rfv6723

6 ms·

61.
by rfv6723 1y ago
RDNA is a dead-end. AMD went down the wrong path by focusing on traditional rendering instead of machine learning. I think future AMD consumer GPUs would go back to GCN.
62.
by rfv6723 1y ago
These ppl are very loud online, but they don't make decisions for hyperscalers which are biggest spenders on AI chips. AMD is doing just fine, Oracle just announced an AI cluster with up to 131,072 of AMD's new MI355X GPUs. AMD ne
63.
by rfv6723 1y ago
I tried thinking with websearch on their website. It has similar speed with o4-mini with search on chatgpt, and o4-mini gave me much better result.
64.
by rfv6723 1y ago
They seem to be very skeptical against Large models. While everyone learned the bitter lesson, apple chose to focus on small on-device models even after the explosion of chatgpt.
65.
by rfv6723 1y ago
Oh, another LLM skepticism paper from Apple. This paper from last year doesn't age well due to rapid proliferation of reasoning models. https://machinelearning.apple.com/research/gsm-symbolic
66.
by rfv6723 1y ago
You can boot an OS on NVME drive from another bootloader on a USB drive. I have been using CloverBootloader for years. https://github.com/CloverHackyColor/CloverBootloader
67.
by rfv6723 1y ago
Why no comparition to gpt-4o-transcribe？ If you don't compare to latest model on the market, how can you claim it's SOTA? According to OpenAI, gpt-4o-transcribe has much better performance than whisper-large-v2. https:/&#x2F
68.
by rfv6723 1y ago
From personal experience, Google translate is fine for translation between Indo-European languages. But it is totally broken for translation between East-Asia languages.
69.
by rfv6723 1y ago
And it's a GPT4.1 warpper. GPT4.1 only cost $2 per 1M input tokens and $8 per 1M output tokens. LLM translation have been cheaper and better than deepl for a while.
70.
by rfv6723 1y ago
SRAM doesn't scale with advanced semiconductor node. Groq is heading to a dead end.
71.
by rfv6723 1y ago
Why not possible with autoregressive model? o4 mini https://chatgpt.com/share/681315c2-aa90-800d-b02d-c3ba653281...
72.
by rfv6723 1y ago
Knowledge cutoff doesn't matter because you are searching from web. Chatgpt is better because it gives links of source websites, so you can easily check them up.
73.
by rfv6723 1y ago
It is a privilege that the United States has sufficient IPv4 addresses; for Asian countries, especially developing nations that lack IPv4 addresses, IPv6 is essential.