tanvach

Alternative explanation - maybe the training data is full of internet discussions, and it's just that we we really bad at self correction online.

If this means we can download and print (injectable) drugs in the future, does this mean we can literally get a virus from dodgy sites?

‘Economically’ is probably the keyword that’s missing?

The actual implementation of the retention analysis is not too hard. If you can compute unique customer id, then yes. But how do you get unique id, if say user is not logged in? What if user uses multiple devices? What if user creates new account? The third party solutions I’ve seen (which is fairly limited), is to embed a code into your product, and they take care of the tracking.

I worked as a data scientist for a few big internet companies, and have done a lot of retention analysis (from the ground up). I'd say, it's a hassle for a reason: accuracy. Third party tools seems to give little validation of correct user tracking, and you may have quite different concept of 'a distinct user' and 'retention'. It gets quite deep if you really think about it, both the tracking part and the analysis part. Happy to get into the weeds if you are interested.

Apart from that, I'd also love to hear what hassle free options are out there.

This short youtube video really helped with me to understand the big pictures of category theory with concrete examples.

The Mathematician's Weapon | An Introduction to Category Theory, Abstraction and Algebra https://www.youtube.com/watch?v=FQYOpD7tv30

If the language has low cost to repeat words then I think the value of gender or tenses is going to be low. Thai, for example, has tones so our words tend to be shorter (more information can be packed per syllable). Languages are solving compression in different ways.

Another fun example - we repeat an adjective to turn it into adverb. Since most adjectives are 1 syllable, it has the same cost as adding ‘ly’ in English.

Working with the Thai government can be a bit of a pain - forms must be filled in Thai only.

Agree with the characters. Although it’s not super hard to learn (my non-Thai wife learnt the characters pretty quickly), I wish we officially have something like pinyin.

Fun fact: written Thai in Latin characters is unofficially called ‘karaoke’ [1].

[1] https://blogs.transparent.com/thai/how-to-readwrite-in-karao...

Also agree on the use of language extends beyond transmission of thought - the art comes into using language as a tool. I bet all languages has these hidden corners of complexity for it to be expressive enough to be wielded, just under appreciated by non-native speakers.

Hmm I think translating that faithfully will need a bit more context. But usually you can just say ตอนที่อยู่ในเมืองจะไปตัดผมด้วย translating to ‘whilst in town (will) go cut hair also’. The ‘จะ / will’ is a universal word that can be added to other verbs to signify ‘future in due course’.

In my thinking, tenses is a bit like compression where you can condense temporal relationships into a single word. In Thai, we are more explicit and add more words. Though Thai words tends to be shorter (since we have tones to disambiguate sounds), adding words are not that expensive. I’m not a linguist, so this is just a hypothesis.

Also, don’t take it the wrong way - the usage of superfluous is strictly logical. Like saying computer languages are superfluous (who ever needs anything other than an infinitely long tape?).

As a Thai, I’m somewhat proud to see that our gramma is quite simple, i.e. no tenses and genders, etc. These extra structures are superfluous if you think about it. This makes it easy for beginners to start forming simple sentences. However, the complexity gets pushed to the vocabulary. It tends to have different words that mean the same thing, but with differing levels of, say politeness. Not to mention a whole ton of slang and double meanings.

Great video!

Thank you for sharing, I enjoyed reading the paper and gained some insights. To me, bivector makes rotation vastly more intuitive (the gyroscope precession example is case in point).

First, like some commenters, I believe you may be depressed. If you have resources to help you, please seek it out to verify. The more personalized (counseling) the better.

Second, any chance you are passing time by watching tv and surfing the internet?

There’s a period of my life I used to watch a lot of tv series and doom scrolling. Looking back, I was depressed with my studies. These mindless activities helped to numb my soul, making life less painful.

Slowly though, I realized the cognitively cheap entertainment was a form of addiction, and that it was robbing me of my creativity and energy.

Third, regarding exercise and diet. Super critical. But! You need to find the right kind that suits you. I highly recommend bouldering. You can do it alone, any time, and there’s a clear and constant progression you can see. Also, people are generally nice and social.

My response to this is that not everyone thinks like you. Art is inherently subjective.

Nice UI, and the approach works quite well (occasional wrong choice of palette compared to ground truth). It’s kind of fun to turn my photos into B/W and run through the model to try and guess the output.

