mkasu

I'm not in the US, but as an immigrant myself I can say it's incredibly stressful to be on a working visa during job changes and during a period of unstable economy. Especially if you have family.

To be rational, I work in academia and I shouldn't need to worry too much about job safety and/or visa applications, but I still semi-regularly woke up at night from nightmares of my visa expiring, not being renewed, suddenly being deported, or similar stuff.

I got my permanent residence earlier this year and all of it stopped. It gives you some sense of stability/security. It also makes one feel a bit more accepted in society by not needing to leave it if a work agreement (for whatever reason) ends.

> if they trained the model on me visualizing a bear, a fish, and a bird, and then the neural net still outputs "horse" when I visualize a horse

Well, the failure cases figure says it does not work if "training and testing datasets do not overlap". So, it'd just find the closest trained class and then generate new image from that class (i.e., in your example the bear might look more similar to horse than a fish or bird, so it'll generate a random bear).

Yeah, I thought the same after seeing this. It's kind of a fun use-case of Diffusion models in this context, but as a scientific paper it seems too overselling. Well, it surely is the kind of clickbaity content to anticipate lots of retweets from.

I only skimmed the paper, but from what I understood, it is essentially a diffusion model pre-trained on a handful classes. The brain information is then used to largely "pick which class to generate a random image from".

The paper itself even picked the "better" examples. The supplemental materials show many more results, and many of them are just that, a randomly generated image of the same object class the person was seeing (or, the closest object class available in the training data).

"Reconstruct" seems a pretty bad word choice. I think the results are presented in a way vastly overselling what they actually do. But that's a worrisome trend in most of AI research recently, unfortunately.

(I have a PhD in a field of Applied Machine Learning. I work at a university in Computer Vision.)

Well to be fair, three of the four apps are Apple ecosystem apps (although Craft now expanding to other platforms).

Maybe Roam, Obsidian, Logseq would be better examples of booming apps note-takers jump ship to? But then, I think all of these apps are rather niche compared to Evernote.

> IMO Perfomance wise CloudFront and Cloudfront are quite similar.

Yes I don't think the capitalization has too much impact.

My portfolio website is 4-lingual and the initial setup was pretty confusing (a bit trial and error until I understood how it works). But now it works very conveniently.

Hugo has a multilingual mode[1] which I am using on two websites. It's a bit confusing at first but works pretty well for me.

For content, you basically put blogarticle.en.md and blogarticle.fr.md next to each other it will treat it as the same article in different languages. Per default it puts different language websites into different subdirectories (with a default language at root level) but it also supports multi hosts setups where you can define different domains for each language (e.g. blog.de and blog.fr). There's support for generating "this article in other languages" type menus, which allows you to crosslink the website across languages. There's also i18n support for translating stuff in templates etc., where you provide a dictionary file with translated versions of each string, and it'll replace them based on the language of the current article (useful for stuff like navigational menus etc.).

1: https://gohugo.io/content-management/multilingual/

There's a paper which does something similar including an analysis of different statistical models on top of a CIELAB color space.

Estimating Color-Concept Associations from Image Statistics. Ragini Rathore et al., IEEE VIS InfoVis 2019. https://arxiv.org/pdf/1908.00220.pdf

For people who are interested in this kind of idea, there is some research on estimating Word-Color associations.

A popular crowd-sourced dataset for this is [1] which contains average color association for 14,000 words. There is a demo available at [2].

There's also a recent work[3] trying to estimate such association using image data from Google, similar to the OP project, but a bit more sophisticated than just taking an average.

1: Colourful Language: Measuring Word-Colour Associations, Saif Mohammad, ACL 2011 Workshop on Cognitive Modeling and Computational Linguistics (CMCL). https://www.aclweb.org/anthology/W11-0611/

2: http://www.lexichrome.com/

3: Estimating Color-Concept Associations from Image Statistics. Ragini Rathore et al., IEEE VIS InfoVis 2019. https://arxiv.org/pdf/1908.00220.pdf

While I think general awareness of ethics concerns is needed, I think it might also sometimes bias research directions in itself.

I.e., dealing with ethics concerns and/or ethics committees becomes a huge additional workload in itself, so the research is prioritized to minimize dealing with it.

For example, one might stop a research which might help to treat cancer but dealing with the necessary approvals for patient data makes it unfeasible. Instead you switch to a general purpose target domain where it suddenly (unintentionally) could be used for war instead, but being general purpose it does not need to be approved by the ethics committee..

> It would be a shame if ML also became a field people avoided because they didn't want to contribute to evil, in their own view.

Both ethics and privacy considerations have recently become pretty regular at Computer Vision and Multimedia Processing conferences.

One very popular object detection model (called YOLO) had the main author recently leaving the field because he got concerned about military applications using his research results.

https://medium.com/syncedreview/yolo-creator-says-he-stopped...

There's very good advice in there for people in leading or mentoring positions.

I am in academia and I noticed many professors are a mixed bag when it comes to such communication skills. They don't really compliment students much for their work or results. Basically all feedback is target at existing issues and how to improve things. I feel many people can quickly get discouraged by this.

When I give students feedback I always start with thanking them for the work and pointing out some things I think they did very well.. Most students seem to appreciate it a lot.

Yes, the seeming performance of (especially) neural models compared to traditional models is probably the main factor. Although, some voices[1] argue that traditional or much simpler approaches still often do a similar job compared to super over-engineered models, especially when going even slightly beyond an existing target-dataset or task.

