pastefile1
@@ -0,0 +1,207 @@
+The lolcow of g4f
+🎵 https://www.youtube.com/watch?v=T6eK-2OQtew 🎵
+
+meow_18838 on discord, 1205039741754671147 id
+alt account: 1275892277662716098 id
+another alt account id: 1200861224104116425
+
+==============================================================================================================================
+
+— CSAM-Suspicious behavior —
+https://files.catbox.moe/6gksas.png
+https://files.catbox.moe/djzx5c.png
+https://files.catbox.moe/nh3yfz.png
+https://files.catbox.moe/3uli1f.png
+https://files.catbox.moe/09vfef.png
+
+==============================================================================================================================
+Ownership change fake-outs:
+https://files.catbox.moe/25tk8k.png - Lucifer account is Meow, or at least a co-resident: <Proof is on Doxbin, so not mentioned here.>
+
+LukasHabtoch: Faked a hospital visit in means of “avoiding.” https://files.catbox.moe/vvii2u.png https://files.catbox.moe/dayjwx.png
+Connection: https://web.archive.org/web/20240913223331/https://www.youtube.com/channel/UCAAk_Qtpue_1EQmF5sMidbA
+- https://files.catbox.moe/bcf2r3.png
+* https://files.catbox.moe/6tqtji.png
+* https://files.catbox.moe/7qzp4l.png
+* https://files.catbox.moe/yy3mpv.png
+
+
+
+==============================================================================================================================
+— Own models, purported to be some "amazing models", are tiny finetuned of existing models. Breaching licenses. —
+
+There is a pattern of faking models, or at least writing off models to be “better”, even though such are merely system-prompted other models.
+* Seen in SplitticAI (https://files.catbox.moe/g7firb.png & https://www.youtube.com/watch?v=kJDq0pS-8D8&t=2027s )
+* Seen in AnyAI (a later project) https://files.catbox.moe/799h5a.png
+* And seen in CablyAI (most current project, see below)
+
+-- Cably-80b Incident: Faked model benchmarks --
+https://files.catbox.moe/mpamr2.jpg
+https://files.catbox.moe/ktylqh.jpg
+https://files.catbox.moe/pzfb7j.jpg
+https://files.catbox.moe/f7a5no.jpg
+https://files.catbox.moe/1l5m3w.jpg
+https://files.catbox.moe/rjpz2h.jpg
+
+—> 01-preview faking:
+https://files.catbox.moe/cf4s8j.png
+https://files.catbox.moe/2cf1y8.png
+https://files.catbox.moe/nsbvl9.png
+https://files.catbox.moe/vnden8.png
+
+—> Backtracks after realizing that it is fake:
+- https://files.catbox.moe/d0yqyy.png
+* https://files.catbox.moe/id5id0.png
+
+Repeat offender of ‘system-prompting ChatGPT to act a way’:
+https://files.catbox.moe/b51ca7.png
+https://files.catbox.moe/4j9oxu.png
+
+
+Other Text models likely to be put under scrutiny.
+https://files.catbox.moe/4pd49q.png
+https://files.catbox.moe/k71yf6.png
+
+
+Image models:
+
+Overall, go note the similarities of "Size" "Text" and "Prompt" across all models too.
+
+-- Interdiffusion 4: --
+https://web.archive.org/web/20240913104538/https://huggingface.co/cutycat2000x/InterDiffusion-4.0/commit/c866a5849598f86ca72936119efc94b3ee38a7bd --> "retroactively changes the base model"
+
+-- Interdiffusion-3.5 nonsense, especially: --
+https://web.archive.org/web/20240913104907/https://huggingface.co/cutycat2000x/InterDiffusion-3.5/commit/0b2896da1015bab3f2cbd230a1953f476cc9efa8 --> "2.5" with "-3" as a base on the "3.5" repo.
+
+https://web.archive.org/web/20240913105147/https://huggingface.co/cutycat2000x/InterDiffusion-3.5/commit/1aa88af0194b54780d066de64ba3caa149075a0f --> "example image" copy/pasted from his wrong model.
+
+https://web.archive.org/web/20240913105312/https://huggingface.co/cutycat2000x/InterDiffusion-3.5/commit/b246c16913f472857e0e6d3ef13fa8c2564d4a04 --> images "improved" over time.
+
+https://web.archive.org/web/20240913105459/https://huggingface.co/cutycat2000x/InterDiffusion-3.5/commit/61939723750aea983a4d31ba8958e8bce08d3de0 --> ends up having renamed the model to "3.2" to then "3.5" and purge all of the fake images. Hahhahahah
+
+https://web.archive.org/web/20240913105821/https://huggingface.co/cutycat2000x/InterDiffusion-3.5/commit/fa7da06141d0b58f25da695b5f78ed2ac763d9b1 --> Images were stored/generated on his own cloud.
