How to stop shoplifting: 𝗔𝗜 𝗜𝘀 𝗪𝗮𝘁𝗰𝗵𝗶𝗻𝗴 (𝗮𝗻𝗱 𝗜𝘁'𝘀 𝗚𝗲𝘁𝘁𝗶𝗻𝗴 𝗥𝗘𝗔𝗟𝗟𝗬 𝗚𝗼𝗼𝗱) 𝗜𝗺𝗮𝗴𝗶𝗻𝗲 𝘁𝗵𝗶𝘀: You walk into a store, casually pick up an item, and… BAM! Security gets alerted because an AI spotted a suspicious movement. Sounds like
MACHINE LEARNING
-
Eight Generative AI Buzzwords Explained and Simplified
By
–
8 Generative AI buzzwords explained! https://
datasciencedojo.com/blog/generativ
e-ai-use-cases-and-concepts/
… #AI #MachineLearning #DeepLearning #DataScience #GenerativeAI #ChatGPT #OpenAI #OpenScience #LLM #LLMs #Python #Code #100DaysOfCode @DataScienceDojo Source @jblefevre60 -
llm.c optimization: balancing historical accuracy and training speed
By
–
This is cool!! I'm not exactly sure how to upstream these changes to llm.c… Part of me wants to reproduce GPT-2/3 using their exact hyperparameters just for historical aesthetics, but part of me also wants to just train things as fast as possible. Probably both. – lr 3X is
-
Testing LLM Accuracy with Mathematical Problems
By
–
Mathematicians, unite! What are your go-to problems for testing the accuracy of LLMs?
-
Etymology and Tokenization in Language Models Explained
By
–
Ah this makes sense. So etymology is baked into the tokenization in this way. I guess there's similarity to phoenetic languages in that sense.
-
72B Model Benchmarking: MixEval and MMLU-Pro Performance Analysis
By
–
The 72B model looks amazing but wow what an excellent benchmarking! Super happy to see MixEval and MMLU-Pro there. I have high hopes for MixEval as a proxy for Arena Elo. Also curious to see if these models behave better than Llama 3 for SFT.
-
OpenAI publishes new research to understand GPT-4 internals
By
–
OpenAI has come under fire for letting its superalignment team disappear but the company just published work from @ilyasut
, @janleike
, and others on a new way to peer inside GPT-4 to better understand how it works. -
SAE for LLM Feature Extraction and Behavior Conditioning
By
–
Y lo que ha sacado OpenAI, pues bueno, se le parece demasiado… → Explorar el uso de SAE.
→ Para extraer features de un LLM.
→ Que podrían servir para condicionar el comportamiento del modelo. Ajam -
Anthropic’s Scaling Monosemanticity: Breakthrough in Transformer Interpretability
By
–
Es el tema del próximo vídeo y es FAS-CI-NAN-TE, pero el crédito va a Anthropic que no sólo ha hecho un increíble trabajo de interpretabilidad, sino también de documentación que os invito a leer si os interesa. https://
transformer-circuits.pub/2024/scaling-m
onosemanticity/index.html
… -
Sparse Autoencoders: Anthropic’s Recent AI Research Breakthrough
By
–
Todos esperando GPT-5 y nos salen con un artículo sobre Sparse Autoencoders… 😅
— Carlos Santana (@DotCSV) 6 juin 2024
Lo más gracioso es que JUSTO me ha pillado animando para el próximo vídeo un Sparse Autoencoder. Y es que esto que acaban de presentar es una copia del trabajo reciente de Anthropic. 🌉 https://t.co/WhOMFxAWlR pic.twitter.com/bUe63lh4ywTodos esperando GPT-5 y nos salen con un artículo sobre Sparse Autoencoders… Lo más gracioso es que JUSTO me ha pillado animando para el próximo vídeo un Sparse Autoencoder. Y es que esto que acaban de presentar es una copia del trabajo reciente de Anthropic.