Hacker News: kuprel

New comment by kuprel in "Claude Fable 5"

kuprel — Tue, 09 Jun 2026 19:23:00 +0000

https://artificialanalysis.ai/evaluations/humanitys-last-exa... Not bad

New comment by kuprel in "DARPA's new X-76"

kuprel — Mon, 09 Mar 2026 18:48:49 +0000

From the image it doesn't look balanced for VTOL when the propellors are vertical. Also are the jets enabled during VTOL?

New comment by kuprel in "Gemini 3.1 Pro"

kuprel — Thu, 19 Feb 2026 21:23:21 +0000

GPT 5.2 loses at everything but they included that

New comment by kuprel in "Gemini 3.1 Pro"

kuprel — Thu, 19 Feb 2026 19:43:45 +0000

Why don't they show Grok benchmarks?

New comment by kuprel in "./HelpGod"

kuprel — Wed, 27 Aug 2025 17:45:12 +0000

Sounds scary haha

./HelpGod

kuprel — Wed, 27 Aug 2025 17:31:33 +0000

Article URL: https://github.com/kuprel/serenity

Comments URL: https://news.ycombinator.com/item?id=45042495

Points: 4

# Comments: 2

New comment by kuprel in "God Simulator in Bash"

kuprel — Wed, 27 Aug 2025 04:34:53 +0000

A simulation of God's decision tree for the Serenity Prayer: "God, grant me the serenity to accept the things I cannot change, the courage to change the things I can, and the wisdom to know the difference."

God Simulator in Bash

kuprel — Wed, 27 Aug 2025 04:34:53 +0000

Article URL: https://github.com/kuprel/serenity

Comments URL: https://news.ycombinator.com/item?id=45035498

Points: 4

# Comments: 2

New comment by kuprel in "New research reveals the strongest solar event ever detected, in 12350 BC"

kuprel — Mon, 19 May 2025 02:23:36 +0000

This was my first thought. It seems this solar event happened over 1000 years before Younger Dryas

New comment by kuprel in "OpenAI Audio Models"

kuprel — Thu, 20 Mar 2025 20:27:46 +0000

They released a crippled model: https://github.com/SesameAILabs/csm/issues/63

New comment by kuprel in "OpenAI Audio Models"

kuprel — Thu, 20 Mar 2025 18:48:18 +0000

OpenAI doesn’t have voice cloning

Show HN: AI Tim Draper (With Permission)

kuprel — Thu, 20 Feb 2025 21:51:40 +0000

Article URL: https://timdraper.ai/

Comments URL: https://news.ycombinator.com/item?id=43120752

Points: 2

# Comments: 0

New comment by kuprel in "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL"

kuprel — Mon, 27 Jan 2025 04:26:55 +0000

I wonder if a language model can be treated as a policy over token-level actions instead of full response actions. Then each response from the language model is a full rollout of the policy. In math and coding, the reward for the response can be evaluated. This is not how DeepSeek works now, right? It treats full responses from the language model as the action if I understand correctly.

New comment by kuprel in "Llama 3.1 405B now runs at 969 tokens/s on Cerebras Inference"

kuprel — Tue, 19 Nov 2024 03:54:21 +0000

I wonder if Cerebras could generate video decent quality in real time

New comment by kuprel in "1 in 4 U.S. Teachers Say AI Tools Do More Harm Than Good in K-12 Education"

kuprel — Thu, 30 May 2024 23:45:42 +0000

So then 3 in 4 say AI tools do more good than harm?

New comment by kuprel in "Egypt's pyramids may have been built on a long-lost branch of the Nile"

kuprel — Thu, 16 May 2024 17:54:19 +0000

What if they’re even older than 11.5ky? Neanderthals had larger brains and have been around for half a million years

New comment by kuprel in "Egypt's pyramids may have been built on a long-lost branch of the Nile"

kuprel — Thu, 16 May 2024 16:29:42 +0000

So maybe the pyramids are older than we thought?

New comment by kuprel in "55x Speedup of Andrej Karpathy's Minbpe LLM Tokenizer with PyTorch/CUDA"

kuprel — Fri, 23 Feb 2024 20:16:52 +0000

Using int16 and an H100 the speedup is actually 108x over the M2 air

New comment by kuprel in "55x Speedup of Andrej Karpathy's Minbpe LLM Tokenizer with PyTorch/CUDA"

kuprel — Fri, 23 Feb 2024 17:06:10 +0000

The M2 Air was actually much faster than whatever CPU was on the cloud RTX4090 machine I rented. I chose the stronger benchmark to compare to

New comment by kuprel in "55x Speedup of Andrej Karpathy's Minbpe LLM Tokenizer with PyTorch/CUDA"

kuprel — Thu, 22 Feb 2024 21:56:41 +0000

This adds PyTorch/CUDA training support to Andrej Karpathy's minbpe. It takes 2min 28sec (148 seconds) on an RTX4090 to train the BasicTokenizer with a vocab_size of 512 on 307MB of Enron emails. The original code takes 2hrs 15min (8076 seconds) on an M2 Air with Python 3.11 to do this. That is a 55x speedup.