+
+-- Interdiffusion 3: --
+https://web.archive.org/web/20240913105921/https://huggingface.co/cutycat2000x/InterDiffusion-3/commit/3cc06ff0194d76ddfe8ea2932fa9e007c93377c9 --> again, "2.5", even though its 3. Also shown are datasets, the only time they are shown, likely the only reason his stuff performed.
+
+
+Claims to have recently created a model to generate images “very fast”, but it is just würstchen ( https://huggingface.co/blog/wuerstchen , model released 7 days before the original VOD.)
+- https://youtu.be/kJDq0pS-8D8?t=2034
+
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+
+
+==============================================================================================================================
+— Highly engaged with malware, for at least some time. —
+
+The Connection:
+https://web.archive.org/web/20240322222408/https://pypi.org/project/Splittic/
+https://web.archive.org/web/20240324185407/https://pypi.org/user/Dart2.0/
+https://github.com/Dart2004/Splittic-AI
+https://web.archive.org/web/20240913111357/https://github.com/Dart2004/Splittic-AI
+
+The malware:
+https://web.archive.org/web/20240913113516/https://github.com/Dart2004/jupitergrabber/blob/main/inject.js
+https://files.catbox.moe/nt0xx1.png
+https://files.catbox.moe/w25p8y.png
+https://web.archive.org/web/20240913113803/https://github.com/Dart2004?tab=repositories
+
+
+celestialscape.com is the site hosting his shit malware —> https://files.catbox.moe/ttx2j9.png
+
+
+Found in discord server https://discord.gg/ufvyg5F2j4 asking for help with skidding the malware he chose to use
+
+Malicious files downloaded from his site at celestialscape.com can be found in the ./distributed-malware-files/ directory
+A decompiled version ceche_installer.exe (which is a file dropped by the main installer.exe file).
+This decompiled version HILARIOUSLY has plaintext python code, where the discord webhook listed below can be found
+https://web.archive.org/web/20240129180859/https://celestialscape.com/
+https://web.archive.org/web/20240129212331/http://celestialscape.com/installer.exe // MALWARE CONTAINING WEBHOOK LISTED ABOVE
+
+
+==============================================================================================================================
+Singing. No further context.
+https://web.archive.org/web/20240913121316/https://www.youtube.com/@cutycat2000
+https://files.catbox.moe/17atk2.png
+https://gofile.io/d/riD14r
+
+
+==============================================================================================================================
+— His First Attempt: ImagineAI —
+
+The Connection: https://files.catbox.moe/6qswf3.png (same PayPal as current)
+
+https://web.archive.org/web/20230623223819/https://huggingface.co/ImagineAI-Real
+https://files.catbox.moe/zej7ur.png https://files.catbox.moe/lxpom6.png - splittifcai connection
+https://files.catbox.moe/ikgplp.png - File storage weirdness
+
+https://files.catbox.moe/oxo49d.png
+https://files.catbox.moe/6iexh9.png
+https://replit.com/@CodingLehrer/ImagineImages?v=1
+https://web.archive.org/web/20240910025608/https://huggingface.co/spaces/ImagineAI-Real/ImagineAI-Image-Generator3/blob/main/app.py
+Ludicrous claims to be proven absolutely false (its all just stable-diffusion fine tunes)
+
+
+==============================================================================================================================
+— The Path to SplitticAI —
+https://web.archive.org/web/20230325131810/https://huggingface.co/Daniton
+https://web.archive.org/web/20240913111357/https://github.com/Dart2004/Splittic-AI
+SplitticAI “Super” Chatbot: https://web.archive.org/web/20240913112040/https://huggingface.co/spaces/Daniton/superai-chatbot/blob/main/app.py
+
+
+Main Situation:
+Meow covered up his supposed “top tier” peak model with apis of other free models. These were not his own model, but repackaged other models that were all freely available ( https://youtu.be/kJDq0pS-8D8?t=1078 ) elsewhere.
+tldr: https://www.youtube.com/watch?v=kJDq0pS-8D8&t=4109s
+
+
+Donation money has been used “for those models” which do not exist. https://youtu.be/kJDq0pS-8D8?t=1033
+
+https://youtu.be/kJDq0pS-8D8?t=1128 - 500 Token Input “Thats why the edit would not have been possible. Bug with mp3 generation (?)”
