3 ms·

I can tell you one impact -- there are huge amounts of ChatGPT helped, or entirely written, junk papers appearing at conferences, which is causing serious probl

I can tell you one impact -- there are huge amounts of ChatGPT helped, or entirely written, junk papers appearing at conferences, which is causing serious problems getting reviewers. Some large events have introduced a pre-review phase to filter out junk, but it's getting harder, as LLMs are good at producing plausable looking nonsense. Honestly, I'm not sure of the long term solution here. Conferences and journals may have to introduce a system where you need an existing member of the community to "vouch" for your work to be allowed to submit (they don't have to say it should be published, just worth reviewing).

fastneutron 3y ago
LLM-written content has a certain gestalt to it that’s pretty easy to sniff out once you’ve seen it a few times, just look at Quora and LinkedIn these days. While I think it’s probably fair for people with English as a second language to use LLMs as a writing assistant (for example), there definitely needs to be some kind of author disclosure statement at the very least. This could be similar to how more papers are now including contribution statements when there are many coauthors.
- corethree 3y ago
  That gestalt can easily be smoothed out. You simply need to prompt LLM to do so in your query. Tell the LLM to depart from its default style. Say make your response more brief, well written and add a witty conclusion at the end Then you say rewrite the final paragraph to be slightly shorter and incorporate a certain example. Yada yada. Likely you can even do this "write your response in a style that is not typical for an LLM"
  - throwup238 3y ago
    It’s easier to illustrate with prose: ask it to write in the style of Terry Pratchett or Hunter S Thompson and the style changes drastically. You can even ask it to rewrite your scientific paper in the style of Shakespeare: https://chat.openai.com/share/498804da-6d59-4a0c-91d1-da4cea38c261 https://chat.openai.com/share/498804da-6d59-4a0c-91d1-da4cea...

fastneutron 3y ago

LLM-written content has a certain gestalt to it that’s pretty easy to sniff out once you’ve seen it a few times, just look at Quora and LinkedIn these days. While I think it’s probably fair for people with English as a second language to use LLMs as a writing assistant (for example), there definitely needs to be some kind of author disclosure statement at the very least. This could be similar to how more papers are now including contribution statements when there are many coauthors.

corethree 3y ago
That gestalt can easily be smoothed out. You simply need to prompt LLM to do so in your query. Tell the LLM to depart from its default style. Say make your response more brief, well written and add a witty conclusion at the end Then you say rewrite the final paragraph to be slightly shorter and incorporate a certain example. Yada yada. Likely you can even do this "write your response in a style that is not typical for an LLM"
- throwup238 3y ago
  It’s easier to illustrate with prose: ask it to write in the style of Terry Pratchett or Hunter S Thompson and the style changes drastically. You can even ask it to rewrite your scientific paper in the style of Shakespeare: https://chat.openai.com/share/498804da-6d59-4a0c-91d1-da4cea38c261 https://chat.openai.com/share/498804da-6d59-4a0c-91d1-da4cea...