|
https://twitter.com/sharifshameem/status/1283322990625607681 This is just a toy for now, but in ten years it might not be.
|
#
?
Jul 15, 2020 23:03
|
|
|
|
| # ? May 9, 2024 10:07 |
|
|
All you need now is the end user to know exactly what they want and how to describe it.
|
|
#
?
Jul 15, 2020 23:14
|
|
|
A quick feedback loop can help them get there, at least.
|
|
#
?
Jul 15, 2020 23:22
|
|
|
Can’t wait for the brain implant that will let me generate entire websites with a thought.
|
|
#
?
Jul 15, 2020 23:38
|
|
|
Protocol7 posted:Can’t wait for the brain implant that will let me generate entire websites with a thought. I'll preorder that.
|
|
#
?
Jul 15, 2020 23:57
|
|
|
Xik posted:All you need now is the end user to know exactly what they want and how to describe it. Until there's a descriptor implemented for making it "pop" I don't think anyone has to worry
|
|
#
?
Jul 16, 2020 00:48
|
|
|
Xik posted:All you need now is the end user to know exactly what they want and how to describe it. And something that implements all the backend logic...
|
|
#
?
Jul 16, 2020 00:49
|
|
|
Can't wait for the job listings requiring 6 years of experience of with GPT-3.
|
|
#
?
Jul 16, 2020 02:06
|
|
|
I haven’t really been following this except for the first announcement which I think was posted here or the Coding Horrors thread, and this tweet. Have there been any demonstrations of GPT-3 other than short demo videos by the creator? I’m not saying it’s full of poo poo but I’m thinking it really hard.
|
|
#
?
Jul 16, 2020 02:24
|
|
|
It’s always good to be skeptical but at the same time something like this is inevitable. The skeptic in me says it’ll probably get rolled into some Wix-like site builder and stamped with some Powered By AI marketing bullshit though.
|
|
#
?
Jul 16, 2020 05:09
|
|
|
Oh I believe there's parsing that's translating the demonstrated sentences to CSS or whatever to make the two buttons have more padding. It might even be a decent NLP engine, but whooooo boy is there some poo poo-rear end vague language employed by programmers of every description. I'm just highly skeptical of the ability to handle to anything remotely underspecified and/or actually complicated. I think the first canned demo I saw was "is a string a palindrome," and the Google homepage isn't exactly a model of complexity.Xik posted:All you need now is the end user to know exactly what they want and how to describe it. "Okay GPT-3, make the padding between buttons like Google but from 2007. I took a picture on my phone and you can download it here: C:\Users\Dipshit\Desktop\GoodGooglePicture.doc"
|
|
#
?
Jul 16, 2020 05:47
|
|
|
csammis posted:"Okay GPT-3, make the padding between buttons like Google but from 2007. I took a picture on my phone and you can download it here: C:\Users\Dipshit\Desktop\GoodGooglePicture.doc" I laughed at this uncontrollably for way too long. Oh dear this is too real.
|
|
#
?
Jul 16, 2020 08:12
|
|
|
csammis posted:C:\Users\Dipshit\Desktop\GoodGooglePicture.doc" This except it's a blank outlook saved message which has that document as an attachment. The doc itself has two page breaks and on the third is a shrinked screenshot with all the important text unreadable.
|
|
#
?
Jul 16, 2020 09:59
|
|
|
Protocol7 posted:It’s always good to be skeptical but at the same time something like this is inevitable. The skeptic in me says it’ll probably get rolled into some Wix-like site builder and stamped with some Powered By AI marketing bullshit though. The language processing is cool and all, but I don't think it's the future. Something that can take mocks and turn it into a page is probably more likely to be useful, and even then only for static pages.
|
|
#
?
Jul 16, 2020 13:58
|
|
|
It's 2020. Screenshots from clients come in as phone camera pictures of another phone, over text message, at 12:45 AM
|
|
#
?
Jul 17, 2020 13:10
|
|
|
Vulture Culture posted:It's 2020. Screenshots from clients come in as phone camera pictures of another phone, over text message, at 12:45 AM How far we’ve come from 2000’s: “email has an attached word doc, that contains a screenshot of an image to be uploaded to the website.” What a brave new world!
|
|
#
?
Jul 17, 2020 17:24
|
|
|
He's building basic application functionality into the GPT-3 description-based builder now: https://twitter.com/sharifshameem/status/1284095222939451393
|
|
#
?
Jul 17, 2020 20:13
|
|
|
Time to break into the factory and break all the GPT machines
|
|
#
?