+-> Note: Official StableDiffusion only supports 77 tokens natively at the time for the most part. https://github.com/huggingface/diffusers/issues/2136
+-> 500 tokens, ~1 per 4 characters, appears to approximately align with the max limit able to be generated by https://playground.com/create
+https://youtu.be/kJDq0pS-8D8?t=1184 - This is confirmed by Meow as the “supposed issue why the faking had to occur.”
+-> Ex-Co-Owner confirms playground usage: https://youtu.be/kJDq0pS-8D8?t=3390
+-> Further confirmation is available through url scanning. Images generated by “SplitticAI” are from playgroundAI. https://files.catbox.moe/dayjwx.png
+-> https://github.com/search?q=%22storage.googleapis.com%2Fpai-images%22+AND+%22playground%22&type=code
+
+Meow claims the models were real at the time of the video: https://youtu.be/kJDq0pS-8D8?t=2527 , and claims to “send it later” to MattVidPro ( https://youtu.be/kJDq0pS-8D8?t=1418 ).
+-> Matt (https://www.youtube.com/channel/UC5Wz4fFacYuON6IKbhSa7Zw), himself, disproves this:
+https://files.catbox.moe/b94dcb.png
+https://files.catbox.moe/fudawd.png
+https://files.catbox.moe/1o271y.png
+
+
+The code, as revealed by witness testimonies, used BingImageCreator.py (Dalle-3) and “OverAI” which likely refers to OpenAI.-> https://youtu.be/kJDq0pS-8D8?t=1377
+Meow claims it to be a “base64 images model” trained by GPT-2 https://youtu.be/kJDq0pS-8D8?t=1416 which is NOT POSSIBLE TO ACTUALLY DO (GPT-2 not only cannot generate base64 images ‘reliably’, but also 200 base64 images would need a massive number of tokens in output, something that GPT-2 did not support, and modern ones still don’t. ( https://youtu.be/kJDq0pS-8D8?t=1461 + https://youtu.be/kJDq0pS-8D8?t=1581 ).
+Meow claims it to be a model that “generates 200 base64 images and picked the some, all that work, and pick the best one of them […] it is trained on only valid base64 images”: https://youtu.be/kJDq0pS-8D8?t=1512
+Meow claims it to be a model of “multiple terabytes”: https://youtu.be/kJDq0pS-8D8?t=3083
+Meow claims this to the GPT-2 Splsync and Splgen: https://youtu.be/kJDq0pS-8D8?t=3141 of “multiple terabytes”, ( https://youtu.be/kJDq0pS-8D8?t=3168 ) even though GPT-2 itself is 3.25GB ( https://huggingface.co/openai-community/gpt2-large/blob/main/model.safetensors )
+The model name Splgen is the same model name as also shown to be using images from PlaygroundAI (as shown above).
+
+Meow also later claims it is 1.3TB “specifically” which is complete nonsense at the time, as the largest public model now, not then, is Llama-3.1-405b-instruct, which is ~800GB, still much smaller than what was claimed. https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-Instruct/tree/main
+https://youtu.be/kJDq0pS-8D8?t=3313
+https://files.catbox.moe/o4kyg1.png
+
+This model, is later also claimed to be able generate “images and songs” which is also, fundamentally, not possible. ( https://youtu.be/kJDq0pS-8D8?t=2274 )
+-> The “Music Generation” element also had been shown to be faked, as it was Elevenlabs calls, not an actual model. ( https://youtu.be/kJDq0pS-8D8?t=3587 )
+-> Faking is also supported by witness testimony from those times ( https://files.catbox.moe/tu7ax5.png )
+
+Meow, after already having provided absurd claims of this “GPT-2 Finetune” went into talking about the size of it.
+-> (In Chat), claims to need a 512GB RAM-capable CPU to run this 1.3TB model ( https://www.youtube.com/watch?v=kJDq0pS-8D8&t=3760s + https://files.catbox.moe/1o7eh5.png )
+-> To clarify, a server of this size would cost in the tens of thousands, assuming the model is real, and has these requirements ( https://news.ycombinator.com/item?id=8065635 )
+-> Given the claims that he is “running it at non-profit” ( https://files.catbox.moe/nlpjch.png + https://youtu.be/kJDq0pS-8D8 ), this is impossible.
+-> This fact is repeated later, the ability to store “200TB of data” is not possible without 10s of thousands of dollars worth of servers ( https://youtu.be/kJDq0pS-8D8?t=3827 )
+
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+===> HE ADMITS TO THIS BEHAVIOR. https://youtu.be/kJDq0pS-8D8?t=2097 <====
+
+
+
+Paypal, the one simple thing that connects across all servers.
+https://www.paypal.com/donate/?hosted_button_id=TRYUC22GQ3HFG
+https://files.catbox.moe/u65my8.png
\ No newline at end of file