High bandwidth communication across space using laser beams - ~30 mins max for sending a high bandwidth video from
earth to Max.
youtu.be/xxdOWEgaEJo?si…
Recent talk by Nick lane. It is always a pleasure to hear a person who understands an area articulate concepts with such clarity . @karpathy you may like it - since you have posted your thoughts on his books before
youtu.be/FLaTU-t1CQM?si…
The mainstream means news coverage of the recent Tesla recall is an example of the thirst for sensationalist coverage. Almost none of the news articles care to mention that the remedy for this recall is an OTA update which customers are used to already!
tesla.com/support/vehicl…
Not surprisingly, a comedian accurately sums up the current state of AI. However, the debate on regulating the creation and access to LLM/LMM weights is worth it, for the long term.
Not surprisingly, a comedian accurately sums up the current state of AI. However, the debate on regulating the creation and access to LLM/LMM weights is worth it, for the long term.
For those of us wondering how much of a model is memorizing and how much is generalization - this conversation is worth watching. Even if the insights are only from toy models - they are striking - a model starts with memorizing and then gradually switches to generalization.
For those of us wondering how much of a model is memorizing and how much is generalization - this conversation is worth watching. Even if the insights are only from toy models - they are striking - a model starts with memorizing and then gradually switches to generalization.
An alternative approach to reduce hallucinations without SFT or RLHF. The approach is model agnostic too. They even have a workaround solution for closed models like GPT4 where they don’t have access to model logits for tokens
An alternative approach to reduce hallucinations without SFT or RLHF. The approach is model agnostic too. They even have a workaround solution for closed models like GPT4 where they don’t have access to model logits for tokens
Directly addresses the credit assignment problem we often see these days in AI - identifies those who are materially contributing to its progress with papers and code
Directly addresses the credit assignment problem we often see these days in AI - identifies those who are materially contributing to its progress with papers and code
These are the true catalysts fueling AI progress. It is worth noting they are largely underrepresented in popular top 100 lists in part because those snapshots do not take into account the contributions made by researchers and practitioners over the years.
These are the true catalysts fueling AI progress. It is worth noting they are largely underrepresented in popular top 100 lists in part because those snapshots do not take into account the contributions made by researchers and practitioners over the years.
If you'd asked me a year ago, superposition would have been by far the reason I was most worried that mechanistic interpretability would hit a dead end.
I'm now very optimistic. I'd go as far as saying it's now primarily an engineering problem -- hard, but less fundamental risk.
If you'd asked me a year ago, superposition would have been by far the reason I was most worried that mechanistic interpretability would hit a dead end.
I'm now very optimistic. I'd go as far as saying it's now primarily an engineering problem -- hard, but less fundamental risk.
An alternative to Nvidia for computing at scale
Condor Galaxy 1 AI supercomputer specifications:
4 exaFLOPS of AI compute at FP16 with sparsity
54 million AI optimized compute cores
82 terabytes of memory
64 Cerebras CS-2 systems
Base configuration supports 600 billion…
I was hesitant to watch this video at first. But then I did. It is nearly impossible not to feel both despondence ( the absurd brutality of war) and hope (seeing doctors work on the front lines saving lives). This is raw footage - as real as it gets.
This video is also a stark…
Inspirational talk by Andrej to a young audience on the importance of building AI agents. An interesting line snuck in between- “and I got distracted a bit with self-driving…” he later adds self-driving belongs to a class of problems that appears easy but is very hard to…
Inspirational talk by Andrej to a young audience on the importance of building AI agents. An interesting line snuck in between- “and I got distracted a bit with self-driving…” he later adds self-driving belongs to a class of problems that appears easy but is very hard to…
i might have heard the same 😃 -- I guess info like this is passed around but no one wants to say it out loud.
GPT-4: 8 x 220B experts trained with different data/task distributions and 16-iter inference.
Glad that Geohot said it out loud.
