install, also dismissed RetailNext for using their own hardware, etc.

Phew. Now, the positives! I do believe that someone will crack this at some point. There are several promising trends including a massive jump in the share of IP camera installs, massive drop in the price/performance ratio of GPUs, and steady increase in commercial broadband. And everybody loves data. Those trends, plus a pivot to optimizing physical business processes, are why I'm still working on camera software at my company, Perceive (www.perceiveinc.com). We got an interview at YC several years ago for this idea but were rejected.

So Jonathon and team, despite how this post started, I really do wish you guys the best. Your tech looks really good. I would love to talk.

I can't believe this is the only post with Surely You're Joking Mr. Feynman! Such a fun, creative book. Parts that stuck with me: lock picking, finding safety hazards at Oak Ridge, playing drums, working intensely.

Also curious about this.

I found this to be a great read on what I think the parent is referring to: https://lilianweng.github.io/lil-log/2018/06/24/attention-at...

I agree with this. I think people are looking at the wrong layer for automation. Frontline workers will exist for a long time because of their flexibility and dexterity. I think automation will largely impact one level up - at the supervisor/manager level. My bet is on cameras and computer vision to be the enabling technology for this.

Love seeing so much Patrick Collison content (should I say PC content?) lately. His Tim Ferriss podcast was thought provoking. I think that's what's important - he has new ideas and he's talking about them.

If there were safety advantages with cooperative driving and ultimately efficiencies I bet a public relations campaign could convince people. Or having dedicated lanes for cooperative driving. I would certainly turn this on for interstate driving.

> Lift code from GitHub to restrict delta computation so that an object that exists in one fork is not made into a delta against another object that does not appear in the same forked repository.

Must be pretty great to be a Github engineer that contributed this to the product to see it upstreamed.

The industrial base, and the fact that production could continue unabated by bombings on the American continent for years, was a deciding factor in the war. As were the Russians repelling Hitler. My historical perspective on this has shifted massively since first learning about the war in school. It wasn't D-Day or the Battle of Britain or Midway. All of those battles were key wins but the war overall was really organized and executed by tens of thousands of typists and mid-level managers.

Nightmares.

I’m not sure using WWI era factory data is the best way to support the point, but I agree generally. For me, decision making ability is vastly reduced if I’m tired or burned out.

I agree. An international competition over “who can do the best research” is certainly something to welcome.

I think the focus on regional effects is a good one because people relate to them much more easily. Today there is also a great article in the NYT on climate change in Yellowstone. It talks about snowpack, drought, and invasive species of weeds and bugs that are more prevalent because of shorter warmer winters. That kind of specificity hit home for me living in the Midwest. In the past few years our ash trees were ravaged by the Emerald Ashbore and just this year a horribly difficult weed, the yellow nutsedge, began spreading. Both of these are new in our area because of warmer temperatures.

Agreed. FAANG has the government over a barrel.

Single player mode ftw. I'm really excited about this app.

File systems suck anyway. They’re useful for programmers and software but pretty awful for people organizing data. My exclusive approach for finding things these days is Spotlight on Mac or google drive search. Drive search is so good that it’s not even necessary to name documents correctly or put the right content in them. Sometimes it seems like the search is context aware and you can type ideas into the search bar and get a list of documents that might be relevant to them. I think of my document memory more as a hashmap than a tree and typing in part of the file and going straight to it is extremely satisfying.

I really like Viterbi. It has applications all across computer vision and filtering and it takes full advantage of the magic of dynamic programming.

Agreed. There's a lot to like here.

In my experience "do both" on any big decision is a recipe for disaster. Very hard to avoid tho. I've known this and still pushed for it just because choosing one of two hard options is itself so hard.

My tmux conf: https://github.com/EverettBerry/dotfiles/blob/master/ubuntu/...

I especially like prefix + & to kill the whole window, prefix + , to rename window, and prefix + ":break-pane" to split the current pane into a new window.

I have found Rekognition to be quite good. Maybe the most interesting machine learning service result for me from the past year was IBM being much better than Google at speech to text. Specifically they offered some key features, like speaker attribution, that made their offering standout. I also found their accuracy rates to be very good.

When I first started exploring these apis I just assumed Google would be amazing but that is _not the case at all_. I suspect they save their best stuff for their own products and the api solutions always lag a little behind. IBM may be better because their apis are such a core offering. Microsoft's stuff is a clear afterthought and the only Amazon service I've had success with is Rekognition. That said, Transcribe just launched so it may/will improve with time.

Beautiful and profound. Not all decades in life are created equal. Don't waste a moment!

I basically only use Dropbox because it is supported on Linux. Also its syncing technology is sooo much better than Google Drive you would think Drive was built by interns. However, they are taking their sweet time with newer features like Smart Sync. It's disappointing because I'm paying for these features and yet they're not supported on one of the platforms my whole company uses. All in all I really hope Dropbox doesn't keep chipping away at Linux - I fought hard internally to use it and I don't want to be proven wrong.

I just came here to submit this. Really cool paper! The counting thing has been devilishly hard for neural networks.

Well, this person "missed computer science" and picked up SICP to scratch that itch. Then he did the examples in clojure instead of scheme! It would seem to me that his colleague's signaling was correct. Would your average frontend engineer engage in either of those two activities?

Exactly. This jumped out immediately to me.

Some of his more famous paintings like Starry Night and Wheatfield with Cypresses aren't here. I guess I'll have to settle for Wheatfield with Crows! The sketches are fantastic tho - you have to scroll quite a ways to get to them but they are worth it.

Good for them. I couldn't be happier with the XPS 13 Developer Edition I just bought. Even though I had to wait for a firmware update a few weeks ago for the webcam to work :|. But the best part of that? It was a matter of 'apt-get dist-upgrade'.

Yep. Subscribe to the “morning paper” too. Nothing like sipping coffee, reading about generative adversarial networks, then going to work and building web apps all day.

No discussion at all of the actual content of what he was saying. In my experience he was right on the money. Problems just keep coming at you and you gotta deal. The only way to truly be successful and be able to look back and relax is to leave the business. Seems most founders aren't built that way though.