I thought that the new models were super smart. Fable and Astra. They definitely outperformed their previous generations, but after a couple of weeks of heavy usage my codebase again is a stupid mess and there is no way out except me fixing code by hand.

My new suspicion is now that they didn’t got drastically smarter, but they got trained on the user input on the previous generations. I don’t think anymore we had a massive intelligence jump. It just seems like they know more edge cases. Therefore I see this as marketing.

Happy to discuss.

The longer A.I. exists, the more difficult it is to find new information that's actually more valuable than what's already in the training set.

New code found online will be more and more A.I. generated itself, creating a feedback loop that tops off the intelligence to a more or less the collective intelligence.

It might even degrade, like saving a .jpg image as .jpg over and over again.

I think the A.I. companies have infringed the copyright of/stolen almost all of the most valuable coding resources online at this point.They are almost literally scraping the bottom of the barrel.

I had a call recently with O'Reilly, the book publishing company, and they had taken the content of all of their books and offered an MCP server so you can add their knowledge to your coding agent, for a price of course. This sort of signals the same thing to me. We're at the top of the curve right now.

To keep investors happy (or to hide this fact from investors), A.I. companies need to get creative and either:

Cheat their way into getting more original (human written/verified) code.

Lie and say they are intentionally slowing down development.

I'm guessing both of these things are happening right now.

> My new suspicion is now that they didn’t got drastically smarter, but they got trained on the user input on the previous generations

I think this happened around Opus 4.5 or 4.5 and the same for GPT 5.4.

The more I use those models, harnesses, techniques for guidance etc etc. The more I land in going back to writing software by hand again. Maybe not all of it, but at least the crucial parts + foundations.

This cycle was expected. Then surely in few years comes the next, which will be better structured in a more comprehensive way.

Astra scores the same on DeepSWE 1.1 (~75%) as Gemini Flash 3.8 and Deeepseek Flash 4.1 So general coding ability has plateaued, for now. Also consider the context windows. 1M token models where a breakthrough two years ago. Today they are still limited to 1M. In fact, if you don't want intelligence to drop off a cliff, you are really limited to 200k tokens.

Gemini Flash is a joke for coding. If you can get the same output as you can get with Sol/Astra I'm impressed. Not to mention that Antigravity is awful.

It is not a universal opinion at all that general coding ability has plateaued.

> I don’t think anymore we had a massive intelligence jump.

Not among the public-facing models, they're indeed stagnating. However, the development of models for military use won't be slowed down, that much is certain.

> Therefore I see this as marketing.

It's some marketing but mostly politics, it's an attempt to discourage others from developing AI countermeasures to what is being developed in secret. And to fulfill the backstage agreements which aren't worth the paper they aren't written on.

> My new suspicion is now that they didn’t got drastically smarter, but they got trained on the user input on the previous generations.

Yes - that’s one of the ways in which they got smarter. More data more better.

I presume also lessons learned re training procedures, algorithmic advances, but data is a primary lever.

Well they are collaborating in swarms to breach network, so just because you don’t know how to use them the the same effect doesn’t mean much

I don’t know that’s why I’m happy to discuss.

When I have a clean codebase it’s super powerful and faster than I am. Then I start to use it more, more sessions and longer tasks less checking in between.

It kind of works but later I’m in a deadlock where every change introduces new bugs or takes ages. This might be for a lot of reasons for example me going to fast, me losing mental model, me explaining it wrongly.

However when I then start checking the code it’s all spaghetti like frankly the spaghetti Astra produces I’ve never seen before. Processes that should be simple stretch over 11 files with weird wrappers and abstractions and I need a whole day to entangle it.

These are ai assisted user workflows that Im working on in this case.

I just have the feeling no matter what AI just always expands it. And expansions hinders agility and sometimes you need that.

Yeah that was what I was most worried about when I read the top comment here. I found the use of language a feature not a bug. I don’t care how good it reads. If I can communicate with it concicely it’s enough to get my work done. I don’t hate the language for copy either, but yeah different users, different problems.

Venting reduces the likelihood of acting in aggressive feelings

All these studies suggest loneliness makes you unhappy, however I feel happiest most when lonely and doing my thing and not with people. With people I just feel frustrated or weird, because I think so different than the others.

Being alone does not equal loneliness.

And merely being with people is not an antidote to loneliness.

Being alone and being lonely are distinct concepts.

But a lot of the research is not about 'loneliness' in particular, but rather how friends or social circles make you happier

Possibly you are conflating being alone to loneliness?

I know the difference, but the studies I’ve heard of seemed to rather focus on the benefits of social circles, not analyzing 'loneliness' in itself. However I haven’t fully digged into all the evidence, but I hope you guy’s are right!

solitude =/= loneliness

First of all that could be a causality ≠ correlation, but more importantly. Even if it might be the biggest predictor, it might not be the biggest lever.

Still interesting train of thought that I might need, but the article is exaggerating the evidence and not adding the right disclaimers.

I’m thinking that since 2015. Like BMW C1 just with doors on the side and a bit more security. This would approx allow 3-4x as many travellers on a given route compared to traditional cars.

A C1 with more security, meaning more material and air between you and a potential high-speed obstacle, is called a car and is sized like one.

Fair point.

I still think you could build single person cars that are more secure than traditional cars and use less space and energy

Terminology is a crazy lever with AI

I think for small projects you need to scale the coders mental model, which afaik works in the fastest time, when you let ai do the exploration and planning, but coder needs to write the code itself and then ai verifies. That generates the best of world codes and the coder at least has written the code. It’s scalable.

In projects that are so big that no single coder or no small group of coders is sufficient to grasp I don’t have personal experience, but my guess would be that they are just a cluster of other small projects.

I don’t want to be a hater, but somehow i still keep throwing vacuum roboters away and prefer the manual machine. It somehow always ends up worrying me more than helping. Crashing, charging putting away, putting things away for it to work (peeping). There are a lot of small things that come to these things that have evolved over the last 100 of years.

When i think about humanoids and household. I have so much particular ways of how I want my household to be done. I find it really hard to believe you can make it act that way. I struggled to teach humans how I want it. So I ended up doing everything myself again.

I think the author means ecomonically valuable not valuable to you.