pugio

It's a fair point. Qualified ML people working at big companies command huge salaries. Work at big tech: get lots of $$ and access to big compute.

Our value prop is different. Work with us, and do something really meaningful. I think that's a pretty normal tradeoff between for-profit and non-profit companies, no? (Higher pay vs possibly higher goal satisfaction).

I didn't mean that "a better education is one which focuses on programming". We'd like to improve education across the board, but have chosen to focus our efforts on STEM topics at the moment.

Humanities are by no means lesser, they're just not what we're focused on, presently. (Also, a really good humanities education probably looks very different from a really good STEM education setup. Different challenges, different problems to solve, at least initially.)

In short: we're looking for people with overlap between instructional design experience and programing+ml industry experience, to create instructional content.

A masters in education/instructional psychology isn't needed, so long as you can demonstrate pedagogical aptitude. In other words, can you put yourself in the mind of a beginner, and craft an explanation which is clear, intuitive, and anticipates common student questions/pitfalls?

Random shot here, but I work for a non profit interested in building a better kind of education (focused on programming, ml, and data science).

We've been having a hard time figuring out how to hire qualified people to build top-notch educational content because we pay less than industry rates.

The upside is that we do meaningful work, have good health care, decent pay, good work environment, (edit: also fully remote-able), and job stability (we're funded by philanthropists and don't need to make a profit).

If anyone hit by the recent layoffs is passionate about good education and would like a change of pace, feel free to email me (address in profile) to start a conversation.

It's worth noting that the Pydantic author stated he's rewriting parts of the validation in Rust for the next major release, with the aims of drastically improving performance.

I'm definitely interested in a simple type-safe validator library, but that planned performance boost is pretty tempting.

What about a commercial source but fully source-available product?

I'm currently building an app I hope will appeal to the HN crowd, and I was planning on making the full source available to all customers - this would be billed as a feature of the product.

I can't count the number of times I was happily using a closed source product only to find a bug or missing feature that I really wished I could just make a small tweak to add/fix. I didn't need the app to be open source or anything, I just wanted to be able to see what was running and make a small change myself. For many of these situations I would happily commit the patch back to the commercial product, just so that my experience would be nicer.

I want to be able to charge for and sell my app, but I would like all my users to be able to see exactly what's running under the hood, and be able to tweak and modify it for their own personal use. (I also plan on including an extensive public extension API. Extensions can also be open source of course.)

To me this seems like a fairly good sweet spot, as I really do need to charge to be able to support the development. I could even see committing to always keeping the source available to users, or to open source it if I ever stop commercial development of the product. I hope this appeals to folks here, because I can't see a better model that will support a single developer as I go.

Coincidentally, the latest episode of The Rookie (S05E02 - a show which portrays the LAPD as highly caring/moral) was partially about a group of dirty cops. Essentially the good cops were (sheriff plus deputies) like: "Cops can hang out together all they want, but as soon as they give themselves (their group) a name, they are destined for jail. 'Never fails'."

Great show, funny how that works out. I'm sorry that it isn't remotely accurate it seems (when it comes to the rest of the cast).

Sweet! I'll keep an eye on the repo. Thank you for open sourcing DocQuery. I agree with your reasoning: my current attempts to find an NER model that covers all my use cases have come up short

Your commercial product looks very cool, but my use case is in creating an offline-first local document storage system (data never reaches a cloud). I'd like to be enable users to search through all documents for relevant pieces of information.

The templates sound very cool - are they essentially just using a preset list of (natural language) queries tied to a particular document class? It seems like you're using a version of donut for your document classification?

Wow this is exactly what I've been looking for, thank you! I just wish with these transformer models it was possible to extract a structured set of what the model "knows" (for e.g. easy search indexing ). These natural language question systems are a little too fuzzy sometimes.

Just published an app on both app stores using Svelte and Ionic. Svelte was a pleasure to use - everything seems like just enough of what you need, without adding any additional mental overhead or complexity.

I'm really looking forward to the stable release of SvelteKit to use as the backend for my next project.

I really appreciate the use of pip > conda. Looking forward to the update for the repo!

Can you describe how you did (/ are doing) this? Do you now need to use conda (as opposed to OPs pip only version)?

Me at the end of last year: "Should I really go for the full 64GB on this M1 Pro? What could I possibly use this for? mmbml mumble... something about unified GPU... something Deep Learning, one day..."

Me now: "a red juicy apple floating in outer space, like a planet" --H 768 --W 768

Uses about 27GB. 1.81s/it.

Can't do 1024x1024 yet because of some hardcoded Metal issue (https://github.com/pytorch/pytorch/issues/84039.

It sounds like we have a very different idea of what "ethical" means. In answer to your question about "what's stopping them?" - well, just that: ethics, morals, the desire to act decently even when there might be some additional benefit to being sleazy.

Waiting for a law to prohibit immoral behavior doesn't seem like a sustainable way for a society to function, even though it does seem to be the way we're trending.

Related to the copyright discussion in another thread... does anyone have a sense of legalities if someone recorded audio versions of these and released them for free? I suspect you'd need the author's permission.

I tried to use Thunderbird for this for a while, but there's a weird bug where it creates copies of emails that have the deleted attachments. You then have to delete those copies to fully delete everything.

