Is Signal Processing The CURE For AI's ADHD?
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Related videos
How DeepSeek V4 Broke AI’s Cost Curse
bycloud
101.8k views
There Are Signs Something BIG is Coming...
Stefan Burns
249.3k views
The Cure For Being Ticklish
Good Mythical Morning
146.2k views
Bitcoin Is Winning And That's The Problem
Coin Bureau
16.2k views
Microsoft's AI business is falling apart
The Friday Checkout
83.3k views
The design process is dead. Here’s what’s replacing it. | Jenny Wen (head of design at Claude)
Lenny's Podcast
41.8k views
What’s the best programming language for AI?
Theo - t3․gg
75.8k views
The RL Irony in LLMs
bycloud
23.0k views
Trump SCHEME is a TICKING TIME BOMB For CRIMINAL PROBES
Legal AF
19.8k views
The Chinese AI Iceberg
bycloud
107.4k views
Top Comments (10)
btw how is Mamba doing? There was a ton of hype surrounding her back in the day
0:11 I was so sure the truck was gonna transform into Optimus Prime
delithful memes! Man where do you find energy to post so many supreme quality videos so often? My deepest respect
It's just a hunch, but the biggest unrealised gains are at the token embedding stage. If we can formalise language by pre-processing whole words into semantic word senses (on a multilingual / language independent word sense knowledge graph like bablenet) using contextual and linguistic analysis, then each token will actually "mean something" and the sentence as a whole will become "formal language" like code instead of natural language. It's basically insane that we cut words up and assume the parts have meaning. It's a miracle that it works at all, even if one token encodes "duck" as a nown or verb, they're not at all related. So much attention wasted untangling composite embeddings.
Plot twist: The truck was Optimus Prime The Transformer.
You caught me off guard with that locked in meme
Thanks for covering that paper. I was amazed when I read it and really wanna see someone implementing it into an open model that can be tested by the community. Would be great for a small 3B model on your phone.
That intro tho? Absolute cinema
8:51 either "62.2% of its parameters" or "62.2% less parameters"
Check out HubSpot's Free ChatGPT Bundle! https://clickhubspot.com/jgv5 maybe I'll also cover ring attention and tree attention next time
Unlock the Data Inside
Turn Videos into Knowledge
- Get FREE 10/day: transcripts, summaries, chats
- Chat with videos, export text & PDF
- $1 free API credit for RAG, chatbots & research
Free forever plan • All features unlocked
Top Comments (10)
btw how is Mamba doing? There was a ton of hype surrounding her back in the day
0:11 I was so sure the truck was gonna transform into Optimus Prime
delithful memes! Man where do you find energy to post so many supreme quality videos so often? My deepest respect
It's just a hunch, but the biggest unrealised gains are at the token embedding stage. If we can formalise language by pre-processing whole words into semantic word senses (on a multilingual / language independent word sense knowledge graph like bablenet) using contextual and linguistic analysis, then each token will actually "mean something" and the sentence as a whole will become "formal language" like code instead of natural language. It's basically insane that we cut words up and assume the parts have meaning. It's a miracle that it works at all, even if one token encodes "duck" as a nown or verb, they're not at all related. So much attention wasted untangling composite embeddings.
Plot twist: The truck was Optimus Prime The Transformer.
You caught me off guard with that locked in meme
Thanks for covering that paper. I was amazed when I read it and really wanna see someone implementing it into an open model that can be tested by the community. Would be great for a small 3B model on your phone.
That intro tho? Absolute cinema
8:51 either "62.2% of its parameters" or "62.2% less parameters"
Check out HubSpot's Free ChatGPT Bundle! https://clickhubspot.com/jgv5 maybe I'll also cover ring attention and tree attention next time