If you asked Co-LLM to name some examples of extinct bear species, two models would draft answers together. The general-purpose LLM begins to put together a reply, w/the switch variable intervening where it can slot in a better token from the expert mode (i.e. adding the year
@mit_csail
-
Co-LLM Switch Variable Routes Tasks Between Base Expert Models
By
–
To decide when a base model needs help from an expert model, Co-LLM uses machine learning to train a "switch variable," or a tool that can indicate the competence of each word w/i the two LLMs’ responses.
— MIT CSAIL (@MIT_CSAIL) 17 septembre 2024
It’s like a project manager deciding when to call in a specialist. pic.twitter.com/E2UMzf9b4oTo decide when a base model needs help from an expert model, Co-LLM uses machine learning to train a "switch variable," or a tool that can indicate the competence of each word w/i the two LLMs’ responses. It’s like a project manager deciding when to call in a specialist.
-
MIT Co-LLM Algorithm Enables Specialized Model Collaboration
By
–
Can LLMs learn to "phone a friend?" MIT CSAIL’s new "Co-LLM" algorithm can pair a general-purpose base LLM w/a more specialized model & help them work together. It reviews each token & sees where it needs to call upon an expert, leading to more accurate & efficient replies to
-
90s Teachers Wrong About Calculators in Pockets
By
–
“You won't always have a calculator in your pocket!” — Lying 90s teachers v/
@engineers_feed -
IBM’s 1956 RAMAC: First Disk Drive Revolution
By
–
#otd in 1956 IBM unveiled the first disk drive, the 305 RAMAC. It was 16 sq. ft., weighed over a ton & had to be transported by plane — but it allowed a computer w/only a few KBs of memory to access the equivalent of 64K punch-cards. https://
bit.ly/3z9A3c0 Image v/Ed Thelen -
ScribblePrompt: AI-Powered Interactive Image Segmentation Tool
By
–
Authors: Hallee Wong (
@HalleeWong
), Marianne Rakic (
@MarianneRakic
), John Guttag, and Adrian Dalca (
@AdrianDalca
)
Paper: https://
arxiv.org/abs/2312.07381 Website: https://
scribbleprompt.csail.mit.edu
Video: https://
bit.ly/3Tn76Ac -
ScribblePrompt Outperforms SAM in Neuroimaging Annotation
By
–
ScribblePrompt reduced annotation time by 28% compared to Meta’s Segment Anything Model (SAM). Its self-correcting, interactive capabilities made ScribblePrompt the preferred tool among neuroimaging researchers at MGH in a user study. 93.8% of these users favored the MIT
-
ScribblePrompt: AI Image Segmentation Tool from MIT
By
–
ScribblePrompt’s interface is simple: users can scribble across the rough area they’d like segmented, or click on it, and the tool will highlight it as requested. Check out the web app here: https://
scribbleprompt.csail.mit.edu/demo -
ScribblePrompt: AI Algorithms Simulate Human Annotation in Medical Imaging
By
–
To train ScribblePrompt, the team used algorithms to simulate how humans would scribble & click on different regions in medical images. In addition to commonly labeled regions, the team also used superpixel algorithms to identify potential new regions of interest to medical
