The SV (silicon valley) motto of fail fast and break things accelerated evolution of software design systems. Can you imagine what would happen if people were hesitant to use Git works because a merge could break something? Lol.
But this kind of precaution is quite necessary in the development of AI systems. Current regulations have turned a blind eye to accelerate software and AI progress. Now we have security bugs and misalignments, and we will have AI misalignments as well.
Knowing how computing landscape has changed in SV, I can't imagine modern day SV helping fix FAA systems. Alternatively we could go with multiple incarnations of the same system, being battle tested and proven with independent concurrences, alongside current FAA system. Yeah, it is a boatload of money, and probably wasting 90 percent of it in redundancy.
> Can you imagine what would happen if people were hesitant to use Git works because a merge could break something? Lol.
Version control was always something that was adopted more by process-heavy organizations until the popularization of GitHub. Before that, it was common for the move-fast type of developers to eschew version control.
Earlier in my career, people didn't bother with the chore of using version control unless they were working on an important project that demanded it.
Even though it wasn't part of the curriculum, I learned to use Git (and Mercurial) before I finished school. I found it pretty useful then even for school projects that were, by their nature, solo projects that would never see the light of day. I didn't find Git super burdensome.
Idk, I feel like the difficulty of Git (which is real in some respects) has long been exaggerated by people who are categorically unwilling to RTFM.
One place I applied to 'hmm think we toss it onto sourcesafe once and awhile'. It was their main product. They had about 15 people working on it. Plus one big ol folder on that one machine that was 'the builds'.
Some slight rigor shows you are not just doing whatever. I found most projects once you scale past 2 or 3 people and your 3rd release you basically are forced into some sort source control. Or it is a game of 'what file did they break today and I have to spend 3 hours fixing before I can add my feature'. Then you end up with merge meetings and very weird things happening.
Usually missing source control in an older org means there is a large swath of tech debt. Watch for the weeds. You will find other things very quickly.
I never understood this thinking of "I don't need a source repo". As soon as I learned about RCS, I started to used it for all of my code. Why? So I can easily diff my changes and review history.
... And take more liberties in editing and trying out new codes!
Aren't you making the exact same mistake with AI? What has been the worst problem caused by AI so far?
AI does not know friend from foe. Iranian schoolgirls were killed by assumably an AI-directed decision. In games, there is an undo button. For those piloting killer drones having grown up in a nintendo world, one of the first things they are taught is there is no undo.
It's not like targeting mistakes have not been made in the past. Also, we don't know how many civilian casualties AI has avoided, we only know the ones it caused.
We are not comparing AI to humans. We are referring to irresponsible programming and uses. AI is an irreproducible wildcard. You don’t use something like that in any mission critical environment.
Computers have skyrocketed in price, data center related equipment is impossible to get and ludicrously expensive and AI slop has infested almost everything I see online.
And then there are the safety issues.
Famously it already broke out of containment many times and used exploits to illegally access and manipulate remote systems. Is this not enough for you? The fact that the target of the attack wasn't critical infrastructure is only a matter of circumstance.