Aedelon

4 ms·

1.
by Aedelon 6mo ago
Sorry, I wrote wrong link for several source. 1. Appfigures $2.1M = https://appfigures.com/reports/app-profile/338340235920 2. Watermark bypass = https://www.404media.co/sora-2-watermark-removers-f
2.
by Aedelon 6mo ago
Full quality Sora yeah probably needs serious hardware. distilled version on a 4090 though? maybe. danjl earlier in this thread made a solid case for just distributing the weights and letting people run locally. the SD/ComfyUI crowd al
3.
by Aedelon 6mo ago
your $120-150/month gut feeling is basically where the math lands. $1.30/clip, 100 clips/month, you need $130 just to cover compute. $150 with margin, checks out. Problem is at 10x the price you lose 90%+ of users and the who
4.
by Aedelon 6mo ago
Anthropic doesn't break out per-product numbers but they're at $19B annualized, Claude Code being a big chunk. The thing is text/code inference is just way cheaper. A Claude Code session runs maybe $0.01-0.05 in compute. A So
5.
by Aedelon 6mo ago
This is the model that makes sense to me and I'm surprised nobody at OpenAI pursued it. Yeah a 4090 would take hours for 10 seconds of video, but people already do this. The SD/ComfyUI crowd runs overnight batch generations on con
6.
by Aedelon 6mo ago
Yeah fair, the $65 is for someone cranking out 50 clips/month. Most users were probably doing 5-10, so more like $6.50-$13 in compute. That's fine at $20/month. Doesn't change the bigger picture much though. OpenAI'
7.
CUDA Tile is the biggest GPU programming shift in 20 years(pub.towardsai.net)
2 points by Aedelon 6mo ago|0 comments
8.
A $20/month user costs OpenAI $65 in compute. AI video is a money furnace(aedelon777.substack.com)
76 points by Aedelon 6mo ago|42 comments
9.
Anthropic's Mythos leak: 3k files in a public CMS, and what the docs revealed(medium.com)
38 points by Aedelon 6mo ago|5 comments
10.
Testing Nvidia's FP4: Running 70B LLMs on a Single RTX 5090 with Real Benchmarks(ai.gopubby.com)
2 points by Aedelon 7mo ago|0 comments
11.
China's AI progress by the numbers: GLM-5 benchmarks, robotaxi, and Huawei chips(medium.com)
2 points by Aedelon 7mo ago|1 comments
12.
AI Broke ARC-AGI-2 at 84.6% – But the Key Trick Is from 1972(ai.gopubby.com)
1 points by Aedelon 7mo ago|0 comments
13.
Pure LLMs Score 0% on ARC-AGI-2. Why the Third Wave of AI Looks Like the First(ai.gopubby.com)
1 points by Aedelon 7mo ago|0 comments
14.
by Aedelon 7mo ago
Yeah, that HF dataset page is rough. 247+ threads, mostly DMCA reports, archive-locked fics scraped without consent, dataset reuploaded after takedown. The AO3 community had every reason to be furious. Not RWKV-specific though. Most large c
15.
by Aedelon 7mo ago
Author here. The core claim: RWKV-7 (2.9B params, RNN) scores 72.8% avg across standard benchmarks vs LLaMA 3.2's 69.7% — trained on 3.1T tokens vs ~9T. Same parameter count, one-third the data. The more interesting result is architect
16.
RWKV-7 beats Llama 3.2 with 3x fewer training tokens and formally exceeds TC^0(ai.gopubby.com)
2 points by Aedelon 7mo ago|3 comments
17.
Reasoning Models Fabricate 75% of Their Explanations (ArXiv:2505.05410)(ai.gopubby.com)
4 points by Aedelon 7mo ago|0 comments
18.
by Aedelon 8mo ago
Both angles are real but they play out differently.
19.
by Aedelon 8mo ago
Survey of 65+ papers on model collapse. Key finding from Dohmatob et al. (ICLR 2025): even 0.1% synthetic contamination in training data causes measurable degradation. No major dataset (FineWeb, RedPajama, C4) currently filters for AI-gener
20.
0.1% synthetic data is enough to degrade AI models (Nature, 2024)(medium.com)
2 points by Aedelon 8mo ago|3 comments