Though, at this point, GPT-4 is…
i might have heard the same 😃 -- I guess info like this is passed around but no one wants to say it out loud.
GPT-4: 8 x 220B experts trained with different data/task distributions and 16-iter inference.
Glad that Geohot said it out loud.
Though, at this point, GPT-4 is…
I have never encountered this before. Bing+GPT offers a result to a question that is completely unrelated to the input question. The odd thing it happens after a proper start.
📣 New dataset drop!
Introducing SlimPajama-627B: the largest extensively deduplicated, multi-corpora, open-source dataset for training large language models. 🧵cerebras.net/blog/slimpajam…
30 Followers 18 FollowingWith 65 million global subscribers and an AI platform running on 125M videos, our App infuses viral ingredients into creator content: https://t.co/IGdtll5aS9
496 Followers 7K FollowingTech Entrepreneur & Digital Marketer by Day 🚀
Music Producer by Night 🎧
Join Me in Unlocking the Secrets of AI, Digital Marketing & Online Business.
26 Followers 100 FollowingOn a journey to Be and Build the Infrastructure between Minds. Life is littered with paradox and I intend to resolve them all
186 Followers 436 FollowingGeek. Dad. Independent. Work hard. Be nice. Do right for it's own sake. Anti-tribe, pro-integrity, pro-wonder. #GoodnessCounts! bsky: @chustonai.com
375 Followers 4K Followinghttps://t.co/DNa1eMyOq4 & https://t.co/gWroBpclUd from IIT Madras, PhD from the National University of Singapore, former Consultant at Ernst & Young, and currently a researcher in public health.
36K Followers 16K FollowingFounder/CEO of @Vectara (Trusted GenAI for the Enterprise). Founder/ex-CTO @Cloudera, ex-VP at Yahoo & Google. PhD EE Stanford. IG, FB, LI: @awadallah.
712K Followers 288 FollowingTogether with the AI community, we are pushing the boundaries of what’s possible through open science to create a more connected world.
30K Followers 123 FollowingMechanistic Interpretability lead DeepMind. Formerly @AnthropicAI, independent. In this to reduce AI X-risk. Neural networks can be understood, let's go do it!
44 Followers 471 FollowingPassionate explorer of applications of ethical use of machine learning in healthcare, climate change and process optimization.
498 Followers 1K FollowingDeep learning @nvidia
MS in Data Science from GalvanizeU/University of New Haven.Working on deep learning for computer vision and language understanding.
35K Followers 189 FollowingCo-founder and CEO https://t.co/efv72CKpAG (@WaveFormsAI) - Ex @OpenAI GPT-4o/AVM Audio Research Lead - #Her #TARS - Ex @AIatMeta, @Polytechnique (X11)
30 Followers 18 FollowingWith 65 million global subscribers and an AI platform running on 125M videos, our App infuses viral ingredients into creator content: https://t.co/IGdtll5aS9
186 Followers 436 FollowingGeek. Dad. Independent. Work hard. Be nice. Do right for it's own sake. Anti-tribe, pro-integrity, pro-wonder. #GoodnessCounts! bsky: @chustonai.com
375 Followers 4K Followinghttps://t.co/DNa1eMyOq4 & https://t.co/gWroBpclUd from IIT Madras, PhD from the National University of Singapore, former Consultant at Ernst & Young, and currently a researcher in public health.
36K Followers 16K FollowingFounder/CEO of @Vectara (Trusted GenAI for the Enterprise). Founder/ex-CTO @Cloudera, ex-VP at Yahoo & Google. PhD EE Stanford. IG, FB, LI: @awadallah.
3K Followers 2K FollowingMachine Learning🧠, Natural Language Processing📖 and Information Retrieval🔍 Search Engines, Recommenders, Chat-Bots💬 Empowering Researchers and Engineers
412 Followers 234 FollowingCo-founder, CTO & VP Engineering at @NexusflowX | Ex-Director of Machine Learning at @SambaNovaAI | PhD in machine learning at @Stanford