The Speedy Challenger to Stable Diffusion 🚀

Today’s top AI Highlights:

1. Deci’s New Model Faster than Stable Diffusion with Same Quality

2. Multi-Agent Planning in Gaming with LLMs

3. “Bard Extensions” and Fact-Checking its Responses

4. New AI Chip for Handling 5T Parameter Models

& so much more!

Read time: 3 mins

Latest Developments 🌍

Deci’s Text-to-Image Model Races Past Stable Diffusion 🚅

Deci introduces DeciDiffusion 1.0, an opensource text-to-image latent diffusion model with 1.02 billion parameters, achieving equal quality to Stable Diffusion v1.5 with 40% fewer iterations.

Key Highlights:

• DeciDiffusion 1.0 outpaces Stable Diffusion v1.5, producing top-tier images in under a second. Deci's Infery SDK further enhances DeciDiffusion's speed, producing images 3x faster than Stable Diffusion v1.5.

• DeciDiffusion uses U-Net-NAS, a more efficient variant of U-Net, reducing computational demands. Specialized training techniques are employed to shorten training time and achieve high-quality results in fewer iterations.

• Its superior computational efficiency ensures a smoother user experience and boasts an impressive reduction of nearly 66% in production costs.

Quick Updates from OpenAI 👇

• OpenAI has introduced "gpt-3.5-turbo-instruct," a new ‘instruction-based’ language model optimized for efficiently following specific instructions, different from the chat-focused GPT-3.5 Turbo, with improved comprehension and task-oriented functionality.

• OpenAI now has a fine-tuning user interface, allowing users to view and create fine-tunes and increased the concurrent training limit from 1 to 3 models.

Multi-Agent Planning with LLMs 🕹️

Researchers at UCLA, Microsoft and Stanford introduce MindAgent, infrastructure designed to assess and enhance Large Language Models' (LLMs) multi-agent planning capabilities for gaming interactions.

Key Highlights:

• Through MindAgent, LLMs like GPT-4 demonstrate zero-shot multi-agent planning with advanced prompting techniques, and showcase generalization skills to coordinate more agents in diverse game domains, including Minecraft.

• The MindAgent infrastructure extends beyond testing, as it can be seamlessly deployed in real-world gaming scenarios.

• Researchers also introduces CUISINEWORLD, a new gaming scenario and benchmark, enabling the evaluation of LLMs' ability to coordinate and schedule multiple agents while collaborating efficiently with human players.

Bard Gets Smarter and More Reliable 😎

Google has updated Bard’s features making it more versatile, collaborative, and capable of delivering high-quality responses across multiple languages and Google services.

Key Highlights:

• "Bard Extensions" now allows Bard to find and display relevant information from Google tools like Gmail, Docs, Drive, Maps, YouTube, Flights, and Hotels within a single conversation. User content from Gmail, Docs, and Drive is not accessed by human reviewers.

• "Google it" button that enables users to verify its responses by searching the web for supporting or contradicting information, helping users double-check the accuracy of Bard's answers.

• Bard's features, such as image uploads, Search images in responses, and modifying Bard's responses, are now available in more than 40 languages.

SambaNova's 5 Trillion-Parameter Model Solution 🔌

SambaNova, a company working on full-stack AI solution, including hardware and software, has unveiled the SN40L, a new custom AI chip designed to handle 5 trillion parameter models.

Key Highlights:

• SN40L chip is claimed to be 30 times more efficient, reducing the number of chips required to run trillion-parameter models from 50-200 to just eight chips.

• The chip is available now and is backward-compatible with previous generation chips.

Tools of the Trade ⚒️

• Graphologue: Transforms LLM’s such as GPT-4 responses into interactive diagrams in real-time with Inline Annotation.

• Dubbah: AI dubbing solution for videos that can clone your voice in 28 different languages, keeping the voice quality, tone and emotion intact.

• FleetWorks: AI-powered logistics that automates phone calls and emails for freight brokers, forwarders, and logistics teams.

• Cardinal: AI-powered product backlog that continuously enriches features with customer feedback and revenue data, syncing customer data, mapping feedback to features, while keeping features' progress up to date.

• Release AI: Enables developers to have meaningful conversations with their AWS and Kubernetes infrastructure, providing instant access to DevOps expertise.

Hot Takes 🔥

• > the year is 2025 > all consumer GPUs have been banned > you can't even "game" without using the "cloud" > frames are streamed to your device > everything is behind an API > requires approval before using the latest "AGI" model > all requests are logged and monitored ~ anton

• Alright I’m convinced. High probability we have AGI in ~ 2 years If our small 5 person team is getting extremely close to an autonomous agent , I can only imagine what folks at openai are up to ~ Sully

Meme of the Day 🤡

That’s all for today!

See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!