It gave me the numbers for the various trials. I don't have them handy and I didn't run them manually - so the AI could totally have made it all up. I did not keep the numbers so I have nothing to share.

Sure - it's just a hypothesis.

I do not recall the details! But I did watch as the damn AI did repeat wall clock and sample tests to see where the time was going and try 2-4 different approaches and clocked each one in detail. I felt embarrassed because I *could* do all that and certainly had done it in the past - but it would have taken me a *long* time to set up each tests and analyze the results and try over and over and track the results for each.
As you imply - none of this is particularly tricky - it's mostly just labor intensive and time consuming. And those are both things that AI has no trouble with.

Related - when I tell an agent to optimize for performance - especially if the request is well targeted - it does an *amazing* job. Doing performance work is so well defined and described that it can follow the patterns and do all the checks and get it really right. To do a good job it's really repetitive and there are often a lot of angles to check and .. it just does it really well.

This is like Trump promising $5K if republicans win the midterms. It's just a PR stunt.

Imagine trying to reason out the tax implications of something like this. The privacy implications. Everything.

Elon: "I really wanted to, but lawyers wouldn't let me!"

No way to see that coming.

I predict it could cost as much as $150 Billion!

Do I hear $200B? We're making shit up - go large!

My day job is a (saas) app that's about 20 years old and has been worked on continuously by a team that has varied in size from about 5 to now about 15. It is complicated and connects to several external services. There are parts of the app that are too hairy for AI to reason about. But we have found, recently, that AI is surprisingly good at explaining how some rather complicated (but well written) parts of the app work.
The 2-3 really hairy parts of the app are written in an arbitrarily complicated way that I would never have used and long to rewrite - but they work. Nobody on the team is happy with it - and there is no corporate memory of anyone being happy working with these parts.
We have exceptional test coverage (>95% by lines), so it's just a matter of time before we decide to lay out an AI agent team to rewrite some of the horrible stuff in a way that is much more palatable to both human and agent. This is not something I would have suggested as possible 4-6 months ago.

I'm not sure how to answer that question. My flip answer is "the usual way." But nothing about software is the usual anymore.

100% of the code that .. makes up this project is code that I did not write. But I absolutely told AI what I wanted and in many cases *how* I wanted it. And any code that had "interesting" design elements is code that I reviewed and in some cases told AI to change. Where interesting refers to meaningful architectural decisions.

At this point it is a very light project - somewhat less that 10K lines of code. It is in its minimal viable product phase - and I imagine it will grow to be more than 15 or even 20K LOC. Eventually becoming projects on 3 distinct platforms and becoming substantially larger.

All of which is ridiculous. Fewer than 200 prompts so far, I'm sure. Maybe fewer than 100.

I've got a black swan for ya!
I just shipped software that AI wrote [practically] entirely that I would not have shipped otherwise. I don't think I'm the only one.

Pfft. Morons. It is *super easy* to make predictions.

Accurate ones, however...

Yes, because users understand those kinds of choices.

Too soon!

Way more than cheaper rides.
* App to summon
* Automatic driver mapping to pickup
* Predictable arrival for pickup
* Predictable arrival for drop off
* No BS routing and you knew the cost before pickup/dropoff.

Back in the day I called for cabs ahead of time and had them fail to show up 25% of the time. I've never lived in a dangerous area, so that wasn't it. They just were not reliable or predictable. Uber saw a huge opening for a service and they took it. They also saw a huge opportunity for destroying the existing market solution in an anti-competitive and illegal way and took that, too - banking on the notion that government would not act fast enough to stop them. And they were right.

And because cars are a certain size. Making it half the size for just one person would be nearly as expensive and would be worse for taxi'ing 2 people (which are still a reasonable number of fares).

Account closed.
I created my twitter account when the service was practically new and only used it a handful of times - so I might not be the target audience. Even so, I miss the original idea of it.

Oh, no. I have it on good authority that computer vision is about to exceed human performance.

Wait - what year is it, again? :-p

We have companies that are making the statement that their AI has escaped the lab and run amok. That brings to mind a ton of questions: what were the prompts that started this? What evidence do they have? And who was it presented to? Why hasn't the FBI started interviews? If a single person in their basement had created an AI Hacker wouldn't that person run afoul of every anti-hacking law? Why isn't it the same for these large corporations? And how much compute time and tokens did it take to create these hacking models?

That last question is the most important because as it stands not one of these companies of produced anything that is of value. Meaning none of these models have increased human productivity, none are doing anything without prompting, and models being left to run alone and "escape" is not just unethical, and unscientific but if those models began hacking competing companies then why aren't these companies asking for criminal or civil.

There are a lot of stories stating how it happened but that begs more questions like those above. And just how much are these companies already spending on the extreme compute times. I mean, didn't Amazon lose millions of dollars to Anthropic because of run away computing time & tokens? Where are these companies, who have yet to make a dime, get the extra money to spend on all this compute time & tokens?

Do I believe these hacks happened? Yep absolutely but they were all intentional to try and show what these models can do and to cover for the crazy financing schemes that all these tech companies are doing to look like they will at some future time make money but Oracle is tending to junk status, every other tech company is trading the same billions of dollars around. And we still don't have a product that will actually do anything that will save any company money. At least 2 trillion dollars will be spent without anything showing for it.

This is worse than Meta's bet on VR. The, if we build it, they will come mentality hasn't caught up with reality. These models cost more than the humans they are meant to replace. And with no central cost saving product that will blow everyone minds and become a commercial success this seems like just more snake oil.

And by not holding these companies criminally liable under all those anti-hacking/cracking laws for these models "hacking" their competitors just re-enforces that this is all more for show. I'm unsurprised that I have yet, please link to an article if I'm wrong, one politician raise the idea to hold these companies criminally responsible.