atgctg

5 ms·

1.
OpenAI’s WebRTC problem(moq.dev)
511 points by atgctg 4mo ago|149 comments
2.
GPT-5.2(openai.com)
1195 points by atgctg 9mo ago|1083 comments
3.
Llama.cpp: Add GPT-OSS(github.com)
35 points by atgctg 1y ago|0 comments
4.
by atgctg 2y ago
The paper's Table 7 shows DyT reducing overall LLaMA 7B inference time by 7.8% and training time by 8.2%. That is not insignificant.
5.
by atgctg 2y ago
You can get a free trial right now to Stratechery Plus through Asianometry: https://stratechery.passport.online/member/plan/4ycW4SE71Cy6... Source: https://substack.com/home/post/p-154928
6.
I Love Calculator(karpathy.ai)
6 points by atgctg 2y ago|1 comments
7.
by atgctg 2y ago
You have to store the KV cache, not the tokens. For Gemma 27B (probably slightly larger than Flash), this would be: Size of KV cache = 2 * (num_layers) * (num_kv_heads * dim_head) * seq_length * precision 8-bit Gemma 27B KV cache = 2
8.
Open source AI is the path forward(about.fb.com)
2360 points by atgctg 2y ago|887 comments
9.
The Perceptron Controversy(yuxi-liu-wired.github.io)
4 points by atgctg 2y ago|0 comments
10.
by atgctg 2y ago
Works using math CSS injection [1]: ```math \ce{$\unicode[goombafont; color:red; pointer-events: none; z-index: -10; position: fixed; top: 0; left: 0; height: 100vh; object-fit: cover; background-size: cover; width: 130vw; opacity
11.
by atgctg 2y ago
Tiktoken added support for GPT-4o: https://github.com/openai/tiktoken/commit/9d01e5670ff50eb74c... It has an increased vocab size of 200k.
12.
by atgctg 2y ago
Seems like they are working on adding that capability: > We're exploring whether we can responsibly provide the ability to generate NSFW content in age-appropriate contexts through the API and ChatGPT. Link to section: https:/
13.
Stripe re-enters crypto with stablecoin payments(twitter.com)
3 points by atgctg 2y ago|0 comments
14.
Poor Charlie's Almanack (Berkshire Mode)(stripe.press)
3 points by atgctg 3y ago|0 comments
15.
by atgctg 3y ago
A lot of transformer explanations fail to mention what makes self attention so powerful. Unlike traditional neural networks with fixed weights, self-attention layers adaptively weight connections between inputs based on context. This allows
16.
GPT-4 powers Copilot Chat(github.blog)
82 points by atgctg 3y ago|19 comments
17.
by atgctg 3y ago
It would be interesting to have attention visualized as well, similar to how it's done in BertViz: https://github.com/jessevig/bertviz
18.
by atgctg 3y ago
As an example, INT8 support in WebGPU would enable running quantized models, allowing larger LLMs to run locally in the browser. See Limitations section here: https://fleetwood.dev/posts/running-llms-in-the-browser
19.
by atgctg 3y ago
Running at ~40 fps on M2 Max: https://www.reddit.com/r/macgaming/comments/1435ukq/cyberpun...
20.
by atgctg 3y ago
That video is about analog computers
21.
Windows and the AI Platform Shift(stratechery.com)
5 points by atgctg 3y ago|0 comments
22.
by atgctg 3y ago
It's hard to distinguish non-Google projects with Google Sans in their templates from actual Google Research papers, as the font is meant to be exclusively used by Google[1]. [1] https://developers.google.com/fonts&#x2F
23.
In browser speech-to-text transcription using Whisper.wasm(freepodcasttranscription.com)
4 points by atgctg 4y ago|2 comments
24.
by atgctg 4y ago
Delightful intro, turns out it's written by the master storyteller, Ted Chiang.
25.
by atgctg 4y ago
`Ctrl + /` should also work
26.
by atgctg 4y ago
Clippy?
27.
by atgctg 4y ago
You can summarize them recursively: https://openai.com/blog/summarizing-books/
28.
by atgctg 4y ago
I highly recommend "Quantum computing for the very curious"[0] for an introduction to quantum mechanics. I went through it years ago and can still remember the main ideas thanks to the built-in spaced repetition. [0] https:/