k8si

Again, vegans almost always have to read the ingredients/labels, on every processed food product they plan to consume. The little 'vegan' icon on the back is new and not consistently used. Choosing a plant based lifestyle is A LOT more burdensome than not doing that. I know because I've switched back and forth many times and am married to a vegan. Whey, casein, random cream, honey, they're in everything.

Even so: reading ingredients is honestly not that hard.

I imagine this was probably user error on your part. The vegan cheese is usually on the other end of the grocery store from the real cheese, and also usually says 'vegan' prominently, and also there are only like 2-3 brands that make vegan cheese (Miyoko, Daiya, maybe 1 other?). And vegans still usually have to read the ingredients to make sure. Non-vegans could probably do the same if it were that important.

There's no competition between linguistics and ML/NLP, they have completely different goals as fields.

I meant that my linguistics background helps me understand & solve problems: studying linguistic field work has helped me design crowd labeling jobs, knowing about morphology helps me understand why BPE tokenizers work so well (and when they might not), knowing about syntax/dominant word order makes me think that multilingual Bert should probably do something more intelligent with positional embeddings, methods from psycholinguistics are useful for understanding entropy/surprisal wrt LM next-word probabilities... just a few examples but the list could go on.

Language models as a solution to what problems?

Yes, you can easily use AutoModel.from_pretrained('bert-base-uncased') to convert some text into a vector of floats. What then?

What are the properties of downstream (aka actually useful) datasets that might make few-shot transfer difficult or easy? How much data do your users need to provide to get a useful classifier/tagger/etc. for their problem domain?

Why do seemingly-minor perturbations like typos or concating a few numbers result in major differences in representations, and how do you detect/test/mitigate this to ensure model behavior doesn't result weird downstream system behavior?

How do you train a dialog system to map 'I'm good, thanks' to 'no'? How do you train a sentiment classifier learn from contextual/pragmatic cues rather than purely lexical ones (example: 'I hate to say it but this product solves all my problems.' - positive or negative sentiment?)

How bad is the user experience of your Arabic-speaking customers compared to that of your English-speaking customers, and what can you do to measure this and fix it?

My linguistics background really helps me think through a lot of these 'applied' NLP problems. Knowing how to make matmuls fast on GPUs and knowing exactly how multihead self-attention works is definitely useful too, but that's only one piece of building systems with NLP components.

I read in a 'how to improve executive functioning' book that it can be helpful to have one unified TODO list. So that's what I do. I don't think I should make myself a JIRA ticket for 'buy peanut butter and milk'. I also don't think I should make a Confluence page about 'random idea for blog post I had last night while I was falling asleep'. So for things that I need to share that are relevant to work, sure. Otherwise, my notes are generally a very very very long TODO list (bullet journal method has worked for me), punctuated by things that won't quite fit into working memory that I know I'll need to remember in like 20 mins. Also random doodles that I use to help myself pay attention during meetings.

You can turn off the part where it logs into the cloud and insert your own logging backend fairly easily.

Also if your research/biz needs are satisfied by sklearn, then why not just use sklearn? But for a lot of NLP systems, BERT is actually really really useful. And if you don't want to use a pretrained BERT, you can easily initialize their BERT implementation randomly and train it on your own data.

Yep - "Huggingface Enterprise" just like there's "Github Enterprise" seems like a straightforward way to make money, at least to me? Does Microsoft make good money from Github Enterprise?

It's expensive to hire NLP labor right now, and has been for awhile. Seems like one strategy could be: HF provides a cheaper & more scalable alternative to having to hire an in-house NLP team. Basically NLP becomes synonymous with HF.

And they amortize the cost of hiring their own NLP engineers by developing a few models/model-based services that lots of businesses would be willing to pay for. E.g. 'foundation models' for different verticals like healthcare etc. Then it'll also be a lot easier to either fully automate or at least scale up work that's specific to each paying customer (because fine-tuning should go much more quickly, just essentially be a hyperparameter tuning cycle in as many cases as they can get away with).

There's a huge difference between understanding tools/libraries/frameworks/programming languages/APIs etc vs understanding how to build a system on top of any of these things that scales well, is maintainable, is easy to collaborate with others on, that can be extended quickly, that doesn't cost that much time/money to build in the first place, etc.

Yes the former changes every 1-5 years. Doing the latter well is much harder, no single tool can solve these problems, and I think years of experience really does help.

