Chisel and this project are different; chisel is a programming language to design hardware while this is a build tool for hardware design.
In software we have a nice compilation process that transforms code into machine code. However, to "compile" an ASIC you go from a hardware description language like verilog (which basically describes the design) and pass it through a complicated pipeline that composes a number of different tools. Right now, engineers use ad-hoc flows that use bash/TCL to glue together all the parts --- the project posted above is an attempt to cleanly specify + control the "compilation" process.
Not sure if I follow your JFT argument, but there's a large body of work on both (a) studying whether chasing ImageNet accuracy yields models that generalize well to out of distribution data [1, 2, 3] and (b) contextualizing progress on ImageNet (i.e., what does high accuracy on ImageNet really mean?) [4, 5, 6].
For (a), maybe surprisingly the answer is mostly yes! Better ImageNet accuracy generally corresponds to better out of distribution accuracy. For (b), it turns out that the ImageNet dataset is full of contradictions---many images have multiple ImageNet-relevant objects, and often are ambiguously or mis-labeled, etc---so it's hard to disentangle progress in identifying objects vs. models overfitting to the quirks of the benchmark.
> I've seen TikTok videos go viral being really critical of the US government including its foreign policy. The kind of video that would just be invisible or outright banned from Youtube or Facebook video.
Do you have any evidence of this happening? I've never heard of pro-US censorship before on any of these platforms (or any social media platform..)
I'm a computer science PhD student, and I mostly maintain/create open source projects (https://github.com/lengstrom/) in the evenings for fun --- its particularly great you're working with friends!
I've never viewed it as a potential full time job. However, I recently got GitHub sponsors and make about 5$ a month!
Hi HN, author here! As students facing online instruction, we built ZoomerBackgrounds to solve firsthand the everyday problems of having subpar backgrounds and spoke with several friends working in industry about their background needs. The result is a product we really think you're really going to love. We had a lot of fun making this over the weekend and hope you all enjoy :)
For the tech stack we used react on the front end with a Node.js server + Firebase for persistent storage. This was my first time writing Javascript in about 5-6 years and it's really neat to see how the ecosystem has matured - there are even real classes now! We also used the Google Cloud Vision API for detecting inappropriate images.
On the database side, we found that Firebase Store works really well until it doesn't -- when you want to do anything that isn't basic reads/writes (e.g. SQL-like IN queries, query pagination, programmatic queries like sorting by the HN ranking algorithm ~ points/(time since post)^(1.5)) the workarounds can get complex and a little weird.
One question we had is around AdSense approval! The Google Adsense approval team rejected our website for not having enough meaningful content / not having "valuable inventory" - it would be great to hear some insight about how to show Google we are a real website.
We took a look at adversarial examples for linear classifiers (and in general, we looked at properties that adversarial training induces) here: https://arxiv.org/abs/1805.12152 For $\ell_\infty$ adversarial examples on linear classifiers we found that adversarial training forces a tradeoff between the $\ell_1$ norm of the weights (which is directly associated with adversarial accuracy) and accuracy.
It looks like this article works through something vaguely similar for $\ell_2$ adversarial examples. It would be interesting to compare the author's approach with explicit adversarial training.
Whoa, I made Hextris (https://github.com/hextris/hextris, one of the games removed from the store) a few years ago! Is there any precedent in OSS developers being held responsible for misuse of their code?
Another author here - adversarial training is not sufficient to protect against white-box attacks but it seems to be the best method we have so far (https://arxiv.org/abs/1706.06083).
It seems that all current state of the art architectures are vulnerable to adversarial examples; to the best of my knowledge there are no image classification networks that researchers have failed to reliably produce adversarial examples for.
I hadn't seen that, thanks for the link - it looks useful! Right now the kernel size is not divisible by the step size - I'll try changing the transformation network.
Currently not significant (few dollars a month), but it used to be a few hundred a month when it initially became popular. Monetization could have been much better if we had put in interstitial ads, etc.
Someone posted my game (http://hextris.io/) on Hacker News - from there it spread to a few popular outlets (major tech news companies, subreddits, obscure but popular blogs), all organically. Initially I tried submitting the game to game journalists / iOS app review websites, none of which responded to me - wouldn't recommend that route unless you have ins somewhere.
Cryptographers are computer scientists/developers, though. If you want to do something "real," other than make pretty buttons, that is, you're going to have to do some math.
With 0% interest loans (lots are available for students), 10k tuition + 10k a year in living sounds pretty reasonable considering that college graduates make about 98% more per hour [1] on average than those with only high school diplomas. Not to mention that if you're doing something in CS, you could pretty easily pay off each year by doing extra work on the side / through summer internships.
In software we have a nice compilation process that transforms code into machine code. However, to "compile" an ASIC you go from a hardware description language like verilog (which basically describes the design) and pass it through a complicated pipeline that composes a number of different tools. Right now, engineers use ad-hoc flows that use bash/TCL to glue together all the parts --- the project posted above is an attempt to cleanly specify + control the "compilation" process.