The AI crash is impossible - Change my View

[Joshe]

1 hour ago, Leo Gura said:

I don't care about code quality, I only care that it works.

Your AI slop will not work in live business use cases. Your AI slop could not even handle this forum traffic without corrupting the forum database and destroying itself.

One error in writing to the database would destroy this forum.

Yes, dropping agents into an existing (brownfield) production codebase is very risky, but a lot of that risk can be mitigated with good strategy. Getting agents up and running in a brownfield codebase is largely a project management and system design problem that requires lots of iteration and creativity. The problem is largely solvable, you just have to figure it out. 

You could instruct CC to write comprehensive test coverage for every database operation. Before anything touches a production db, you'd have a test suite confirming everything works right in staging. That's just one guardrail you could build into your agentic workflow. 

Edited 3 hours ago by Joshe

[Lyubov]

As someone who uses AI in my work, it sure does make a ton of ridiculous errors and some truly odd reasonings. It almost is like a toddler you need to constantly coax and hold it’s hand, it gets exhausting, I don’t feel like I’m working less or am so much more productive. Maybe I am marginally more efficient? This shit is NOT replacing us en masse anytime soon. 

Edited 3 hours ago by Lyubov

[Lyubov]

The people hyping AI the most are effectively unemployed people who talk about it online a lot and the tech crowd believing their own hype and propaganda, they raise funds with their buddies and the dopamine rush they get from that blinds them of the truth.  

I can’t say I’ve met anyone with a real job or anyone actually employed having a substantial portion of their work taken over by the AI. They are burning a fuck ton of fresh water just to write their emails a bit faster. 

All those layoffs are speculation and because we are in a quiet recession on par with 2008. It’s not because AI has actually replaced anyone. 

Edited 3 hours ago by Lyubov

[Peo]

16 minutes ago, Lyubov said:

As someone who uses AI in my work, it sure does make a ton of ridiculous errors and some truly odd reasonings. It almost is like a toddler you need to constantly coax and hold it’s hand, it gets exhausting, I don’t feel like I’m working less or am so much more productive. Maybe I am marginally more efficient? This shit is NOT replacing us en masse anytime soon. 

True, but it's accuracy is developing rapidly. It gets better over a few months rather then years. AI is accelerating. Just compare some AI image generation from 2023 to 2026. It gets a lot more realistic and makes a less anatomy mistakes. 

[Lyubov]

1 hour ago, Peo said:

True, but it's accuracy is developing rapidly. It gets better over a few months rather then years. AI is accelerating. Just compare some AI image generation from 2023 to 2026. It gets a lot more realistic and makes a less anatomy mistakes. 

Does it though (develop rapidly)? How does that actually pan out in relation to all the real world promises made? I’ve heard this before about how AI advances fast but it seems the only metric often cited is AI slop stuff like this. Yes it has gotten better at generating clips you can make funny videos from. Replacing people and radically changing all sorts of industries on a scale never before seen? I’m not holding my breath. I have no problem eating crow if I’m wrong. I’m calling the bluff now. It’s a bubble and there’s a huge wake up call coming. 

Edited 1 hour ago by Lyubov

[RamBam]

@Leo Gura You are massively underestimating how much AI will transform society. 

Game Dev is some of the most complicated code. Nobody is vibe coding multithreaded pipelined cpp. You can't just go "Claude re-write MongoDB in Rust, make no mistakes." There is certainly a lot of work AI cannot do. But only 10% of CS people work on anything of substance, and that is being generous. Most of the industry is grinding out low priority Jiras in React or Java, things AI can already do quite well. Human work in general is, for the most part, highly dull and repetitive. 

For example I think you are wrong that AI couldn't replicate this forum. What is so complicated about it? There are ~1M total posts. Even at 10M reads & 1M writes monthly it would be possible to run everything on a cloud hosted DB for under 10$ a month. Everything is in English so you probably need only 3 deployments for US, UK, AUS.  There is no complicated nested reply structure. Its a CRUD web page that has fairly low traffic. I could easily vibe code a forum with a Firebase backend over the weekend. 

I think it is possible that valuations will collapse. But this would be because Wall Street got detached from reality and not because the tech is not transformative. Same thing happened with the internet. 

AI is already turbo cracked at Math+CS. There are maybe 100 competitive programmers better than AI; you will see a Deep Blue moment where AI becomes #1 in competitive programming and IMO. I challenge you to post any math or programming problem that AI cannot solve. Yes these are compartmentalized repetitive problems in well documented fields, still impressive. AI can translate in real time, generate custom images and video, search 1000 page documents for relevant information, work with applications such as PowerPoint or Excel. Massive funding is flowing into robotics. Expect to see self driving, automated shipping ports, Amazon fulfillment centers completely automated, aspects of food preparation and farming automated, AI powered diagnostics and surgery, military drones and drone defense with no human in the loop. There is going to be massive disruption to the work force and societal structure in general.  

Even in research AI is showing some promise. For example Donald Knuth, who is a CS legend and Stanford professor, published a recent paper "Claude Cycles". This is from the abstract: "Shock! Shock! I learned yesterday that an open problem I’d been working on for several weeks had just been solved by Claude Opus 4.6— Anthropic’s hybrid reasoning model that had been released three weeks earlier! It seems that I’ll have to revise my opinions about “generative AI” one of these days."

Edited 8 minutes ago by RamBam