Could you explain to us in details what kind of analytics data are you collecting? Also for which purposes are these collected?

I am interested (and I bet more people here) in getting a comprehensive list minus the marketing BS.

Great question. We're in the midst of updating our privacy policy, but I'll give a full explanation here.

It may be useful to differentiate between what data we collect and what data we don't collect in order to clear up any confusion.

1. What we collect

We collect a variety of usage analytics that help us understand how you use Kite and how we can improve the product. These include:

* Which editors you are using Kite with

* Number of requests that the Kite Engine has handled for you

* How often you use specific Kite features e.g. how many completions from Kite did you use

* Size (in number of files) of codebases that you work with

* Names of 3rd party Python packages that you use

* Kite application resource usage (CPU and RAM)

You can opt-out of sending these analytics by changing a Kite setting (https://help.kite.com/article/79-opting-out-of-usage-metrics).

Additionally, we collect anonymized "heartbeats" that are used to make sure the Kite app is functioning properly and not crashing unexpectedly. These analytics are just simple pings with no metadata, and as mentioned, they're anonymized so that there's no way for us to trace which users they came from.

We also use third party libraries (Rollbar and Crashlytics) to report errors or bugs that occur during the usage of the product.

2. What we don't collect

* Contents (partial or full) of any source code file that resides on your hard drive

* Information (i.e. file paths) about your file system hierarchy

* Any indices of your code produced by the Kite Engine to power our features - these all stay local to your hard drive

How do you guys plan on monetizing this? This list of analytics seems fine at a quick glance, but I'm concerned that there's not a transparent path to profitability here.

Just like any other service, there's no guarantee that you won't start collecting snippets of source code or other metadata to start selling once the VC's start applying pressure to generate income. What's your strategy?

> You can opt-out of sending these analytics by changing a Kite setting (https://help.kite.com/article/79-opting-out-of-usage-metrics).

Shouldn't this be opt-in according to GDPR?

> * Line-of-Code Completions - Kite's completions engine can now predict several tokens of code at a time, powered by the most sophisticated AI code models available.

Won't this result into copy-paste-like functionality of StackOverflow? This seems great for small projects, but doesn't look useful at for anything else it seems. I'm skeptical that Kite provides anything of value compared to all of the other tools I have.

The website has mentioned that other languages "are coming soon", though it has said this since the last Kite scandal. Are these just empty words?

> Shouldn't this be opt-in according to GDPR?

No, because it isn't personal data.

But cookies that save your search queries are personal data and require opt-in? You can consider anything not to be "personal data". I still really want this to be opt-in, not opt-out.

Here’s some feedback: pretending you’re grateful for users screaming at you for breaching their trust is not a good way to regain that trust.

> If you're already a user, Kite has been auto-updated and is now working locally without sending code to the cloud. If you previously uploaded code to our servers, you can remove your data via our web portal.

> We're grateful to our users for helping us reach this point.

Why aren’t you scrubbing all surreptitiously uploaded data proactively?

Why are you pretending you are glad you got caught doing sketchy things?

You’re going to have to do a lot better than that to convince us that your company suddenly operates with integrity now simply because you got caught.

Thanks for moving it off the cloud and making it local.

That is a huge step towards making us comfortable with the potential privacy aspects.

Do you collect any information?

And how will you make money? (I want you to be around, pay staff, do well etc)

Quick question: Is it doing much differently from TabNine[1]? For me TabNine has been incredibly accurate at guessing what I'm wanting to type.

[1] https://tabnine.com

Great question!

TabNine is a different approach that uses far less semantic information than Kite. It really shines when you're talking about syntactic repetition, e.g. if you check out the screenshots on their homepage. They have a page about semantic completions, but the semantics are very shallow -- basically what attributes are on the instance you're accessing.

In contrast, we've spent ~50 eng-years semantically indexing all the code on Github, building statistical type inference, and rich statistical models that use this semantic information in a very deep way. The result is that Kite can help more often, in ways that reflect a deep understanding of the semantics of the code you're writing.

Not to mention by pushing malware to people via IDE addons, pretty good strategy for training those datasets.

Interesting - is the model that learns from (semantically indexed) code bound by its license? One could argue it is derivative work.

Interesting question. I’m not a lawyer, but I presume it’s derivative enough not to be copyrighted. For example, one could argue Google’s giant n-gram corpus is derived from many copyrighted webpages.

> Our core belief is programmers spend too much time on repetitive work like copying and pasting from StackOverflow, fixing simple errors, and writing boilerplate code.

Do you have data backing this core belief up or is it just an intuition? I don't feel any of these things are problems I need a solution for. Maybe this depends on the type of programming someone does or the language they use but as a working programmer this is not a compelling sales pitch to me. You're offering solutions to problems I don't have. They're also not problems that I see to any great extent in junior programmers who report to me.