Jul 17, 2020 23:59
|
|
|
Protocol7 posted:He's building basic application functionality into the GPT-3 description-based builder now: Ah, yes, that solves the hard problem of software development, actually getting your users to tell you what they loving want.
|
|
#
?
Jul 18, 2020 00:18
|
|
|
Bruegels Fuckbooks posted:Ah, yes, that solves the hard problem of software development, actually getting your users to tell you what they loving want. Clients have no problem describing things. The problem is that what they actually want and what they describe are two different things. So as long as it supports phrases like “make it pop” or “make an app that does x... but not like that” it’s fine! 
|
|
#
?
Jul 18, 2020 00:38
|
|
|
This thing seems cool, but I can’t help but think it’ll get all out of whack once you take it outside softball demo cases a la Hotdog Not Hotdog. As a UI designer in the enterprise data analytics space, I can imagine a lot of scenarios where it would be easier for me to just write the drat component directly than to dick around with natural language input that basically needs to legalese-level watertight to correctly get your point across to the AI. And then once you have the AI generate you a component, how does manual tweaking work? If you wanted to be able to tweak it and still be compatible with modification efforts by the AI, I imagine it would have to keep track of manual changes so it continues to be in the correct context. And what if I wanted to make something more complex than an “online business card” type website or toy app? Would it be able to build you an app with CMS functionality, like for a portfolio you can add projects to? What if I want the component’s parameter to be a complex data object? Would I have to describe how the app/component is to interpret the object? Or does it even work for that level of description? How can I instruct it to talk to the API? Does it consider security? If I’m a restaurant and I want to post an online menu, do I have to spell out how it’s to format each entry or can I give it a text doc or CSV with the menu items and just tell it the display rules? The more I think about it the more questions I have. Anyhow, the natural language space in AI is fascinating to me. At a previous job I got some exposure to bleeding edge agricultural/industrial robotics and would daydream what the human-robot interface would be like - the ultimate one was just natural language and maybe showing it a map, just like how you communicate with a human worker. I think we are still a ways off from that, though.
|
|
#
?
Jul 18, 2020 14:33
|
|
|
I can’t help but notice that all the output is class components, which seems like a bit of a regression since functional components have been the new hotness for a while now. The earlier versions output raw HTML which probably would be more useful in the long run. I mean it wouldn’t take a skilled React developer very long to write a fully functional increment/decrement or todo app, and I agree beyond that it’s limitations are likely to be painfully obvious. I’m more interested in GPT-3 itself, seeing as it supposedly can generate articles that are nearly indistinguishable from human-written pieces.
|
|
#
?
Jul 18, 2020 14:46
|
|
|
Queen Victorian posted:This thing seems cool, but I can’t help but think it’ll get all out of whack once you take it outside softball demo cases a la Hotdog Not Hotdog. Protocol7 posted:So as long as it supports phrases like “make it pop” or “make an app that does x... but not like that” it’s fine!
|
|
#
?
Jul 18, 2020 15:53
|
|
|
There's a lot of opportunities to use this kind of tech to make websites accessible, but I'm certain this tool will only ever be used to create inaccessible websites. Would be interesting to see how it fares in the hands of someone like Glitch though
|
|
#
?
Jul 18, 2020 16:04
|
|
|
Isnt it just just ML text generation, where the text happens to be JSX?
|
|
#
?
Jul 18, 2020 21:18
|
|
|
smackfu posted:Isnt it just just ML text generation, where the text happens to be JSX? It uses GPT-3’s NLP features and probably some very very very strict rules on how to build and structure a “React App.” At this point I’m not convinced the app won’t just break or just generate useless code in 90% of situations and all we are seeing is just the best case examples to generate hype. Macichne Leainig fucked around with this message at 04:32 on Jul 19, 2020 |
|
#
?
Jul 19, 2020 04:28
|
|
|
Bruegels Fuckbooks posted:Ah, yes, that solves the hard problem of software development, actually getting your users to tell you what they loving want. Bro, bro, bro! I have this awesome idea. Make, like, Twitter, but faster.
|
|
#
?
Jul 19, 2020 04:41
|
|
|
Protocol7 posted:It uses GPT-3’s NLP features and probably some very very very strict rules on how to build and structure a “React App.” I can fix this issue, all we have to do is put it on the blockchain.
|
|
#
?
Jul 19, 2020 06:06
|
|
|
Che Delilas posted:I can fix this issue, all we have to do is put it on the blockchain. Why stop there? I won’t be satisfied until we have serverless artificial intelligence IoT applications powered by the blockchain to provide actionable insights into natural language processing.
|
|
#
?
