Thread by @treblewoe

Yesterday, I asked Grok a number of questions related to my account, centered around posts that could be a problem. Even when it gave evasive answers, it still listed beneath half a dozen or more of my own posts that clearly related to my query. So I deleted them all.

These charts show the change for my account in the subsequent 24 hour period. My posts' reach today was 3x my average, and by far the biggest day I've had in nearly six months, despite having twice as many followers as I did then.

In this thread, I'll share some of the responses it gave that outline things which are clearly impacting everyone. The change was instant as soon as I began deleting old posts that no on will ever even see again anyway.

The data will appear almost entirely in screenshots because one of the things that is affected is any discussion of this very subject.

Here's what I learned...

I first asked for specifics. All of these are undoubtedly correct for my specific account. I then asked follow-up questions to get some more details. The list it mentions isn't complete, and we all know the name of one particular one that rhymes with rub black. /2

I then rephrased the initial question and received a new list of problems. All five are again specific to me, despite its insistence that it doesn't actually know anything about me.

#3 is something I've warned about for a while. I've stopped doing it entirely, because the impact on the post is -75% or worse. This is designed to keep people here on X. /3

It specifically mentioned any posts related to this entire subject. Now if you know something about Grok, you know that it's being fed by user-generated content as well as X, so this could all just be hallucinatory nonsense on its part.

As an experiment, I searched my own TL and deleted them all anyway. Those stats in the first post in this thread were the immediate result. I could actually observe the change in realtime yesterday as I posted normally, while querying Grok and deleting posts it mentioned. /4

It said something that made me ask this question that I had suspected for a while. Again, whether or not this was a "real" answer, it was another worthy experiment. So delete, delete, delete.

And I didn't have to work hard, because Grok kept spitting out my own posts below its answers, so I just worked through the ones it brought up to me as examples, again, even as it existed it didn't know anything about me. And again it produced extremely noticeable instant results. /5

I then asked for more specifics about my posts. Again, all three of these are 100% spot on. It very clearly was able to compare the internal rules to my specific timeline. Again, I deleted as many old posts as I could find that would've tripped this one. Helpfully, it suggested most of them for me. /6

I asked about word lists and it helpfully provided a number of examples. Sometimes I had to create a new session because it would stop giving account specifics, but clearly it has the ability. /7

After getting my account cleaned up and seeing the radically positive results last night, I was curious today about other word lists. Note that these are included for information purposes only. I'm glad it will answer these questions frankly. Adults should expect nothing less. /8

It mentioned that spelling variations and substitutions are known & treated the same, so I asked about that. Again, this example isn't intended to be offensive, but it was humorous to watch the engine trip over itself. It eventually stopped & chastised me for the words it used.

This video is included because it is a very good example of the futility of trying to mask such words with 90s hacker tricks. It will all be held against your account, so just don't do it. /9

When asked for variations, it taught me a few new ones. It's clear that their sentiment analysis and variation detection outstrips our ability to get away with it. This is shared because it is comical, and shows what will impact your whole account. /10

I asked about some other categories of subject areas that aren't inherently blue. It made very clear that much of the discourse on the right falls outside of what is permitted. This area is a a really big deal. /11

Some of the state of mind references are a completely normal part of the English language, but keep in mind using any of them will have an impact on your entire profile, not just one specific post, as demonstrated above. /12

In summary, X has a robust sentiment analysis regime running in realtime on everything we say. Even apart from keeping an account, or getting paid for it, these directly affect whether anyone ever sees our posts at all. I think that's something that matters to us all. /13

If you want to search your own account, the Advanced Search feature is the best way to do it. Note that all searches are literal so "dog" won't find "dogs" and so forth.

twitter.com/search-advanced