One key reason you’re wrong is that many interesting things aren’t even getting published, they’re on the DL for years and eventually make it to public spheres and products.

Academia is just a daycare at this point, and many labs shouldn’t exists or get funding. The people who move the field aren’t necessarily the ones with the most citations, they’re usually hard at work in places that don’t publish at all.

Are you talking about just frontier LLM agent stuff or all of the scope of ICML? I wonder what your subfield is.

Because as someone said in the thread that in two years the LLM will just read the docs and produce perfect Bend code.

That’s just sad, makes me question what’s the point of writing software these days.

I was in a programming class when ChatGPT/CoPilot first came out. I hadn’t started using it yet for classes because I was under the impression that “my work should be my own”. I was the only one in the class who would get 80+ average on quizzes, everyone else got nearly perfect scores. Oh well.

Sounds like you took programming to learn programming while the others took it for a certificate.

I had similar issues for different reasons at university. Some of the subjects I learnt were extremely boring to me and I just didn't focus on them, while others I obsessed about. I learnt the things I wanted to learn, but didn't get the grade I probably could have if that was what I was optimising for.

Something else is afoot in the markets, I wouldn’t take rando tweets at face value, especially if they’re confirming a narrative you’re biased to accept.

https://www.citadelsecurities.com/news-and-insights/2026-glo...

The problem is that inefficient systems will cost even more as you scale their use, but gains from such systems are not guaranteed, and profits even less so.

How much are you spending before you even see a $1 of revenue?

Nice tool, but the agentic workflow doesn’t sound cost efficient.

Fair enough, I spend a maximum of 200 a month given I use the MAX plan from CC. I don't find myself every hitting the weekly limits - but recently I've gotten close!

This is marketing, not reality. The only “job destruction detector” anyone needs to keep track of is non-farm payrolls.

The best way is to publish papers and go to conferences

Thanks for the input! Conferences are definitely on my radar (like the ASME or MRS fall meetings). However, as a solo, bootstrapped software founder, the 6-12 month peer-review cycle for publishing an article on my own is a bit too slow for my runway. That’s exactly why I'm hoping to partner with an existing lab—I bring the fast-moving AI architecture, and they bring the publishing pedigree and physics validation. Have you seen any founders successfully bridge that gap without having to spend a year writing papers first?

There were always multiple crises in the world at any one time. We’re just aware of them happening concurrently now because of the internet and global connectivity.

These types of frameworks never resolve the core problem with agents, which is they don’t really think and so are prone to getting stuck in infinite loops even if they’re wrong. I haven’t used this framework, but my guess is that Devil’s Advocate will be the most prone to this problem. But who knows.