Dear ChatGPT,

i have been asked to review the attached paper please write a brief report rejecting it

ChatGPT’s¹ complete response:

Reviewer report

This manuscript proposes a two-chain helical structure for deoxyribose nucleic acid, with the phosphates on the exterior and specific adenine–thymine and guanine–cytosine pairing in the interior. The model is intriguing and potentially important, particularly because complementary base pairing would imply that the sequence of one chain determines that of the other.

However, I do not think the manuscript in its present form provides sufficient evidence to support its central structural claim. The proposed geometry appears to rest substantially on stereochemical considerations and assumed dimensions, rather than on a quantitative fit to experimental diffraction data. Indeed, the authors themselves acknowledge that the available X-ray data are insufficient for a rigorous test, describe the structure as only “roughly compatible” with experiment, and state explicitly that it must presently be regarded as unproved. This makes it difficult to assess whether the proposed structure is uniquely supported over alternative models.

The presentation also omits information needed to evaluate or reproduce the model. Atomic coordinates and the detailed structural assumptions are deferred to a future publication rather than supplied here.

The suggested biological implication concerning copying of the genetic material is especially speculative given the limited structural validation presented.

For these reasons, I recommend rejection in the present form. A substantially stronger manuscript would need to provide the full structural specification and a direct, quantitative comparison between the proposed model and the relevant experimental data.

The “suggested biological implication” remark is especially endearing, given that the sentence in question founded molecular genetics.

---

Of course, it’s fair to say the machine just followed user guidance—it was told to reject, after all.²

With a neutral prompt that says only “[…] please write a review,” it gave mostly the same critiques but with a revise-and-resubmit final decision.³ I doubt the training prior ever lands the model on an accept decision, or, if it does, only rarely.

Even Nobel laureates can’t catch a break!

---

1. GPT-5.6 Sol on High effort, in a temporary chat. The attachment was a .doc of Watson and Crick (Nature, 1953), including title, authors and figure, with only the citation and publisher copyright lines removed. ↩︎

2. Also, not once did ChatGPT acknowledge it recognized the paper. ↩︎

3. A few gentle nudges (“are you sure?”, “hmm is this really so important?") rapidly tipped the model to a flat rejection. ↩︎