[samuelfekete]

There’s a bias towards coding capabilities amongst testers (and perhaps trainers) of ChatGPT.

I bet it has (or can have) a similar level of capability when it comes to chemistry, biology, finance, law, etc. - all human knowledge that is expressible as text or formulae.

With regards to software engineering, a great amount of time is currently being wasted by coders trying to solve problems that have already been solved by others. Some of the solutions for that previously included libraries, SaaS, and Stack Overflow.

Now we have another tool at our disposal to 10x all software engineers (and perhaps the same for other industries).

[jstx1]

I think the question is what happens when you're able to 10x all software engineers but don't have job market demand to match the 10x increase. 90% layoffs? Maybe not 90% because these processes aren't perfectly efficient or evenly distributed, and also because you can increase productivity without affecting demand to some extent, but what's the %? 30-50%? That's still huge.

[pornel]

Perhaps we'll have Jevons paradox, and instead of reducing the workforce, we'll accelerate software eating the world.

A lot of software is crappy. The extra productivity could go partially into quality improvement.

[samuelfekete]

We haven’t automated 10% of what could be automated, especially in tasks that were previously not cost-efficient to automate or that required AI tech to automate.

[rich_sasha]

My understanding is that much of automation difficulties are actually to do with interactions with the physical world. We still don't have a machine for folding laundry that matches human performance.

Coding is nice for AI to overtake because it's all abstract (apart from context). The problem domain is literally just string to string mapping.

[skydhash]

Coding may be abstract, but execution of the resulting program is not. And results of the execution is driven by real world needs. Truth is that a human can invent things because it can pattern match across whole domains. You can say there is a mechanic solution to that, how can we do an algorithm that have the same result. AI cannot unless the algorithm was already created. I think the current state of AI is great for searching and creating starting point, but it can never get us to the finish line.

[civopsec]

Finally a software engineering silver bullet post-No Silver Bullet.