221–230 of 547 posts
When I read “non-local context,” it really drove home for me just how off the mark they were and changed the whole tone.
It also makes me think were they just hoping the solution would fall out of the sky? Seems irresponsible if that was part of their calculus.
> It includes our data-driven Python type inference engine
I couldn't find which repository this lived in. I am very interested in it, as my team maintains a few open source static analysis and code generation tools. We'd be interested in trying this out.
Credit to where credit is due. I worked in R&D here in a group tackling it for almost a decade ("program synthesis"), and while Copilot has a lot more to do, it solved so much of the usability & basic use case gap of what the R&D community had been attempting for years. Large language models & transformer models have been out for years, and the Github team executed well on adapting them.
(Separately: There _is_ an interesting question whether this space is VC-investable -- how likely will at least 1 startup here make it to 9-10 figures of revenue. But that's another story.)
"Our diagnosis is that individual developers do not pay for tools. Their manager might, but engineering managers only want to pay for discrete new capabilities, i.e. making their developers 18% faster when writing code did not resonate strongly enough." I never used Kite, but I've tried Github Copilot twice, and found it marginal at best (and distracting at worst - which is why I turned it off both times). If Kite wa…
My experience with copilot has been very different. It easily pays for itself, and getting my employer (seed stage startup) to spring for it for the entire team was an easy sell. Yeah it's pretty dumb most of the time. But I know that, and I don't use code from it without carefully checking it out and modifying it. But it's still a huge help. Just the time saved writing tests alone pays for it. And I've had a few spo…
> Our diagnosis is that individual developers do not pay for tools. Throwing salt on the wound here but that’s just false. I mean, there’s copilot and it’s alternative that I can’t think of the name right now. more broadly there’s Jet brains ides, visual studio, Productivity apps, etc. look at product hunt or appsumo or popular show hns. Devs pay for tools, just not Kite. Edit: I should clarify, enough devs pay for t…
Frankly I don't pay for tools. Money is tight at work and at home, so if it's not free it ain't happening.
> Then, our product failed to generate revenue. Our 500k developers would not pay to use it. I don't pay for Kite (or any other proprietary developer tooling like Github) because one day your company can choose to shut down, change its terms, or raise my prices and I'd be left without recourse, while also being locked in to a proprietary workflow. Just like you did today, which validates my hesistation. Kite should h…
I don't think the being locked in a proprietary workflow bit is your real reason, because when you break it down - this doesn't make much sense. Fear of needing to switch workflows down the line outweighs the [potentially temporary if company dies] boost in productivity?
Of course, this assumes kite fits your workflow well and you find it delivers value (you don't cancel immediately)
Earlier quoted context omitted.
My experience with copilot has been very different. It easily pays for itself, and getting my employer (seed stage startup) to spring for it for the entire team was an easy sell. Yeah it's pretty dumb most of the time. But I know that, and I don't use code from it without carefully checking it out and modifying it. But it's still a huge help. Just the time saved writing tests alone pays for it. And I've had a few spo…
Kite has been around for a lot longer, if anything Copilot was Github copying them
From an old programmer perspective, it doesn’t make much sense, but a new programmer will not want to learn the old way, which will be effectively obsolete from lack of updates. If there’s any value to be derived from it, perhaps it is demand for hardware that will run enormously-inefficient code. The way that now you see people doing full sorts to get the third-largest value just because it’s easier to write it that way, you will see code that also does analytics and builds a distributed hash table to accomplish the same task, just because more capability means more usage means more suggestions to carry along that code.
I think it was a mistake to think of computer programs as a linear text language, but I don’t see this turning back. At some point, the concept of programming a machine will merge entirely with the method of interacting with a machine, which is to say, communicating intent, and then I suppose we can relax into a very comfortable full-service 5-star extinction.
"Our diagnosis is that individual developers do not pay for tools. Their manager might, but engineering managers only want to pay for discrete new capabilities, i.e. making their developers 18% faster when writing code did not resonate strongly enough." I never used Kite, but I've tried Github Copilot twice, and found it marginal at best (and distracting at worst - which is why I turned it off both times). If Kite wa…
The first group gets annoyed all the time because Copilot fails to write most code when prompted with comments, or writes inaccurate code at best. They get upset when they see that Copilot can reproduce GPL code when prompted in a specific way.
The second group most prompt Copilot by allowing it to tab a complete a line or two at a time, and they are actually super happy because Copilot is way better than any other existing autocomplete; it's basically in a class of its own. To them, the GPL issue seems a bit more abstract, because they would never use Copilot to do that anyways.
I fall pretty firmly into the second camp (can you tell?). Allow me to soliloquize for a moment. Copilot is an incredibly powerful tool, probably the most powerful one I have, but, just like any tool, you need to really learn your way around it, and understand what it can and cannot do, before you start making judgments. I'm not surprised that you turned it off after using it twice. Imagine saying you stopped using React after making two components!
Maybe I should write up a bit more about how I use Copilot, but in a nutshell I feel that it falls somewhere between a 2x-better autocomplete and (and this bit is even more interesting) a tool similar to google search, but more tightly integrated with the coding environment. The second bit is why it's so good. Imagine if I were to continuously google search everything I was doing while coding, while I was coding it. Sure, most of the time it'd just confirm you were doing the right thing, but... every now and then, Google might turn up a better strategy than the one I was currently trying. That's how I feel Copilot works all the time; it's continuously "google searching" for alternate approaches, and every now and then it'll be like, "aha, did you think of [this thing]" and really take me aback, because I wouldn't have even thought to Google for that particular bit of code / problem / strategy.
Of course, you could continuously google search everything you did as you did it, but it would be a massive waste of time. Just imagine Copilot is doing it for you, and returning what it found. Most of the time I know what I'm doing, but every now and then, the result is remarkable.