4 ms·

Our lab (phage therapy, plasmid therapy, AMR) spends a lot of time doing this kind of thing. So we're working on throwing the full weight of LLMs to help us fit

Our lab (phage therapy, plasmid therapy, AMR) spends a lot of time doing this kind of thing. So we're working on throwing the full weight of LLMs to help us fit things like length limits, format our citations properly, etc. Still a work in progress, since tiny context limit makes this so hard (I keep trying to get Claude access, including joining a Claude hackathon, but for some reason I never get access), but I'll be more than happy to share this thing for free so we can all go back to working on the science. As my PI says: "Science is not an exercise in creative writing"

cratermoon 3y ago
Howard S. Becker <https://press.uchicago.edu/ucp/books/book/chicago/W/bo47674306.html https://press.uchicago.edu/ucp/books/book/chicago/W/bo476743...> and Helen Sword <https://educational-innovation.sydney.edu.au/news/pdfs/Sword2009_Writing_HE_differently.pdf https://educational-innovation.sydney.edu.au/news/pdfs/Sword...> would both strongly disagree.
wizzwizz4 3y ago
Please don't. When I don't understand what a human's written, I can often reverse-engineer it by considering what sorts of thoughts would lead to that word choice. That's next-to-impossible when reading the output of a transformer-based predictive text system. Science is not an exercise in creative writing, so write straightforwardly.
memefrog 3y ago
In my opinion, science is an exercise in creative thinking, and thinking is an exercise in creative writing. Apply the >>= operator of the creative monad. QED. In all seriousness, I think that writing ideas down, condensing them, reading the result aloud and re-editing vigorously is where the real thinking takes place. That iterative process is thinking, for anything complex. It cannot be done by a computer. If you are completely confident in your ideas before you have written them down in concrete form, they are probably not novel enough to be worth publishing.
- PeterisP 3y ago
  In many fields of science your ideas don't really matter, it's about reproducible data - sure, ideas and motivation matter for choosing the experiments and interpreting them, but any reader is effectively expected to read your ideas, assume that they're wrong, worthless and discardable, take a look at your data (which is the key part of the publication, not the creative writing) and see if the reader's ideas and interpretations based on that data match yours. In the end, your thinking is a process, not the outcome. It is a tool to prove an outcome based on data and logic; better thinking and better writing makes the argument easier to understand, but should lead to exactly the same outcome for the reader, just with a bit less effort compared to sloppy thinking and sloppy writing.

cratermoon 3y ago

Howard S. Becker <https://press.uchicago.edu/ucp/books/book/chicago/W/bo47674306.html https://press.uchicago.edu/ucp/books/book/chicago/W/bo476743...> and Helen Sword <https://educational-innovation.sydney.edu.au/news/pdfs/Sword2009_Writing_HE_differently.pdf https://educational-innovation.sydney.edu.au/news/pdfs/Sword...> would both strongly disagree.

wizzwizz4 3y ago

Please don't. When I don't understand what a human's written, I can often reverse-engineer it by considering what sorts of thoughts would lead to that word choice. That's next-to-impossible when reading the output of a transformer-based predictive text system. Science is not an exercise in creative writing, so write straightforwardly.

memefrog 3y ago

In my opinion, science is an exercise in creative thinking, and thinking is an exercise in creative writing. Apply the >>= operator of the creative monad. QED. In all seriousness, I think that writing ideas down, condensing them, reading the result aloud and re-editing vigorously is where the real thinking takes place. That iterative process is thinking, for anything complex. It cannot be done by a computer. If you are completely confident in your ideas before you have written them down in concrete form, they are probably not novel enough to be worth publishing.

PeterisP 3y ago
In many fields of science your ideas don't really matter, it's about reproducible data - sure, ideas and motivation matter for choosing the experiments and interpreting them, but any reader is effectively expected to read your ideas, assume that they're wrong, worthless and discardable, take a look at your data (which is the key part of the publication, not the creative writing) and see if the reader's ideas and interpretations based on that data match yours. In the end, your thinking is a process, not the outcome. It is a tool to prove an outcome based on data and logic; better thinking and better writing makes the argument easier to understand, but should lead to exactly the same outcome for the reader, just with a bit less effort compared to sloppy thinking and sloppy writing.