AI and it's effect on your music, your job and the future

[narad]

But again - the agent doing that reasoning is the programmer, not the computer. The most brilliant algorithm is not itself doing any reasoning. The programmer did the reasoning ahead of time, and embedded the expression of such into the software.

In older AI systems, yes. But larger LLMs induce an ability to perform logical reasoning to the extent that they perform better at benchmark tasks without ever being explicitly programmed to do so. The idea that the programmer imparts their own intelligence / reasoning ability into an LLM like ChatGPT is not accurate.

 

[crushingpetal]

In older AI systems, yes. But larger LLMs induce an ability to perform logical reasoning to the extent that they perform better at benchmark tasks without ever being explicitly programmed to do so. The idea that the programmer imparts their own intelligence / reasoning ability into an LLM like ChatGPT is not accurate.

Yeah, wasn't this covered by one of Turing's early papers, the "creativity objection" or something. The charge was supposed to be that "computers only do what they're told" but Turing points out (even in the 1950s) that computers do things that surprise and surpass the programer. Hell, IBM's Watson is better than all its programmers at Jeopardy.

 

[narad]

Yeah, wasn't this covered by one of Turing's early papers, the "creativity objection" or something. The charge was supposed to be that "computers only do what they're told" but Turing points out (even in the 1950s) that computers do things that surprise and surpass the programer. Hell, IBM's Watson is better than all its programmers at Jeopardy.

Oh, I hadn't heard of that. But I think that's a different thing. Even game AI, which we probably shouldn't even mention in the same breath as research AI, the programmer could lay out a variety of rules, and through their interactions, create behaviors that surprise even the people who made it. But they're still embodying some of their own intelligence in the creation of those rules.

I'm not sure I don't want to consider it as "real" AI because it doesn't learn by itself, or because it's not conceivably scalable to build up human level intelligence manually in that way (so I might have to reconsider that position), but in either case, I feel it is different from modern AI (/GPT) which is both learning what the rules are and implying them. That key combination being what gives rise to a tool that is practically useful across many domains (about as many domains as you can write an effective manual for?)

 

[Drew]

Here are a few ways AI could have impact on our music creation (to be clear, a human is generating this list ):

i) your music could be mastered by an AI service
ii) a few plugins have "digital assistants" that supposedly find areas of eq masking and make suggestions (iZotope has something like this)
iii) logic's drummer can generate drum patterns based on a source audio file
iv) (maybe) the quad cortex profiles an amp with a neural net

I'll think of more. *I don't want to argue if this is _really_ AI or just a smart algo. The difference doesn't really matter for the moment.

I think my point, though, was exactly none of these things has currently happened to me.

 

[lurè]

I only use AI writing tools to help with the lyrics a bit. And that's it.

And I don't use ChatGPT, it's too bad for this. I use Sudowrite, way better. It costs only $19 per month. Here are other AI writing tools: https://writingtools.co.uk/pricing.html. Feel free to browse them, if anyone is interested in a bit of help.

 

[p0ke]

Regarding my job:

Any software developer who's ever in contact with clients will know exactly what I mean

 

[Rubbishplayer]

Regarding my job:

View attachment 147903

Any software developer who's ever in contact with clients will know exactly what I mean

Regrettably, and with respect, I think you are wrong. And I'm speaking as ex-programmer, Scrum master and product owner.

AI is already writing code based on simple textual input: you can do that today with ChatGPT. Moreover, with training, it can learn to understand subjective specifications, especially if expressed as personas and user stories.

On a lighter note, a great excuse for this clip...

 

[lurè]

I only use AI writing tools to help with the lyrics a bit. And that's it.

And I don't use ChatGPT, it's too bad for this. I use Sudowrite, way better. It costs only $19 per month. Here are other AI writing tools: https://writingtools.co.uk/pricing.html. Feel free to browse them, if anyone is interested in a bit of help.

Dafaq happened

 

[p0ke]

Regrettably, and with respect, I think you are wrong. And I'm speaking as ex-programmer, Scrum master and product owner.

AI is already writing code based on simple textual input: you can do that today with ChatGPT. Moreover, with training, it can learn to understand subjective specifications, especially if expressed as personas and user stories.

Yeah, I know ChatGPT is able to write code, but you completely missed my point. Which is that customers will never know what they want to begin with.

 

[Rubbishplayer]

Yeah, I know ChatGPT is able to write code, but you completely missed my point. Which is that customers will never know what they want to begin with.

Spoken like a true programmer.

 

[p0ke]

Spoken like a true programmer.

No but seriously. It always starts with "we just want a basic app" and evolves into a huge system that does everything. I'd say 9/10 projects are like that. And usually the whole focus of the app is entirely different from what it was when development starts.

 

[Rubbishplayer]

No but seriously. It always starts with "we just want a basic app" and evolves into a huge system that does everything. I'd say 9/10 projects are like that. And usually the whole focus of the app is entirely different from what it was when development starts.

And thus it has always been, which is why we have The Agile Manifesto (nothing to do with great value Korean-built 7-strings), which has formalised this indecisiveness into a manageable project management approach.

 

[Randy]

No but seriously. It always starts with "we just want a basic app" and evolves into a huge system that does everything. I'd say 9/10 projects are like that. And usually the whole focus of the app is entirely different from what it was when development starts.

As someone working in marketing/branding, I will second that the client not knowing wtf they want will always be the thing standing in the way the most. And that they likely don't have the awareness to prompt the software into giving it what they actually want.

 

[TedEH]

I've been making this point for a long time whenever it comes up - I don't think AI can meaningfully replace programming because it isn't actually intelligent. It lacks greater context and the ability to collaborate with other teams, and interpret client requests, and meaningfully debug, etc etc. It can do basic versions of these things, with guidance, but there's a lot of human work in being a programmer if you intend to produce anything valuable.

In the same way that AI "can" write songs, and AI "can" make videos, and AI "can" play video games, it's only doing so with a lot of guidance and is mostly producing slop that wouldn't hold against deeper scrutiny or tighter requirements or more complex problems. AI can fake the Software part but it can't fake the Engineer part of Software Engineering. Programmers don't just produce lines of code, they're problem solvers and collaborators, roles that AI is not currently suited to.

 

[Moongrum]

Yeah, I know ChatGPT is able to write code, but you completely missed my point. Which is that customers will never know what they want to begin with.

My product managers don't even know what they want

That said I'd be happy if AI replaced every software engineer, even though I am one. Maybe people would leave Washington and it would be a nice place to live again.