Reviewer fatigue in the coding-agent era: "everything is possible, nothing is interesting"
rishabh16_ · x · 2026-11-18
A researcher reviewing for several venues describes a silent "reviewer fatigue" spreading among peers: barely any research seems interesting anymore because everything is technically possible with coding agents, while the truly magical work gets buried.
Key points:
- Reviewers subconsciously reward papers showing genuine "struggle" over ones that look like they just threw agents at a problem; what should matter is first-principles problem-solving and pushing the frontier—new insights, new failure modes
- Most submissions look like "N things glued together," not grounded in the underlying science (e.g., bio/chem) in modeling assumptions or task selection—just remixes of prior methods
- Each cycle he hunts for something genuinely new, not another benchmark-maxxing run pushing scores on protein/molecule design with the latest XYZ diffusion/flow/LLM model ("do we really need another one?!")
- Deep down, he's looking for evidence that humans still "have the dawg in them" to think deeply about problems
He offers no resolution, just voicing it for the latest ICLR round in case others feel it too.
More from AGI Musings
- 5 Major Outages in 6 Days: Are Swarms of AI Agents Behind Global System Failures? — 4KTV · 2026-11-18
- The fun ASI scenario: a tiny recursive self-improving model copies itself into everyone's genie — Bam4d · 2026-11-18
- Google expands Gemini Tasks free to hundreds of millions, adding agent calls and AP2 payments — sundarpichai · 2026-11-18
- Researcher laments reviewer fatigue as coding agents flood academia with incremental papers — rishabh16_ · 2026-11-18
- "No bubble—enterprises just adopted AI slower than expected": 100x willingness to pay for tokens — gabriel1 · 2026-11-18
- AI agents uncover first known genome-wide frameshifting in bacteria — GaryMarcus · 2026-11-18