I recently discovered that some popular federated instances have been using LLM-assisted moderation tooling that evaluates whether someone has said something bannable. They do this by running a script/app that sends the user’s comment history to OpenAI with the question “analyze this content for evidence of specific political ideology sentiment. Also identify any related political ideology tropes“. (The italic bits are where I've redacted the ideology they're seeking).
OpenAI’s LLM (they’re using GPT-5.3-mini) then responds with something like:
and so on, hundreds of comments.
I have not named the instances or people involved, to give them time to consider the results of this discussion, make any corrective changes they want and disclose their practices at their own pace and in their own way. I have also redacted the evidence to avoid personal attacks and dogpiling. Let’s focus on the system, not the individuals involved. Today these instances and people are using it and maybe we’re ok with that because it’s being used by groups we agree with but what if people we strongly disagree with used it on their instances tomorrow?
The use and existence of this tooling raises a lot of other questions too.
What are the risks? Fedi moderators are often unsupervised, untrained volunteers and these are powerful tools.
What safeguards do we need?
Would asking a LLM “please evaluate this person’s political opinions” give different results than “find evidence we can use to ban them” (as used in the cases I’ve seen)?
What are our transparency expectations?
Is this acceptable and normal?
Should this tooling be disclosed? (it was not – should it have been?)
If you were given a choice, would you have opted out of it?
Can we opt out?
Are there GDPR implications? Privacy implications? Should these tools be described in a privacy policy?
Are private messages being scanned and sent to OpenAI?
How long should these assessments be retained and can we request to see it, or ask for it to be deleted?
Once the user’s comments are sent to OpenAI, is it used to train their models?
What will the effect be on our discourse and culture if people know they are being politically profiled?
Where are the lines between normal moderation assistance tools, political profiling and opaque 3rd-party data processing?
I hope that by chewing over these questions we can begin to establish some norms and expectations around this technology. The fediverse doesn’t have any centralized enforcement so we need discussions like this to develop an awareness of what people want in terms of disclosure, privacy, consent and acceptable use. Then people can make choices about which instances they join and which ones they interact with remotely.
And of course there are the other issues with LLMs relating to environmental sustainability, erosion of worker’s rights, increasing the cost of living and on and on. I can’t see PieFed adding any functionality like this anytime soon. But it’s happening out there anyway so now we need to talk about it.
What do you make of this?
Wthout going into the issue itself, it is such a ridiculous waste to use an llm for something that a far simpler model could do like 100x faster and locally for essentially free...
Just search for "machine learning text moderation" and you will find all kinds of options. Not to talk about the fact that a simple 4B LLM could do this as well.
One thing I really hate is how LLMs have completely overshadowed the entire ML/AI field and people just use them for everything.
Using a trillion parameter LLM model for basic text moderation is like using a gaming rig to play candy crush.
I've been doing a lot of work around this area and the issue tends to pop up around context. There are instances a haiku model will catch something far more accurately than gpt-oss-safeguard. Obviously you pay in costs, realistically, you need a full system here for a proper implementation so some flows to tiny LLM some flows exclusively ML, some flows to a higher intelligence.
People on the fediverse are generally pretty anti ai but it's basically impossible to scale a platform without AI moderation. I would fully welcome any instance trying an implementation of [Osprey](https://discord.com/blog/osprey-open-sourcing-our-rule-engine with LLMs.
I want to train a local ML system to recognize my personal handwriting. Is that possible?
Yes, a Convolutional Neural Net could do it or a plain Neural Net even.
You'll want to create a sample of your handwriting with the letters isolated into their own picture and labelled. Maybe 10 of each letter to start with? If you get bad results with that make more samples. Each picture should be the same size as all the others. The starter course of Machine Learning I took way back when had us using a database of labeled numbers, each picture was 10 pixels by 10 pixels.
Then pick a CNN model (or better yet several) and train them on your handwriting. You can find some here: https://huggingface.co/models?other=CNN
Pick the one that does best. As part of that course I mentioned, I created an evolutionary algorithm to mutate, combine, and propagate CNNs to find out the best configurations for identifying images. The ones that performed the best got to combine with other top performers.
You might also be able to find a CNN specific to handwriting and then fine tune it to yours with your samples.
This is doing it raw and will have a lot of education for you along the way. There may be some prebuilt handwriting model you can just fine tune with easy instructions from the person who made it all wrapped up into a nice bit of python for you. Maybe.