This commit is contained in:
2025-06-05 21:28:30 +02:00
parent 819f6a0cb3
commit 23edf6d6ab
6304 changed files with 466 additions and 708478 deletions
@@ -1,21 +0,0 @@
full-title:: [Nobody Likes a Know-It-All: Smaller LLMs are Gaining Momentum](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df)
site:: [thesequence.substack.com](https://thesequence.substack.com/p/nobody-likes-a-know-it-all-smaller?isFreemail=true&post_id=143886317&publication_id=54309&r=z2w67&token=eyJ1c2VyX2lkIjo1ODkyMTU2NywicG9zdF9pZCI6MTQzODg2MzE3LCJpYXQiOjE3MTQzMDIyMTIsImV4cCI6MTcxNjg5NDIxMiwiaXNzIjoicHViLTU0MzA5Iiwic3ViIjoicG9zdC1yZWFjdGlvbiJ9.QchZ1G1bDqjrnlNOOkP84CYj2gm3Pnuc7Nc8FtKA3B0&triedRedirect=true)
author:: davide.diruscio@gmail.com
labels:: [[Newsletter]] [[LLMs]]
date-saved:: [[28-04-2024]]
date-archived:: [[02-05-2024]]
is-archived:: 10
source:: [[Omnivore]]
state:: [[archived]]
- ### Highlights
collapsed:: true
- > Massively large foundation models are likely to dominate generalist use cases, but they remain incredibly expensive to run, plagued with hallucinations, security vulnerabilities, and reliability issues when applied in domain-specific scenarios. [⤴️](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df#eddd34f6-fd0b-4fe0-b92c-1f49fb50166e)
- > Microsoft released the Phi-3 family of models. [⤴️](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df#159d8388-53a6-4b9e-8244-191498aef203)
- > Apple open-sourced OpenELM, a family of LLMs optimized for mobile scenarios. [⤴️](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df#cdaf6f6d-2542-4728-97e1-8ec1b60e9594)
- > Phi-3, their famous small language model that excel at match and computer science task. [⤴️](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df#c768a0fd-0511-4768-a797-8c808c15bf28)
- > #### MAIA
>
Researchers from MIT published a paper introducing **M**ultimodal **A**utomated **I**nterpretability **A**gent (MAIA), an AI agent that can design experiments to answer queries of other AI models. The method is an interesting approach to interpretability to prove generative AI models to undestand their behavior [⤴️](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df#c82046e9-30c4-4770-b552-e5a2fc793e4e)
omnivore-note:: [[PROJECTS/MOSAICO]] #omnivore-note-color
- > Google DeepMind open sourced Penzai, a research tookit for editing and visualizing neural networks and inject custom logic [⤴️](https://omnivore.app/me/nobody-likes-a-know-it-all-smaller-ll-ms-are-gaining-momentum-18f246410df#21eb992b-ae4d-4ba7-ae83-df75a2cc7d92)