Do you have a privacy policy for the uploaded photos? I’m not keen on uploading anything important without knowing how it’s stored or will be used in the future.

I’ll probably be buried in all these comments, but my position is that data is only as good as how it is collected. Sloppy data collection gives rise to sloppy conclusion through unknown biases.

The key is to understand the ‘data generation process’ so you can identify biases. My experience suggests that doing so side-step some common pitfalls.

I recommend reach out for ‘The Book Of Why’ by Judea Pearl. He includes many real life examples that’s surprisingly applicable to modern data science.

I find these posts very valuable, thank you for sharing.

Hmm I don't see it in the picture. Either it's been updated, or my eyesight is getting pretty bad...

I actually like Apple Pay for the extra security it provides. At least in the US, there’s non-negligible chance that your card can be skimmed (happened to me before). Using Apple Pay feels much more secure, though come to think of it, I don’t have a real proof that it is.

Honestly I like the energy here. Just have to distance myself from a bit from people on sociopathic spectrum. I did 'lucked out' and joined a company that IPO'd, but honestly probably didn't make it much more than other people who started companies outside of the bay area. Also the access to nature has been amazing for me and my mental health.

- Studied in UK, came to US for a major and landed a job in finance then data science.

- Mostly hanging out with fiends from work (or previous work). Not much of a 'normal' social scene except around events.

- Just love the idea of living and breathing hard and meaningful problems. Bay area has interesting stuff going on all the time.

- I thought SF would be a beautiful, clean city. It's really is a dirty place. Oh well, I put up with it. - Dating sucked. Met my wife here in the city outside of dating context (long story).

- Feeling like the industry is losing momentum. Less inspiring, more grinding. We should focus on hard problems, but a ton of money is spent on stupid get rich schemes. Software is making the world a worse place?

Fear of failure. I tend to find comfort in things I know well. Starting a company has many moving parts, it’s daunting.

Lack of meaning. I haven’t found meaning in the ‘business world’. Much of my time is spent doing hobbies I find meaningful and enjoy. I’ve found a balance in my life now to do interesting things at work, while having good work life balance. Starting a company upsets that balance mostly in a negative way.

Lack of cofounder trust. I live in the Bay Area, and a lot of people who have started companies are not the people I trust or enjoy working with. Someone with values I associate with and respect tend not to start companies.

I do want to start a company one day, only if I can add value to it, and it adds value to the world.

I created and opened up the design of a standalone DSP [1] that leverages AutoEQ. The DSP is optimized for our home brewed headphones for VR, but I’ve been using it for my desktop headphones too.

AutoEQ is not a perfect tool, but an amazing (and free) starting point to get the most out of your audio systems.

[1] https://github.com/tanvach/prettygood_dsp

I’m assuming they needed to have very accurate orbit information for each satellite. Was this provided to them by SpaceX, or was it derived from the signal?

An ex-Amazon PM ‘promoted’ the silent reading culture in my previous company. Definitely brought a lot of benefits:

- forced more senior / leadership members to read through and internalize the details. Reduces wasted time asking already answered questions.

- we used google docs. Question and answers were posted as comments. This formed a log of the majority of the discussion. Less note taking needed.

- allowed less assertive members to ask questions or give answers.

The only drawback - the overwhelming silence at the start that I never really got over…

Medlife Crisis has a great video on running a long term experiment comparing the two shoe tying methods: https://www.youtube.com/watch?v=G4R5--MpP0Q

Thunderf00t was involved, and he explained how the study was carried out really well https://youtu.be/lANjwPzISQw

A warning: my father read about DIY air filter just like this one, and made a couple of them at home. The fans are not designed to run continuously, and one failed and caught on fire. The living room burnt down but luckily the fire was put out before it took down the whole house. Damage costed orders of magnitude more than the extra few hundred $ from buying a commercial air purifier.

Sean Carroll has a great podcast episode discussing the results with Netta Engelhardt, one of the co-authors a paper discussed in this article [1]. He also has a great series on Youtube [2] that builds up modern physics from elementary ideas, up to and including AdS/CFT and black hole information theory.

[1] https://www.preposterousuniverse.com/podcast/2020/09/21/115-...

[2] https://www.preposterousuniverse.com/blog/category/biggest-i...