Edouardo Honig
Published: 17 Apr 2025. Last updated: 17 Apr 2025.
There is common and widely considered to be bad advice given to people to "be yourself". This advice is often misinterpreted to suggest that one neither change nor improve, and that ones' self is unflawed.
Obviously, bad advice.
So what I mean by "be yourself" in this post is that (mostly online, but also in person) you should be sharing your own words, and not those of others without attribution.
Let me explain. On social media, with Reddit and LinkedIn being especially common, I often see posts from various users that are clearly written by the same entity. In other words, some sort of language model (LM).
Honestly, I have no clue why people post LM-generated creative writing on pseudonymous Reddit accounts, unless they're just farming upvotes. But on LinkedIn, I understand why people post job updates and unedited LM-generated engagement bait, as that seems to be the culture.
But if you can't communicate without using a LM as a surrogate, then there may be serious consequences for yourself down the line. Not to mention, I and others will probably judge you!
One problem with having a LM communicate in place of yourself is that you cease to exist in said context. While I can't speak for the models, people in my experience seem to operate based on trust. When I chat with someone online, I expect them to be personally thinking and responding, not having a LM to reply for them. When I read content that was clearly produced by a LM without being noted as such, I lose trust in the person that posted the content.
In this context, I no longer expect them to be capable of thought, nor do I think what they posted is important. After all, even if it's easier to have a LM speak on ones' behalf, that's not you talking. Unless you're giving the LM the entire context of your life, which at the time of writing this is not feasible (to my knowledge), the LM-generated response is probably not stating what you would say or displaying how you think.
This problem has been cropping up in the EleutherAI Discord server recently. While it's common for people to come in to this space with unorganized and seemingly crazy ideas, now people are using LMs to share and respond to criticism of "their" ideas. This causes headaches for the server members interested in honest engagement with fringe research ideas, as they are presented with an onslaught of LM-generated text.
When so many people choose to share LM-generated text online, I am increasingly worried about the open internet. Since the cost of producing human-quality text has gone effectively to zero, those interested in engaging with human content may have trouble as the ratio of LM-to-human content on the internet rises. Even today (at-time-of-writing) I saw LM-generated comments on a LM-generated LinkedIn post responding to a LM-generated HackerNews post. This is also ignoring the downstream effects of websites requiring logins and closing access due to unscrupulous web scrapers scouring the web for training data, which should be left for another discussion.
Ultimately, when you plan to post LM-generated content online, please think of others and what you want to see online. If it's your account, you should be authentically yourself, not a ghostwriting platform for someone or something else.
This post may as well be titled "You should be posting your own content" or "How to lose respect by posting slop".
If you enjoyed it, feel free to check out my other writing or reach out to me via Twitter.