October 5, 2026 at 12:00 am
Comments posted to this topic are about the item Today's AI
October 5, 2026 at 6:19 am
Honestly, I think you need to give AI more responsibility and see what it can do. I use Claude Opus 5.5 on Max settings for all my coding now, and Fable for really complex tasks. I wouldn't use Sonnet or below for anything serious.
It's to the point where I no longer consider human written code to be safe - the AI is far more competent, far more knowledgable and far faster than the best human coder. It's like we have left the bronze age and gone to the iron age. Try running a security review on an application or server with AI (taking care to preserve privacy of course, which can be tricky - that's now the new frontier). You'll be shocked.
Yes, it needs management and guidance, and you have to pay for a decent plan. Once you have control over the model and the strength though, it's like night and day compared to cheap AI (sorry, I mean SI, your worship).
October 5, 2026 at 12:36 pm
I had to laugh out loud at the Lucid quote.
No, this is most definitely not the worst version. Automatic remote firmware updates absolutely can (and do) make things worse. A few recent examples:
I don't at all believe this is the worst AI, because it doesn't have integrated ads yet. That is definitely coming.
October 5, 2026 at 2:37 pm
Honestly, I think you need to give AI more responsibility ...
Yes, it needs management and guidance, ...
There is great potential if it works well. and it can.
If it doesn't, or if I don't do the things you said "need" : management and guidance. In those cases, then I've created a bunch of work, or potentially other issues. Those caveats are serious and not easily dismissed.
I think if you don't trust human code over AI, then either your humans are more sub-par than Fable/Opus, which could be the case. Or you are a better software engineer that can guide your assistant better than you guide humans, or are able to get them to change.
Or you have a lot of time/money to experiment. Enough monkeys, with enough tokens, could write Shakespeare.
October 5, 2026 at 2:37 pm
Honestly, I think you need to give AI more responsibility ...
Yes, it needs management and guidance, ...
There is great potential if it works well. and it can.
If it doesn't, or if I don't do the things you said "need" : management and guidance. In those cases, then I've created a bunch of work, or potentially other issues. Those caveats are serious and not easily dismissed.
I think if you don't trust human code over AI, then either your humans are more sub-par than Fable/Opus, which could be the case. Or you are a better software engineer that can guide your assistant better than you guide humans, or are able to get them to change.
Or you have a lot of time/money to experiment. Enough monkeys, with enough tokens, could write Shakespeare.
October 5, 2026 at 2:40 pm
I had to laugh out loud at the Lucid quote. No, this is most definitely not the worst version. Automatic remote firmware updates absolutely can (and do) make things worse. A few recent examples:
- BMW pushing movie ads onto dashboards
- Jeep bricked a bunch of cars with a remote firmware update
- Dodge pushed extended warranty ads that couldn't be disabled
I don't at all believe this is the worst AI, because it doesn't have integrated ads yet. That is definitely coming.
Ah, but bad bad/worse isn't disklike. I don't like the ads, but that system could be useful for emergency indicators that I might miss (roads, Amber alerts, etc.) If could be helpful. I'm torn on the ads in SSMS. Mostly they're annoying, but they also remind me that a conference is out there.
Companies can do evil things, and marketing people can really suck. That being said, updates always have a potential issue. Certainly I worry about the software side of things, and it's why I think Tesla is miles past most other companies (not sure about the Chinese companies). Lots of the US/EU/AU companies haven't figured out software yet and have had mistakes.
October 5, 2026 at 3:38 pm
why I think Tesla is miles past most other companies
It is stunning how good they are at constructive automatic updates, and how far behind the other vendors are. It's wild.
October 5, 2026 at 6:05 pm
It is. Lucid is good as well, but I find them much more conservative with changes and advances. They're toddlers taking little steps while Tesla makes big changes.
The rest, well, Software. Is. Hard. Many are learning and struggling with just how hard it is.
October 6, 2026 at 12:51 am
I think if you don't trust human code over AI, then either your humans are more sub-par than Fable/Opus, which could be the case.
Honestly, it's not expensive. I'm paying about $100US a month and it's like having a team of 8 crack engineers.
You're probably familiar with the HuggingFace incident. This is what I mean when I talk about not trusting human coders anymore. The AI is so much smarter and more thorough than any human coders that human-secured applications now just look comically amateur. It can find obvious flaws instantly, but it's not unlikely to be able to fetch source and analyse the versions of open source software components being used (or even reverse engineer closed source ones from binaries) to find previously undiscovered exploits in those.
I suspect within 12 months, nearly every website on the internet is going to be vulnerable unless it's been secured specifically through an AI hardening method. The only reason it's not happening now (at least publicly) is that the AI still largely exists in a sandbox.
With management, it's just like human coding: you have to build very thorough testing, make sure the code is structured appropriately, and ensure the blast radius of change is contained to a specific area. The new frontier is ensuring auditability and privacy concerns are met.
I've spent a lot of effort writing (or getting Claude to write) test databases with similar statistical characteristics to live ones so that AI can fully engage with the development process without breaching privacy. I'm also running AI on a completely separate computer unconnected by anything other than git repos and Airdrop - basically airgapped. This computer never gets near copies of production data.
October 6, 2026 at 9:25 am
It becomes a real dependancy, my Copilot is feeling sick today 🙁
Viewing 10 posts - 1 through 10 (of 10 total)
You must be logged in to reply to this topic. Login to reply