Multimodal neurons in artificial neural networks

OpenAI Blog2021-03-04 16:000 views

We’ve discovered neurons in CLIP that respond to the same concept whether presented literally, symbolically, or conceptually. This may explain CLIP’s accuracy in classifying surprising visual renditions of concepts, and is also an important step toward understanding the associations and biases that CLIP and similar models learn.

Source：https://openai.com/index/multimodal-neurons

My favorites

PreviousGPT-3 powers the next generation of apps NextUnderstanding the capabilities, limitations, and societal impact of large language models

Meta discussed attacking Jonathan Haidt — now even his haters are annoyed

Three years ago, Meta apparently considered a plan to turn public opinion against one of its most influential opponents,

0The Verge

Distributed Responsibility Is Still Responsibility

Distributed Responsibility Is Still Responsibility Distributed systems create a recurring moral and managerial mistake:

0dev.to

blueprintUE's small footprint hides a password-reset boundary

blueprintUE's small footprint hides a password-reset boundary blueprintUE Self-Hosted Edition is a collaboration service

0dev.to

Wispr Flow: How I Increased My Coding Speed 5x with Voice Prompting

Replacing manual keyboard prompt typing with Wispr Flow voice dictation speeds up AI-assisted development workflows by u

0dev.to AI

Soft UI on the website and a reworked phone tab bar

Over the last few releases of my myOS Portfolio Platform (v2.66.1 to v2.67.2), I brought the Soft UI onto the GA/OS webs

0dev.to CSS

45 browser games tagged "adventure", 34 studios, 10 categories: what the tag actually tells a player

If you are making a browser adventure game, the word "adventure" on your listing is doing less work than you think. Here

0dev.to WebDev