One Researcher's Plan to Fix AI's Broken Review System

You might think scientific papers are carefully checked before they're published. In AI research, that system is clogged. There are too many papers, many are low-quality or even AI-generated, and reviewers often don't really understand the work. The result? A lot of meaningless research that still gets published and cited.

One researcher argues we should replace the slow, gatekept conference system with a continuous online platform. Papers would get a simple 1-to-10 score, like a rating for a movie or product. Reviews from people with proven expertise would count more. This would push researchers to care about quality, not just churning out papers.

The clever part: AI itself could help. Large language models are good enough to check whether claims in a paper are actually reproducible. That would catch mistakes early and speed everything up. Instead of waiting for a conference date twice a year, work could be posted and reviewed any time.

It's not a finished plan, and it faces big obstacles — like academic careers still tied to famous conferences. But the core idea is simple: make research more accountable, faster, and easier to judge. If that happens, you might get better AI products and fewer inflated headlines.