It could also be interesting to define multiple subsets of sneakers - "2021 avant garde sneakers", "skate-inspired shoes", "2010s most popular shoes in the NBA" (based on https://ballershoesdb.com/ or a similar database). It could be cool to start with a model trained on a subset of sneakers, and tweak/combine things from there.
Because it's a university senior design project - they often pair students with related majors together to work on a project over the course of a semester or two. It's about the interesting engineering of the entire project. They never claimed any one part of it was "novel" but the project as a whole clearly is.
The evidence is literally in the original article: "Darker-skinned women were the most misclassified group, with error rates of up to 34.7%. By contrast, the maximum error rate for lighter-skinned males was less than 1%"
Regardless of whether the higher error rate is a combination of race, gender, or both, it's still a huge issue. Granted, that study was from a year ago, and other companies have since improved their facial recognition systems. But an overall precision/accuracy/f1 score doesn't mean much when accuracy varies that much by group. Sure, you can market it as "accurate on white males", but you can't market it as "accurate"
That's a great point. One of my favorite parts of college was small research-based classes with 10-15 undergrads/grad students. The level of personal interaction and feedback from professors was the most valuable part.
That being said, I was fortunate enough to have similar experiences re: research at the state school as well (a professor who went out of his way to mentor me, leading to a few publications). It was definitely harder to find (because of how many other students professors at big state schools have to deal with), but it was more a function of professors' time than their willingness.
I had a unique opportunity of taking a year of classes at a good state school (in the top 100 for most STEM fields), and then went to an "elite" school for undergrad (taking a mix of CS/non-CS courses at each).
You can absolutely find motivated peers and professors at state schools, particularly if you look at places like the honors college. If you want to land a good job, go to a good grad school, etc (which is the goal for most people), all of that is 100% possible from a state school.
However, if you want to do something non-traditional (startups, research abroad, "choose your own adventure" style careers), I found that the "elite" school offered a lot more opportunity. Part of that is due to funding/size, where top schools have more money to allocate to students who ask for it. Students at elite schools have more support on average (from family as well as the school itself), so they tend to have more opportunities open, earlier.
It's like the difference between a big city and a small town. Plenty of people are successful without living in a huge city, so it's definitely possible. You could argue that the average drive of folks in a big city is higher, but a lot of that has to do with resources and opportunities as well.
The top students at a good state school would absolutely fit in at an elite school. The only difference I saw was in the support/resources they had before coming to college.
Thanks for taking the time to reply - this story is a good reminder to treat employees/all people as well as you can, not conditionally.
Do you think there's anything the CEO could have said/done today that would make you reconsider Instacart? I do like the retroactive payments he mentioned, but wonder if the there's anything more they could do to regain a customer like you.
This is really cool, some great logo redesign inspiration.
A couple of thoughts:
1) If you make the logo designs page shareable (step 2, https://mybrandnewlogo.com/logo-designs), I would love to get feedback from a few people. If you don't want to store things persistently, even storing the input params (name, color, slogan, etc) in the URL would be great.
2) I love that the name can be read as my "brand new" logo, or as my brand, new logo.
Great read. Points #5 (on being able to use a longer time horizon when investing in ideas to let them grow) and #6 (on the growing importance of consumers) stood out to me as well.
It's essentially the build a little/sell/build a lot cycle you mentioned, but you want to minimize the first step as much as possible. As tech folks, it's easy for us to overestimate the importance of technology to our end customers. They're not buying code, they're buying a solution that will make their life easier/better.
"Sales" is a long term process, but the first meeting usually starts with understanding the customer's problem. If you build something before that (and demo it at the beginning of the meeting), you can bias the process with your assumptions. Instead, if you try to get inside the mind of your customer, you can get closer to the correct solution.
For a lot of those first sales meetings, you don't necessarily need technology to demo. Instead of showing/telling your customer, you're starting by listening. Then, you can build a small demo of something interesting before the next meeting (often, this is more to keep the conversation going and explore new ideas than to immediately solve the customer's entire problem).
I agree with the other comments mentioning an initial (days/weeks long) MVP build cycle. However, over the past few years, I've found it helpful to adopt this mindset: "my idea for the best possible product is at least slightly wrong, and the more code I write the harder it is to fix". There's a psychological concept of functional fixedness, where you view an object only as its traditional use. I think a similar concept exists in technology. Particularly as a founders-only company, the technology you build has some inertia, and impacts your mental models of the problem you are trying to solve. The more code you write, the higher the "cost" of changing (both in your mental models, and the technology itself).
All this really means is you're right about the build a little/sell/build more cycle - but what you build is less important than what the customer says. The first build is often just a means to start an interesting conversation about the customer's true problem, and you might need less technology for that than you initially assume.
Getting a replacement car key made is a surprisingly slow/expensive process for most modern cars (not as easy as a replacement house key at Home Depot, for example)
This is awesome. It would be cool to adapt the same idea to images which contain numbers (house numbers, highway exit signs, etc) like "1245" to represent 12:45. Since you've identified the number in the image, you could even blur the background slightly and add contrast between the image portion and background (since the numbers may only be a small part of the image).
Not very efficient, but a cool way of combining screensavers and clocks. Not sure how many "aesthetic" screensaver-quality images contain numbers though
The difference between the "time" and "non-time" font styling is more subtle here - when I changed #main_text to have color: gray, it became a bit easier for me to see the time.
Not only do you both get muddy, but the pig wins with the benefit of experience