💠Adobe Teasing Us with New AI Tool, Open Source LLaVA vs GPT-4👀

Hello vTeamers, and welcome to our AI fast-paced ride. Secure your seat belts as we explore:

📈Trend Adobe Teasing with New Tool

👩‍🔬Research LLMs Struggle with Self-Correcting in Reasoning Tasks

💵Investment Saronic Secures $55 Million in Funding

💁Tips & Tricks Creating vector illustrations with Midjourney

👻AI Fun vTeam Weekly Memes

📖 Read time: 5 minutes

So let the journey begin 🚀🚀🚀

And first, check out the vTeam AI Insight Index 👽️

The vTeam AI Insight Index is our nifty tool to give you a heads-up on what’s popping in the AI scene. We’ve got three key markers to track the buzz:

• Research: using arxiv.org, where the fresh AI papers drop, we see how much new AI chatter there is

• Trend: with Google Trends, we gauge how hot AI is right now

• Investment: at Crunchbase, we’re tracking the money flow and buzz in AI

This week’s numbers:

• Trend: 144 - Folks are talking more about AI

• Investment: 44 - Money’s still flowing into AI

• Research: 163 - Lots of new AI stuff being studied

Catch our AI beat every week! 

TREND📈

Adobe Teases "Game-Changing" AI Photo Editing Tool

Adobe's upcoming AI-powered tool, Project Stardust, will revolutionize photo editing, making it as easy as moving and editing objects in an image, removing or inserting new elements, and changing object attributes like colors. This exciting feature is set to launch across Adobe products soon. Watch the teaser video for a sneak peek.

Open-Source LLaVA Takes on GPT-4: The AI Showdown

vTeam: In a collaborative effort, researchers from Stanford, UW-Madison, and Columbia have unveiled LLaVA, a promising open-source AI system.

Highlights: 

• LLaVA combines CLIP, which understands images, with LLaMA, an open-source text AI model.

• Despite having less training data, LLaVA demonstrated its ability to engage in image-based conversations, akin to the proprietary GPT-4.

• Notably, LLaVA is open-source, making its code, model weights, and training data available online, exemplifying the potential of open-access AI in advancing vision-language capabilities.

Results: While it may not match GPT-4's scale, LLaVA's accessibility and strong performance are noteworthy.

Dropbox Unveils Web Interface Redesign and Opens Beta for AI-Powered Dash

Dropbox is introducing significant changes to its file hosting service, including a redesigned web interface with improved functionality and cleaner design.

They are launching an open beta for Dash, an AI-powered universal search tool, expanding Dropbox AI to cover all user content, and introducing Dropbox Studio, a video collaboration tool. Additionally, new subscription tiers starting at $22 per month are being introduced, offering various features and enhanced security. These updates aim to enhance user experience and productivity.

RESEARCH 👩‍🔬

DeepMind Study Reveals LLMs Struggle with Self-Correcting in Reasoning Tasks

A DeepMind and University of Illinois study reveals that large language models (LLMs) often struggle with self-correction in reasoning tasks without external feedback. Self-correction effectiveness depends on the task nature, working best when leveraging external sources, such as human feedback. The study introduces "intrinsic self-correction," where models attempt self-correction based solely on internal knowledge.

While it works well with ground-truth labels, it falters when labels are removed, impacting real-world applications. Researchers caution against blindly using self-correction, emphasizing the need for skepticism and balanced pre-hoc and post-hoc prompting in LLM tasks.

INVESTMENT 💵

🤑 Autonomous Ship Defense Startup Saronic Secures $55 Million in Funding

👾 Deta secures $3.6 million in seed funding to advance its cloud-based operating system.

🤖 Anthropic Aims for $30 Billion Valuation in $2 Billion Funding Round to Compete with OpenAI.

TIPS & TRICKS 💁

Concerned About Protecting Your AI-Generated Images? Meta got an answer

AI-generated images are on the rise, but they can deceive people. Facebook AI Research (FAIR) and Inria have introduced 'Stable Signature,' an invisible watermarking technique to identify AI-created images. It's open-source, robust to alterations, and aims to promote responsible AI use.

How to create vector illustrations with Midjourney

Use /image and type this prompt.

A flat illustration of an office women worker standing in front of a dashboard with a chart, orange and blue, gradients.

More Tips&Tricks on Instagram

AI FUN 👻

That was a fun ride, see y'all at the next stop 🙂