Jul 19, 2020 06:27
|
|
|
Protocol7 posted:Why stop there? I won’t be satisfied until we have serverless artificial intelligence IoT applications powered by the blockchain to provide actionable insights into natural language processing. Can we get some machine learning to help with this? My grandson says that it can play mario, so it seems like it could do business.
|
|
#
?
Jul 19, 2020 06:34
|
|
|
In all seriousness, I think it’s important to note that the Debuild tool has an official site complete with generic landing page elements and a link to a Google Form for “signing up.” Very official!
|
|
#
?
Jul 19, 2020 06:39
|
|
|
Protocol7 posted:In all seriousness, I think it’s important to note that the Debuild tool has an official site complete with generic landing page elements and a link to a Google Form for “signing up.” Was the official site generated by the Debuild tool though?
|
|
#
?
Jul 19, 2020 08:06
|
|
|
https://twitter.com/yoavgo/status/1284645259935461383 I'm not sure that I completely agree with this, but I can't say it's wrong either.
|
|
#
?
Jul 19, 2020 15:09
|
|
|
Queen Victorian posted:This thing seems cool, but I can’t help but think it’ll get all out of whack once you take it outside softball demo cases a la Hotdog Not Hotdog. The devil on my shoulder says that all product owners should have access to this tool so they learn how to ask for things better... My actual opinion is that this will go the way of Dreamweaver for writing HTML. It will work as expected, but the generated code will be so hyper-specific and bloated that it will be unmaintainable spaghetti, and that many devs will be employed strictly to clean it up. There are thousands of low-paying PHP positions doing exactly that for people's Wordpress sites. vonnegutt fucked around with this message at 23:11 on Jul 20, 2020 |
|
#
?
Jul 20, 2020 23:03
|
|
|
Bring back MM_swapImgRestore()
|
|
#
?
Jul 21, 2020 13:42
|
|
|
ultrafilter posted:https://twitter.com/yoavgo/status/1284645259935461383
|
|
#
?
Jul 21, 2020 15:56
|
|
|
ultrafilter posted:https://twitter.com/yoavgo/status/1284645259935461383 using bootstrap and jquery should be more accepted.
|
|
#
?
Jul 21, 2020 16:06
|
|
|
So I'm sitting with what feels like an embarrassingly dumb question rolling around my head. In the project I'm working on, we've got selection controls populated with arrays of key-value pairs - one part to show the user, and one part to use in the logic. For example, let's say the user is asked to select an amount of money from a dropdown. They select "$100.00" and we run with the other part of the KVP, "100". Now on to my dumb question: what should be the key and what should be the value? My gut instinct says the string shown to users should be the key and the logic-related data should be the value, but the opposite is more often true in the existing code (but that's also inconsistent). Am I just being an enormous idiot here? Also, a fun side note: I say these are KVPs, and functionally they are, but for some reason before my time the team rolled their own "IKeyTextItem<T>" interface instead. Which is just a KVP except with "key" and "text" instead of "Key" and "Value". Which, again maybe I'm just wrong, but this seems... stupid. This is in Typescript by the way, but I figure my question has the same answer regardless of language.
|
|
#
?
Jul 29, 2020 11:44
|
|
|
Neither of them is the key - or rather, the concept of a "key" is meaningless without any context around what the key actually means. You have a text string, and a numeric value. In some parts of your code, you may take a text string as input, and return the corresponding value, which would make the text the key in that particular context. In other parts of your code, you may take a numeric value and look for a matching text string, which would make the value the key. In yet more parts of your code, you might just take a collection of pairs and use it to populate a list of options, in which case neither is acting as the key. Not every pair is necessarily a key-value pair, sometimes it's just a pair. Even if your language provides Pairs (or Triples, or higher-dimensional Tuples), it's often still a good idea to create a new type to represent a specific combination of data (in this case, a selectable item in a dropdown). This allows you to give meaningful names to the individual pieces of data (instead of just "first" and "second" or whatever the default names are), and makes code a lot clearer to read. I might quibble with the name they've chosen a bit, but their overall approach seems fine.
|
|
#
?
Jul 29, 2020 13:00
|
|
|
|
| # ? May 9, 2024 10:07 |
|
|
Jabor posted:Neither of them is the key - or rather, the concept of a "key" is meaningless without any context around what the key actually means. So I think that threw me off and I approached the question from the wrong starting point, ultimately. Thanks a lot for the reply, it's a great help! E: also I realize there are threads better suited to questions like this, it's just that this is one of the few threads I manage to keep up with properly even if I rarely post. Woebin fucked around with this message at 13:34 on Jul 29, 2020 |
|
#
?
Jul 29, 2020 13:11
|
|