Akul Mehta
akulmehta.bsky.social
Akul Mehta
@akulmehta.bsky.social
Glycoscientist @theNCFG BIDMC-Harvard Medical School
👨‍💻👨‍🔬⚽📷
(Views are my own)
💯
October 12, 2025 at 11:45 AM
To be honest all places are quite awful right now, for different reasons.
October 11, 2025 at 9:13 PM
This is remarkable and obviously fantastic science. I have always wondered though, wouldn't the transplanted organ just start expressing the blood group antigens eventually in the recipient? And therefore making this a temporary solution rather than a permanent one.
October 8, 2025 at 8:02 PM
5. Abrupt end of the review as the AI probably ran out of tokens.
September 16, 2025 at 7:19 PM
4. Slowly over the course of the review, AI starts to want to rewrite your entire manuscript for you "I moved the statistical significance sentence up because it is a natural follow-up of the sentence …" The current state of AI, it likes to generate text not to be critical of it.
September 16, 2025 at 7:19 PM
Example:
"-line XX: … Whats does it mean to study well? Properly? Using right methods? Comprehensively, covering a wide range of possible candidates?"
September 16, 2025 at 7:19 PM
For context, several of the authors on that paper were native English speakers and all of them had proof-read the paper. But AI used this as an excuse to improve the language, even where it wasn't needed. 40% of the review was just this.
September 16, 2025 at 7:19 PM
3. AI assumes you are a non-native English speaker even if you are. It will try to rephrase everything you say just to make it sound better. In my colleagues review it stated "…. (obviously not proof-read by a native speaker) …"
September 16, 2025 at 7:19 PM
Another example it cites Wikipedia "Elution - Wikipedia Please be aware of the difference between an eluate and an eluent" or other web-based resources often instead of actual journal articles. This is because LLMs have access to web-based resources for training but not as much pay-walled articles.
September 16, 2025 at 7:19 PM
2. AI cites very strange evidences for their statements. In my colleagues review they stated the following "As I checked in Scopus, the authors have an expertise on the biology and biochemistry of … , so I expected to read a sound work."
September 16, 2025 at 7:19 PM
Another example from the review "I have checked the Supplemental table in excel and I saw XXXXX and XXXXX in rows XX and XX, respectively." Newer reasoning models like to “show their work” and hence blurt out such text.
September 16, 2025 at 7:19 PM