The Future of Code Writing. Is there one?

[Thales750]

Large Language Models and The End of Programming - CS50 Tech Talk with Dr. Matt Welsh

 

[Thales750]

Did any of you watch this?
It's a little bit of review, but it also had some tips on using the AIs of the World to be one of the people last standing when the rest of them are gone.

 

[Mike Krailo]

Guess we're all in trouble.

 

[Isaac]

1200/day, wow

 

[Thales750]

1200/day, wow

That is obviously wacka doodoo. It was for some super specialized data engineers, plus support. We won't need those ratios to eliminate jobs.
I'm starting an engineering company and the entire business modal is based on using AIs for everything possible. I don't even know what that means.
But, I am determined to find out.

 

[CJ_London]

If by business model you mean to draw up a business plan - i.e. market analysis, source and use of capital, projections for growth etc, make sure you ask the right questions in the right way.

 

[Edgar_]

The video rises some interesting points, one of which is "no one understands AI", I can think of a probable outcome for that:
> writing software becomes crazy cheap
> legal problems arise
> creator blames AI
> creator is forgiven
> more legal problems arise, now more serious
> creator tries to blame AI
> doesn't work, is punished
> crazy cheap software becomes crazy dangerous to use
> human developers start solving the issues
> AI-written software supervised by human developers becomes the new
> developers are back

But I think it's safer to assume someone will create tools to understand it better.

Another issue is cybersecurity or security in general. If software is written without people understanding how it works because it's been written by a machine, someone with enough dedication will try to reverse engineer that and compromise stuff without people even knowing because, well, no one understands it.

 

[Edgar_]

But I think it's safer to assume AI will create tools to understand it better.

How can the machine possibly know what we can understand and what we can not? It needs feedback. Now, if you replace your last "it" with "itself", that's nightmare fuel, and I know quite a few people who would give the AI the tools it needs for something like that.

We can only wait.

 

[Thales750]

If by business model you mean to draw up a business plan - i.e. market analysis, source and use of capital, projections for growth etc, make sure you ask the right questions in the right way.

I've been trying to learn the best ways to do that. I read something interesting today. 59% of some kind of marketing folks would give up 10% of their paychecks before they would get rid of the AIs that can create images, content, and videos.

 

[Isaac]

If by business model you mean to draw up a business plan - i.e. market analysis, source and use of capital, projections for growth etc, make sure you ask the right questions in the right way.

If AI could replace humans making judgment calls then all people trading stocks would be rich by now given the software they have used that it has incorporated AI for many years

 

[Cotswold]

One thing that I wonder about with AI generated programming is changes.
We all know that the final analysis that is agreed is often flawed, with adjustments being made during development. Then after installing, the client will often request additions and changes.

Will any manual changes to an AI generated program made later by a developer cause issues? Bearing in mind the code being changed will not be at all familiar. And certainly more verbose than they may have created. Alternatively, will you be able to respecify the application for it to be rewritten by the AI program writer to include required changes, with the existing application remaining unchanged?

If the changes by the AI program generator are made a couple of years later. Will it then be a totally different version of the AI generator software and therefore create a mainly different program that incorporate its own updates as well as the ones requested?

In general I find program writers to be a real time saver, even the ones that I have created for myself. But I just wonder about using AI and will the results be to everyone’s liking. A bit like the car makers who all appear to use the same software to design their SUVs and as a result they all have the same boring and near identical appearance. With only the badges being different.

I always had a resistance to buying software from some engineering companies. They had some manager or director who thought they could do an equally good DIY solution. Often mentioning spreadsheets as their solution. I told them that wasn’t going to happen. (a) Because a spreadsheet wasn’t the solution and (b) they didn’t have the time. I’d go back two years later and nothing had changed, except the intention.

So will the effect of AI simply cause companies to defer buying an application for years because they imagine they can do it themselves?

 

[CJ_London]

59% of some kind of marketing folks would give up 10% of their paychecks before they would get rid of the AIs that can create images, content, and videos.

that is because it does 90% of their work

 

[Cotswold]

If AI could replace humans making judgment calls then all people trading stocks would be rich by now given the software they have used that it has incorporated AI for many years

That being the case, can AI pick the all the winners on a racecard?
If so then the next Cheltenham Gold week should be a good few days.

p.s.Has anyone tested it?

 

[CJ_London]

The bookies?

 

[Thales750]

If AI could replace humans making judgment calls then all people trading stocks would be rich by now given the software they have used that it has incorporated AI for many years

I think the trading computers that are really good at it are being used by people that can afford to keep it from the rest of us, and secret.

Price fixing is automated in main stream markets for all sorts of goods. I think it has been a large contributor to the recent round of inflation.

 

[Isaac]

And not only that, but stop and think for a moment about trading data. It's the most empirical, black-and-white, clear, solid data you can get. Every day, every price, every stock, every event.

And still, only a few people get rich.

The incorporation of AI people are trying now is much less well armed. It has googly information including opinion, conjecture, speculation, and a few facts along the way. People need to be careful, as the news stories coming out about embarassing stupid things people have tried to use AI for are growing exponentially

 

[Thales750]

And not only that, but stop and think for a moment about trading data. It's the most empirical, black-and-white, clear, solid data you can get. Every day, every price, every stock, every event.

And still, only a few people get rich.

The incorporation of AI people are trying now is much less well armed. It has googly information including opinion, conjecture, speculation, and a few facts along the way. People need to be careful, as the news stories coming out about embarassing stupid things people have tried to use AI for are growing exponentially

Maybe we should stop calling it getting rich. Maybe we should call it what it really is, becoming King.

 

[Edgar_]

There is a lot of data available publicly, enough to be an absolute danger. It is not enough for Sam and his team, though. I just bought me an API key, I have become their customer now. What I'm seeing is hilarious.

They made it very easy to integrate into any system, just copy paste your stuff and the AI will "know" how to remember all of that without intermediary developers creating databases for it. You literally just paste your data and IT WILL REMEMBER it whenever you ask for it in any format, and that includes conversational format, so, non-programmers can query this too. And if you, as the user, can create queries for that, be sure OpenAI will also be able to do that.

This is very important, because they might say they're not storing your personal data, but who knows that for sure? So, since you can now just paste, I insist, the AI brain will let you query that and you will be able to have your data available anytime. Those scientists created a monster that scraped some public data, but they will now possess private data as well.

How can a backend developer even compete with something that is capable of organizing ANY type of information without the need for some SQL schema, or rules, or anything at all? Anyone and their grand mother can now just paste information into the system and ask for it at will. Very powerful for companies, but very dangerous for humanity.

It's over. Humans need not apply.

 

[Isaac]

Edgar, that may be the most valuable post regarding ChatGPT, or one of the most, that I've seen so far.

We're all talking about how to get data out of it, but you're exploring the inputs it can save. Very interesting, thanks for posting. That is kinda cool. I wonder how it would be using sql to query the input data. Of course maybe theoretically you wouldn't "have" to, but let's say for the sake of argument that you found it easier to write the sql query than you did conversate with it. Funny

 

[Edgar_]

I guess you could tell the GPT to organize your inputs in tables. You'd need to have a tool that actually executes the responses it gives you. Now that you're there, another tool that tests if the code it gave you actually works. And then query with it. Since it sounds more complicated, I'd stick with the conversational format, maybe make an agreement with the GPT to return your results in tabular format whenever you ask for them in a SQL-esque format? I mean, look at this trending github repo of rules for chatbots:
https://github.com/linexjlin/GPTs

Here's one of them:

That's how easy it is becoming...