On an intuitive level, I would guess there are many, many developers debugging a "you forgot to cast your int to a string" error message as I'm writing this.

More concretely, the StackOverflow thread for "Parsing values from a JSON file?" has almost 3 million views. We take this as evidence that programming is repetitive.

(We should probably clarify that when we say 'repetitive' we mean on a global scale, not an individual scale.)

> I would guess there are many, many developers debugging a "you forgot to cast your int to a string" error message as I'm writing this.

I've always avoided dynamically typed languages because of these types of issues so I guess this seems like a solved problem to me - just use a statically typed language. Clearly there are many people who are using dynamically typed languages for whatever reason though so I can see how better tooling might be useful to them.

> More concretely, the StackOverflow thread for "Parsing values from a JSON file?" has almost 3 million views.

I certainly make use of Stack Overflow, I was taking issue with the idea that significant time is spent copying and pasting code from the site though. I often refer to it to answer a programming related question but I almost never then copy and paste code. I'm looking for a pointer to a library or API, to understand some confusing or poorly documented behavior or for a workaround to a bug etc. and getting an answer to my question rarely leads to copying and pasting code in my experience.

I am using dynamic languages and have for years, but I don't see this as something that would help me in my work. I mean, autocomplete is nice sometimes, but when I know what to write and where, those little autocomplete boxes just get in the way. Maybe I'm also not the target audience though. Which is good - after reading how they started I would never willingly install their software.

There's a trend amongst dynamic languages to slowly add type hints (see Mypy for Python, Typescript, etc) so I see these issues becoming less important over time.

I've been using Python on a daily basis for close to a decade now, I can probably count on my hand how many times a bug has been because of casting a string to an int.

"Our core belief is programmers spend too much time on repetitive work like copying and pasting from StackOverflow, fixing simple errors, and writing boilerplate code."

Sounds like you will likely be introducing bugs into people's code. Or at the very least, regressing code quality towards the mean.

Maybe, but these days you have programming languages/libraries with minimal documentation and underspecced parameters/returns, you have to hunt for examples of use on StackOverflow/Github/etc. and many frameworks/tasks have annoying boilerplates everybody has to type all the time.

Moreover, there are common errors you can avoid if their ranking algorithm discourages code reported/identified as bad.

Kite aside, I think both of these things will happen as developers adopt these technologies. We're seeing similar effects with autocorrect on your phone and self driving cars (i.e. there will still be accidents, and they tend to drive more slowly), but on balance these technologies are positive.

For me autocompletion is the second thing I turn off on a phone (right after vibrate on touch). There is nothing more annoying than writing a word correctly, and autocomplete "fixing" it, just because it was slang word, or a word in another language because I forgot the english word in that moment.

A lot of that is because the minimal phone keyboard doesn't give you a good UI to accept or decline a suggested autocompletion.

With a keyboard and mouse and large monitor, you can display the completion prompt and let the user select or ignore from the various options with Ctrl-Space or similar.

When "Select the first autocomplete over what I typed" is the default behavior, yeah, that sucks. I often end up using the "key combo" Space,Backspace,Space to accept the autocomplete on my phone.

But I definitely see utility in offering a correction for, eg. 'DateTime.Format("yyyy-mm-dd" -> yyyy-MM-dd' or any of the million other idioms that we have to remember or look up. What I don't understand is how Kite can offer useful local-only suggestions - Is the default dictionary that comes with it preloaded with a million lines of open source, or is it literally just my code being suggested?

Curious how much better your approach is versus fairly vanilla "most popular ngrams".

N-grams is the first thing you try when you want to statistically model code. We tried it in 2014 and the results are disappointing, if you want to do anything beyond identifying low level syntactic patterns.

The intuition here is that in natural language, context is defined locally. If you want to know whether a 'they' refers to a male or female, for example, you look at the nearby text. In contrast, the flow of data and control through code is highly non-local, which is a lot of why techniques like N-grams don't work.

This has been a very active area of research in academia since we started Kite in 2014. (Suggested google searches: [big code], [ml on code], etc.)

All that said I don't have data on how our approach would compare to N-grams, but I'm guessing if you look at some of the academic research, you'll find that NLP techniques were abandoned, despite the early papers' focus on them.

Thoughtful answer, thank you. People complain about HN having gone downhill...but answers like this one still pop up fairly frequently.

This is excellent! Thanks for allowing me to play with it!

Did you publish any papers on how you approach intelligent code completion at Kite (the parts you can make public or just rough areas I should look into? I understand you have many emerging competitors). Thanks!

Yes, this may be a good starting point — https://github.com/src-d/awesome-machine-learning-on-source-...

Thanks for checking it out!