with that said I have friends publishing 10+ papers before even staring, they are "basically" done before their first day. That could be bit easier with AI, but still more about other things.
doubt it. I'm a much more experienced coder than most PhD students. ~80% of what I do is coding. I vibe extensively. What happens is that you just move the goalpost. My projects are much more ambitious than they would have been without Claude Code, Gemini, and ChatGPT.
Tbh, if I had Claude Code, Gemini, and ChatGPT during my PhD, I’d probably have graduated in 1 year instead of 5.5 years.
My PhD was ~50% coding, 25% writing/polishing my papers, 25% reading others' papers. AI now accelerates each by at least 10×. Nothing will ever be the same.
@antigravity I just had a case were antigravity using Gemini 3 Pro (high) on its own did git add, git commit and git push and then proceeded to rebuild my project before I stopped it. That was unexpected.
I love Google’s AI efforts, but for coding I still feel Gemini 3 Pro is behind. It sometimes solves prompts that Chat/Codex 5 fails on, yet makes too many mistakes on “simple” tasks that Opus 4.5 and Chat nails.
vibing has definitely grown on me. With the Sonnet era it was useful. But I feel I can be a lot more lose with my prompts since gpt 5. Which is scary since this mean I review and rewrite less code.
@svpino@VivixLabs_HQ reminds me of the progress I experienced with Napster, at first you had to wait a long time to get at the music and then you could play it, but after broadband, the magical movement when it was possible to download faster than u could consume it
Google NotebookLM is also fantastic – especially the quiz mode. I'd be surprised if books on university course reading lists continue to sell as well going forward.
The new “Study and Learn” mode in ChatGPT is great!
If textbook authors and publishers felt threatened by copyright violations – they should be even more worried about this.
that would be total AI vibe coding – i.e. never looking at the code. But with you acting as the guardrails it's a fantastic productivity boost for a lot of tasks.
What has clicked for me so far is AI autocomplete, AI debugging, AI as faster docs, AI as Stack Overflow substitute, AI agents, AI agentic workflow.
What has not clicked: AI vibe coding
AI scaffolding is a bit 50/50.
Again, it's so odd to see React influencers dunk on innovative new projects that have limited reach until they give it to them.
People keep on telling me it is for clicks. But I'm not convinced. It takes guts to come out and say something so questionably ignorant in public.
It has been a ridiculous day. 4 times Opus was completely baffled in a way that felt off by Claude standards. I pasted the exact same thing to Codex for it to one shot the task. Sure, high reasoning is slow but failing is worse.