Honestly, I hardly even read papers anymore. Text is littered with too much junk and incoherent garbage. At most I'll sometimes skim the results, and key figures.
If a paper (usually via 𝕏 post) piques my interest, I'll go directly to the NCBI Geo dataset, and interrogate the metadata and raw data myself. If the data is coherent and well organized, then I'll spend time breaking the paper down with AI, and even reanalyze the data to make sure the conclusions are accurate/truthful. Possible to do this now with Codex GPT 5.5/Claude Opus 4.7.
If the metadata is poorly organized or the raw data is messy and incoherent, I'll skip the paper entirely.
Reality is, most papers are written for the sake of CV-boosting instead of reflecting truth. This'll probably get worse in the future. Only way to capture truth in the future will likely be to directly interrogate and re-analyze raw data oneself with AI.