Looking at other people's response made me think of Lord Of War and War Dogs, both about (real life) accounts of people in the business of selling arms.
That was very illuminating! Do you think you'll try experimenting with some sort of adversarial "agent" setup, where the code isn't released until it passes security review by itself, for each model you are comparing?
Honestly? I did not even know tests existed, because i am of 17 and ai never mentioned so how i supposed to tell the ai that write the tests, that's why it broke.
but when i started building a dedicated backend then i know about tests, coverage, webhooks, idempotency and all that, so i am trying to find out that is there anyone who faced this or am i the only one?
Okay, I get it - to be clear, I wasn't being dismissive, I was genuinely curious. And fascinated at the different ways people use AI, especially down the generations :)
I think what you are facing is a sense of how much the tools are just that - tools, and no substitute at all for what an experienced developer would try to plan for. So yes, I am sure everyone who's using AI tools for the first time has made this "newbie" mistake.
BUT I have also heard that the tools come with "modes" (or agents, or skills, or whatever you want to call them) where you can have it act _like_ an architect, that points out the things you ought to ALSO do. Like write tests, ensure idempotency etc.
I am curious what your experience would be if you went through some cycles of having the AI simply review its work and suggest improvements. Clearly you've already learned some things that you ought to add - I think documenting the progress you make as you discover these things would be immensely helpful to others like you who are new to this!
Thanks! Yeah, I also made that newbie mistake and about modes you heard correctly but even when you say to review the code it still hallucinates which clearly seen in websites and even after multiple correct prompts like "Act like a senior engineer experience of..." something like that, they just write and erase codes but never fixes the issues, they miss the critical flaws, where things starting complexing.
Like I know when I was building an automated workflow, it breaks on the same step every time, even using copilot pro it was unable to fix that, so these are the little things which AI often misses even the PERFECT PROMPT.
What works is that I have to debug it to another AI and then tell the original one that "this is the issue at line X with this error Y".
Depends on what you mean by "business," I suppose, but There Will Be Blood is my favorite, about the "business" of exploiting oil in southern CA though it's much more fictionalized than your other choices :)
It feels a bit alarmist as an article - the obvious next step here is for the FOSS community to start adding AI security reviews to their development cycles. But there'll be a bit of a Y2K-style gold rush before that, as everyone panics and white-hat companies spring up touting their gen-AI credentials.
That's not the title of the article. Which is, "Baby Shoggoth Is Listening." I suppose that title would have made sense to the majority of this crowd but I thought I'd make it something more accessible.
If their money making methods are ethical then isn't this a better strategy, to leave the decisions to others rather than impose your values on them while alive? Also presumably Buffett and his cohorts are better than others at growing their money, so in the vein of the EA argument, it would be best to leave the money untouched while it is being actively managed by the donor, then hand out the windfall after they are dead.
For this argument to work, you have to stop at the donor themselves - you can't keep extending it ad infinitum to their descendants or inheritors. But in the case of the pre-committed amounts, like Gates and Buffett, that isn't the case.
reply