We find evidence that LMs have promising potential to help human facilitators and moderators synthesize the outcomes of online digital town halls—a role that requires significant expertise in quantitative & qualitative data analysis, the topic of debate, and writing skills.
RESEARCH
-
Language Models Enhance Pol.is Platform for Diverse Dialogue
By
–
We collaborated with @compdem to research the opportunities and risks of augmenting the http://
Pol.is platform with language models (LMs) to facilitate open and constructive dialogue between people with diverse viewpoints. -
Google Imagen Editor: Text-Guided Image Editing Tool
By
–
Stop by the #CVPR2023 Google booth at 12:30pm today to speak with @wangsu_googleai about Imagen Editor, a novel tool for text-guided image editing, and to learn about EditBench, a benchmark for quality assessment of image-text alignment.
-
CEO discusses future of AI with custom LLMs and open source
By
–
Our CEO @andrewdfeldman joins @bigdata on
#TheDataExchangePod to discuss the future of #AI with custom #LLMs, Cerebras-GPT and how open source, efficient → the new era of AI Tune in here: -
Stable Diffusion XL Base Model 0.9 Released on Hugging Face
By
–
HF base model: https://
huggingface.co/stabilityai/st
able-diffusion-xl-base-0.9
… -
FAIR’s Compute Challenges Compared to Google and Industry Peers
By
–
FAIR has sometimes not been seen as the destination research organization for AI, in part because historically developers don't get the level of compute that peers get at, say, Google. (Though I do hear even TPUs are more difficult to get a hold internally these days.)
-
LLaMA Innovation: Quantization and LoRA Techniques for Developers
By
–
But that sometimes breeds innovation—like in the case of LLaMA. And scrappy developers without access to massive training clusters like Google or OpenAI have to find ways around the problem. That's led to the widespread adoption of techniques like quantization and LoRA.
-
Startups Lower ML Model Training and Serving Barriers
By
–
As the cultures change here—at Meta or otherwise—the net beneficiary is likely to be the startups. Companies like MosaicML/Modal/Anyscale are trying to drop the barrier to training and serving ML models, and a lot of very cool research is being done on compact models.
-
Meta Refocuses on Commercial Products, Cuts Bio Research
By
–
Case in point, the WSJ over the weekend published a piece on how Meta is refocusing its efforts to serve commercial products. My understanding is some of its more "out there" projects will be facing scrutiny. Its bio research team, for example, is one I've heard was impacted.
-
AI Hype Reshapes Research Talent Competition at Big Tech
By
–
The hype around AI products is quickly changing the way research divisions work within larger orgs. Companies like Meta and Google have spent the last decade wooing some of the smartest minds in machine learning out of academia with the promise of working on cutting-edge tech.