AI4Research: A Survey of Artificial Intelligence for Scientific Research
A comprehensive survey - a unified perspective on AI4Research. Specifically, the main contributions of work are as follows:
(1) Systematic taxonomy
(2) New frontiers
(3) Abundant resources
Project: ai-4-research.github.io
I'm writing with replit and I told it to write unit test, run them and fix stuff along they way. It lead to improvements in areas that I hadn't thoroughly tested (like check if an email address is well formed).
Did you check whether the 100 fixed ones made you happy? And 53 were completely new or were only able to appear after the 100? Oh, just as an experiment: did you tell it to write unit tests? And run them?
I'd scratch Keras and tensorflow and add pandas instead.
Yes absolutely. The important thing is to get the specs right and AI can help a lot.
It's actually smart, without good design you need to spend more time with cursor.
49 USD?
so, it's not just letting ChatGPT write the PRD / product requirements, but also architecture, frameworks and data flow? any key learnings you want to share?
depends on the tool you use? like in replit it's one click.
we need to define "validation". if you want to validate software functionality, you need to do that within the vibe coding IDE, if you want to validate with the user, vibe coding is the perfect tool (after the first phase), if you launch software without the user validation: it's your own fault.
nice, with visual framework did your tool use?
maybe use replit and expo?
repomix sounds like a great pre-processor. will definetly give it a spin.
I think that developing a PRD, esp defining the product in steps / blobs / units is the single most important success factor to keep the AI from derailing.
love the title!
until pen-testing becomes automated and is called via MCP
Vibe Coding: the interactive, language-driven creation of an app. Hype or must-have? A balanced review of the current state. assadollahi.medium.com/vibe-coding-... #vibecoding #agentic #LLMs
Same for me with all frontend frameworks
Yeah, I really like the "linguistic user interface" design. "Ah, I see the problem!"
hm, that's unfortunate. Maybe it's a matter of the right prompt.
It's a great prototyping tool, esp when you want to study user flows. And it's really fun for non-programmers.
For (professional) humans, that's true, but AI generated code seems to contain a lot of fluff, no?
On the other hand the providers compete and users will change to the most efficient one.
Did you ask it to write tests and execute them? Another idea: What would happen if you'd ask it to reduce the code base by 30%?
maybe vibe coding is not for software engineers, after all. :-D
Yes, the success of vibe coding lies in chunks of functionality you request from it. This means that you need to plan your features carefully, i.e. it's a training for PMs :-). At the same time, you get a functionaly prototype within 30min and you can test the flow with users quickly.
That's the state in Apr 2025, let's talk in Apr 2027 again.
I think vibe coding is currently great for product managers, for iterating concepts with users quickly.
What do you like best about it? And what are you comparing it to?