I don't see how Elon Musk being at the helm would improve stock price long term. Yes, making Twitter into a libertarian wasteland would lead to a lot of user growth short term -- the Parler people might come back, along with the bot accounts -- but in turn I predict this would lead to plummeting ad revenue and massive disengagement of its existing (very active, dedicated, passionate) user base, as well as a huge amount of noise that just makes its moderation/NLP infrastructure even worse at filtering harmful content. Maybe I'm wrong/missing something though.

Followers != people who take the time to engage with his polls/other content

Most of my co-workers who have kids will not go back to our office unless our company enforces a vaccine mandate. So anecdotally, I can say that at least some workers at some companies are also putting pressure on to implement vaccine mandates.

It seems like the vax every 6 months is a worst-ish case scenario (worst case being, some mutation is like 100% vax resistant, no one can figure it out, and a lot more people die). Well, one is also supposed to get a dental cleaning every 6 months. The nice thing about the vax is, jab takes 2 seconds and no one bugs you about flossing. I don't see why this would be such a chore.

Sure, but in my office and a lot of the others I have worked in, either a) no one was ever too hot, but many people were too cold or b) the HVAC was old/terrible s.t. that everyone was either way too hot or way too cold. My dream is for everyone to just be a little bit miserable together.

1) Heating pad on my hands while I'm typing? 2) Have you tried typing with fingerless gloves? 3) Not allowed to have space heaters in our office per "the guy who comes around and yells at you for things periodically" 4) This does not warm up my hands, I can assure you 5) This does not warm up my hands, I can assure you

One thing that works great for me at home is, compromising on a temperature with my partner such that we're both physically comfortable throughout the day.

Based on this thread, men are generally much less content with their office temperature than women

I've never been to a tech company where this was the dress code but obviously I haven't been to all of the tech companies

How do I wear more clothes on my hands, when my job is 100% typing?

Introvert's nightmare. Also, who can study in a common area with all the noise?

Oh that sounds so great. I walk out the door to class, realize I forgot a pen, then get to take a 15 minute walk all the around the building just to get back up to my room. Wonderful

Not all students are from countries with great internet access. If you're a Chinese student, you might not have access to many apps professors might assume you'd have access to. The living situation you go back to might not be conducive to studying. Time zone changes will pose a challenge for keeping jobs, maintaining friendships, and doing job networking.

Omfg can we stop making these things bigger PLEASE

Like, who cares??

* What I mean is, text gen models are big enough. We need controllable text generation; like, so it can talk about a specific THING sensibly. Rather than spew statistically plausible nonsense.

In any case, public chain cryptocurrencies are not the solution/upgrade you are looking for.

Funny to look back at this now: https://nyti.ms/2vtd04V

NYT on 24 August 2017: "The Flatiron School in New York may have discovered one path. Founded in 2012, Flatiron has a single campus in downtown Manhattan and its main offering is a 15-week immersive coding program with a $15,000 price tag. More than 95 percent of its 1,000 graduates there have landed coding jobs."

two months later

NY AG, 17 October 2017: "However, Flatiron did not disclose clearly and conspicuously that the 98.5% employment rate included not only full time salaried employees but also apprentices, contract employees and self-employed freelance workers, some who were employed for less than twelve weeks. Similarly, Flatiron failed to clearly and conspicuously disclose that its $74,447 average salary claim included full time employed graduates only, which represent only 58% of classroom graduates and 39% of online graduates."

Wait but haven't there been posts on this very site where studies showed that daydreams & wandering minds help creativity as well? e.g. Something in this article must have been linked and upvoted here at some point: http://www.huffingtonpost.com/2013/10/03/mind-wandering_n_40...

Here are some stats and crappy bar charts:

https://docs.google.com/document/d/1504Sy29t1uUw_B8zTbeLIPCH...

Development is very much ongoing and that's really interesting/nice-to-hear feedback :)

Shameless plug: if you're interested in MALLET you might also be interested in FACTORIE: http://factorie.cs.umass.edu/

Recurring payments: Amazon Prime, Spotify, NearlyFreeSpeech and A Small Orange for hosting, planning to cancel Audible and Wolfram-Alpha because I never use them

Paid for once: Threes, Clear, Convert (for all of my unit conversion needs)

I use but don't pay for: Dropbox, IntellijIDEA, Sublime Text 2

I've also spent a shameful amount of money on Candy Crush...

Yeah... I totally agree that this is a huge risk, and in fact I would expect abuse to happen because cops. Also, to be honest, I really don't like the idea of making software for the man. However, I'll probably still work on the idea this summer, just because it'll probably be an interesting project :)