Aedelon's submissions

1.		CUDA Tile is the biggest GPU programming shift in 20 years (towardsai.net)
		2 points by Aedelon 57 days ago \| past
2.		A $20/month user costs OpenAI $65 in compute. AI video is a money furnace (aedelon777.substack.com)
		76 points by Aedelon 57 days ago \| past \| 42 comments
3.		Anthropic's Mythos leak: 3k files in a public CMS, and what the docs revealed (medium.com/ai-advances)
		38 points by Aedelon 62 days ago \| past \| 5 comments
4.		Testing Nvidia's FP4: Running 70B LLMs on a Single RTX 5090 with Real Benchmarks (gopubby.com)
		2 points by Aedelon 80 days ago \| past
5.		China's AI progress by the numbers: GLM-5 benchmarks, robotaxi, and Huawei chips (medium.com/ai-advances)
		2 points by Aedelon 81 days ago \| past \| 1 comment
6.		AI Broke ARC-AGI-2 at 84.6% – But the Key Trick Is from 1972 (gopubby.com)
		1 point by Aedelon 3 months ago \| past
7.		Pure LLMs Score 0% on ARC-AGI-2. Why the Third Wave of AI Looks Like the First (gopubby.com)
		1 point by Aedelon 3 months ago \| past
8.		RWKV-7 beats Llama 3.2 with 3x fewer training tokens and formally exceeds TC^0 (gopubby.com)
		2 points by Aedelon 3 months ago \| past \| 3 comments
9.		Reasoning Models Fabricate 75% of Their Explanations (ArXiv:2505.05410) (gopubby.com)
		4 points by Aedelon 3 months ago \| past
10.		0.1% synthetic data is enough to degrade AI models (Nature, 2024) (medium.com/ai-advances)
		2 points by Aedelon 3 months ago \| past \| 3 comments