dopamine_daddy

3 ms·

1.
by dopamine_daddy 2mo ago
I emailed you just now from e*****o at gmail.com
2.
by dopamine_daddy 2mo ago
I would suspect during pre-training. Even before RLHF was a thing models exhibited this bias. My bet is that it has to do with the training data corpora being composed in large part of left leaning content, maybe from social media platforms
3.
All major LLMs are lib-left. Even Grok, half the time(unslop.run)
42 points by dopamine_daddy 2mo ago|79 comments
4.
by dopamine_daddy 2mo ago
I gave the Political Compass test from politicalcompass.org to the most relevant LLMs 70 times each: 30 times using the original questions, 30 times using polarity-flipped questions to reduce affirmative bias, and 10 times with the question
5.
by dopamine_daddy 2mo ago
You raise a valid point. Just by intuition I'd say if this were true, it would probably just be a small fraction of the actual flagged articles. I will still look into how I can mitigate this when I update the detector. The difficulty
6.
by dopamine_daddy 2mo ago
Yes that is exactly what I suspected. And I see absolutely nothing wrong with this.
7.
by dopamine_daddy 2mo ago
Thank you, I plan to release the arxiv preprint codes. Also don't worry about killing my server, let me know if you succeed :D
8.
by dopamine_daddy 2mo ago
Yes, I’ve thought about this too. The strength of these models is that there is a lot more knowledge encoded in them than the average scientist has in mind at any given time. That means they can explore many more possible combinations of co
9.
by dopamine_daddy 2mo ago
That's honestly so good to hear, thank you.
10.
by dopamine_daddy 2mo ago
We can't know if real science is happening in the background but I'd wager that the majority of these papers is not complete slop but real findings with AI generated text used to communicate it. If it was just straight slop I woul
11.
by dopamine_daddy 2mo ago
Yeah I was careful on purpose with my statement. :D
12.
by dopamine_daddy 2mo ago
I tried my best to avoid leakage. If you're curious about how I trained the detector I have a writeup on it: https://unslop.run/blog/how-our-ai-text-detector-works FYI this is all relatively new so there might be
13.
by dopamine_daddy 2mo ago
Surprisingly I agree with you. My opinion is: if it makes communicating research more effective, while not reducing the quality of the output substantially, I see no issue. A possible conclusion for this could be: If the majority of CS pape
14.
How we measured AI writing across arXiv, and where the measurement breaks(unslop.run)
244 points by dopamine_daddy 2mo ago|170 comments
15.
by dopamine_daddy 2mo ago
I scored the full text of 12,750 arXiv papers from 2021 through 2026 to find out how many of these get flagged as machine written and how much it increased since the release of chatGPT. I purposely tuned the detector to avoid false positive
16.
Slople – can you pass the reverse Turing test?(unslop.run)
1 points by dopamine_daddy 3mo ago|1 comments
17.
by dopamine_daddy 3mo ago
I made a game where you rewrite a sentence to try and make it sound like it was written by an LLM. It scores you with my own mixture-of-experts MoE AI-text detector. I honestly built it to see what creative ways people come up with to fool