I'd argue, that improving the ML models is really the job of ML researchers and should be mainly targeting ML conferences like AAAI (Adv. of AI). In other conferences (directly targeting NLP, CV, Comp. Biology, etc.) it should be the main job to combine those models with the domain-specific characteristics (e.g., language information for NLP) or "traditional" methods to make it an interesting discussion.

I was recently doing reviewing for a multimedia conference and quite a lot of the papers I reviewed were basically pure ML papers. A colleague had the same experience.

1: https://arxiv.org/abs/1907.06902

NLP is not my main field but still relevant to my work because I often use models and resources from NLP as tools. I'm also personally interested in Linguistics and Languages so I follow related news, sometimes attend NLP conferences and follow people in those fields on Social Media.

It is very concerning how few thought is usually put into linguistic or language characteristics when dealing with these topics. I also rarely see cultural considerations etc. Basically everything is considered as "machine learning will hopefully get this right if having enough data" which is unfortunate (ML is a great tool but the conferences are about language processing).

Another big issue I noticed is that a majority of research only targets or evaluates English texts. In many cases the language is not even specified (although it is clear they use English from figures or examples). I even heard people complaining that work on non-English data is treated as too minor by many reviewers so stuff like that often just gets rejected.

I think this is a really weird development for a field which centers around natural languages.

I live in Japan. I once had a package delivered from overseas where their printers couldn’t print CJK fonts and thus the whole address resulted in just small empty boxes. The post office inferred my address from the post code + my name and delivered it correctly. There wasn’t even a (noticeable) delay.

That I was hinting at in the last sentence. Apple Watch 5 and iPhone 11 came out on the same day and needed watchOS 6 and iOS 13 respectively, so basically, those hardware releases forced the buggy *OS to be released across all platforms and devices.

Comparing iOS 10~13 with macOS 10.12~15 release dates, macOS seems to always come out 7 days after iOS.

This year it has been 12 days after iOS 13.0, instead. Wouldn't really call that much of a push back. It's less than one week behind the usual schedule..

This year all their OSes seem to be riddled with issues at release.

- iOS 13.0 was so bad they released 13.1 in less than 5 days, but even now many things are still hit and miss (with 13.2 in beta)

- watchOS 6.0 is also still pretty bad and not yet fixed (with 6.1 in beta)

- macOS 10.15 GM seems pretty buggy

- Well, I think tvOS 13 is ok?

While the situation might be better for people who use the latest betas, it is still a horrible current user experience for all normal users just updating their devices.

Lots of cross-platform features introduced across these updates (like the new iCloud features and new Reminder apps, etc.) are also in a horrible state.

I'm not sure what their QA team is doing this year but it seems almost everything planned for this Fall would have been better off if pushed back a couple of months. Well, if it weren't for device compatibilities... (the iPhone 11/Watch 5 seemed to be more important than stable software across all their platforms and other devices)

Apple also has a blog where they discuss (some of) their machine learning results[1].

1: https://machinelearning.apple.com/

Indeed. I used to use LR Classic on my Mac, but since the new LR got more and more features I switched to using my iPad almost exclusively. I don't think I used LR on my Mac in months.

NordVPN is not in the list[1]

1: https://vpnpro.com/blog/hidden-vpn-owners-unveiled-97-vpns-2...

Isn't a battery replacement just $199, anyway? Might have been the cheaper alternative even without the recall.

I had a similar decision a few weeks ago (contemplating whether to go from a 2014 15" MacBook to a 2019 model because of a faulty battery and touchpad.)

I went to the Genius Bar, they replaced the top-case including the battery, a new keyboard, and a new touchpad, for $199. They also replaced the display for free while they were at it (relating to coating issues,) so I virtually got a new MacBook (from the outside) for $199.

My concern was that there is no clear upgrade path for static websites, if running out of free bandwidth.

All per-site addons focus on different kinds of dynamic features (Forms, Identity, Functions, ...) I suppose a Team account would work, but that also seems like paying for unneeded features.

The support team just linked me to the pricing page.

Most of these smartphones and tablets have homescreens which serve a pretty similar purpose. And people are used to having links to all their favorite apps on it.

I think he is aware. He made a comment about it in the next paragraph.

> hey, did you know that accuracy and precision are different?

It's fun because directly above the Textexpander comment is Day One, and directly below it are 1Password and Ulysses (all 4 apps have almost same subscription model).

YNAB 4 is also a pay-once desktop app. The subscription service is usually referred to as "New YNAB". It's kind of YNAB 5.

Since I began my PhD I started writing a personal journal about my research.

I write down all things in my mind, things I tried out, things that did not work, etc. Every time I get new results or evaluate something, I also try to include a table or graph with intermediate results.

The journal helps me to see my progress over time. I can grep through it and see if I already tried something. I can go back and see how my results get better over time or whether I hit a dead end somewhere.

I also write a small summary directly after meetings and talks and note feedback from supervisors.

I use org-mode (emacs) for this and can recommend it. All my entries are sorted by year and week number, so I try to write a summary at least once a week. As I use Emacs for all my programming related tasks, the org-mode file is open all day anyhow, so I usually add things more often than that.

The OneNote clipper felt a little lacking and sometimes buggy for me. The Evernote clipper really shines in how smooth everything works.

Despite having some domains which I mostly use for gaming purposes and stuff, my main "serious" email address is @fastmail.fm. It is a nightmare to tell it to someone by phone. It goes as far as "dot fm, never heard that, do you mean .com?" Wtf, that doesn't even sound similar. Emails never arrive.