Switch to the featured app here and everything worked seamlessly. Definitely worth it, as this way I don't have to pay for Google storage and can clean out a bunch of pure junk from my emails.

Check out https://withorbit.com - an ongoing research project (and set of tools on github) to embed SRS into any essay/post. If more people adopted it, you might have what you want.

I'm really enjoying the discussion here, as I've been thinking a lot about what a full modern ML/DS curriculum would look like.

I currently work for a non-profit investigating making a free high quality set of courses in this space, and would love to talk to as many people either working in ML/DS or looking to get into the field. (I have ideas but would prefer to ground them in as many real-world experiences as I can collect.)

If anyone here wouldn't mind chatting about this, or even just sharing an experience or opinion, please drop me an email (in my profile).

EDIT: We already have Into to DS, and a Deep RL sequence far along in our pipeline, but are looking to see where we can help the most with available resources.

I really appreciate this Interviews book as an example of what topics might be necessary (and at what level), taking into account the qualifying discussion here, of course.

Watching this run, I wonder if you could use something like this to desensitize yourself from notification addiction. Since this is stimulus without any reward, ever, may it break the normal stimulus-response-reward cycle of real notifications?

In The Scout Mindset, Julia Galef calls a very related state "the soldier mindset" - this is similar enough to the term you proposed but I think you would enjoy reading the book.

> "The responsible life before God" -- I don't think any modern person is capable of understanding what that meant in the 1940s. We just don't think that way anymore.

I would suggest that perhaps you are limiting your scope of possible "we" (21st century human) thought states. The passage seems to me perfectly clear. For reference, the full quote is:

> "The word of the Bible that the fear of God is the beginning of wisdom declares that the internal liberation of human beings to live the responsible life before God is the only genuine way to overcome stupidity."

He's discussing one way out of the trap of conforming to group/tribal beliefs. If you put your fear of God (in other words, some power higher than, and distinctly external to, any social group) you begin to liberate yourself from caring too much about what other people think.

If you believe that God demands that you act as morally (for some definition/code) and as responsibly as possible, and you place that demand/requirement higher than any other motivating feeling, you will ask about any belief or course of action "Is this Right?" (ie. Does this align with God's command), and consider that question more deeply than the creeds or slogans of your fellow mortals.

Note that he's not saying that the above state of mind is easy, it's merely something just strive for. Many devout religious people still try to the best of their ability to work towards that exact state.

Check out The Scout Mindset. It's the best book I've found on how to think clearly (on these kinds of potentially tribal thoughts), and how to productively interact with people who may not be trying to improve in this area (the "stupid" people).

We're actually making it required reading at my workplace.

I've really been enjoying the posts from Chris unpacking the TruffleRuby optimizations. I've been hearing about Graal and Truffle for years, but didn't understand exactly what made them so fast.

Paired with my current reading of Crafting Interpreters, these posts make me really excited for all the cool things you can do when implementing a language.

Coming from traditional statically typed languages (Java, C++, Typescript, even Python or Ruby...), you might think of a type like "int" or "Float". He's calling stamps "micro-types" because they're more fine-grained than that. A stamp might tag a value as being between 0 and 255, for example, which is much more specific than "int".

It would be really cool to work with a language that allows the user such fine-grained types, which is one reason I keep eyeing Idris.

My colleagues and I discuss, off and on, what it would be like to introduce chavruta learning to other domains. We dream of an environment like that for chemistry, physics, etc.

I think I've seen it develop in Math circles (two folks going at it on a blackboard for hours on end), but there's something really powerful about the format which doesn't seem prevalent elsewhere.

As many social systems evolve and optimize over time, Jewish Beit Midrash learning has evolved over millenia to be an effective educational tool, and I think (/hope/suspect) there's a lot which can be extracted from it and applied to other learning contexts.

As an educator, I think about this a lot. One approach I'm prototyping for a (hopefully freely available) course on reinforcement learning is to translate dense math expressions into English sentences, and then link each piece of text<=>math with a corresponding color. (Quick example I just uploaded: https://gfycat.com/favorablemintyharrier ).

By doing this consistently, I can highlight the link between math symbol and corresponding concept, and help the student learn how to read and interpret dense constructions as we go.

(Note that this approach is intended to be used as part of a larger explanatory strategy, so please assume the example is embedded in an appropriate context.)

I actually just reverse engineered the Victoria app yesterday. I don't really have any reason to do this, but if anyone does want to submit a whole bunch of check-in requests, it's pretty easy. A single CURL call will do it.

I apologize for my lack of clarity. I intended my capitalization of "Freedom" to stand in for a particular philosophy, which I did not clearly explicate.

To restate: I think the COVID lockdowns are a very different case from these electronic restrictions, in scope, duration, reason, and advisability. A critique which cites both as similar examples of government overreach is missing its mark.

You're conflating two very different things. The lockdowns are instituted at the state level, are different in every state, and are with the specific aim of (and in response to) preventing covid transmission. Adhering to a public health lockdown has nothing to do with Freedom, or lack of guns.

By making a false comparison between the two, you are muddying any argument made against the electronic freedom restrictions.