The Rundown AI
Posts
Open-source AI on the rise

Open-source AI on the rise

PLUS: GPT-4V crushes 'faked' Gemini demo...

Rowan Cheung
December 12, 2023

Sign Up | Advertise | Jobs | Tools

Welcome, AI enthusiasts.

French startup Mistral just flexed some serious AI muscle — dropping a powerful new model alongside a whopping $415M raise at a $2B valuation.

With an impressive model and major funding — is there a new AI powerhouse brewing in France? Let’s investigate…

In today’s AI rundown:

Mistral announces new open-source model
Creator replicates Gemini demo with GPT-4V
9 new AI tools to try
How to level up your notes with NotebookLM
Google explores using AI for your life story
Stanford researchers release W.A.L.T. for AI video
Meta debuts Audiobox demo for AI voice, sound effects

Read time: 4 minutes

LATEST DEVELOPMENTS

MISTRAL

🚀 Mistral announces new open-source model

Image source: Mistral

The Rundown: French startup Mistral AI just released Mixtral, an open-source 45B parameter AI model using advanced “mixture of experts” architecture to match rivals with far lower cost — while also announcing an eye-popping $415M Series A round that values the company at $2B.

The details:

Mixtral matches or outperforms models like LLaMA 2 and GPT-3.5 on most benchmarks while running 6x faster.
It significantly improves on past Mistral multilingual and scientific reasoning abilities in areas like math and code generation.
Mistral also introduced La Plateforme, which provides API access to hosted versions of Mixtral and its other models to quickly build apps.

Why it matters: With lean team of top researchers and rapid iterations, Mistral continues pushing boundaries on efficient yet powerful open models. Notably, the new $415 Series A round comes about 6 months after a $113m seed round — showing serious money is flooding into open-source.

You can try compare Mistral’s new model to GPT here.

TOGETHER WITH NOTION AI

🧠 A search bar for your second brain

The Rundown: Unlock new levels of productivity with Notion Q&A - your personal assistant in the digital realm.

Notion’s new Q&A feature lets you ask any question about your entire Notion workspace:

What action items came out of my meeting with my manager last week?
Where can I find my class notes on the nervous system?
Summarize the activities I have planned for my trip to London

Elevate your workflow with instant knowledge – ask, explore, and let Notion Assistant be your key to seamless productivity. Dive into your workspace and test it out today.

GEMINI OPENAI

🎥 Creator replicates Gemini demo with GPT-4V

Image source: Greg Technology on YouTube

The Rundown: In the wake of Google’s backlash around the ‘faked’ Gemini demo (covered yesterday), creator Greg.Technology just replicated the video using GPT-4V in real-time — prompting without edits to highlight how far ahead GPT-4V is when even compared to the staged Gemini demo.

The details:

Greg recreated key parts of the polarizing Gemini demo, which wowed viewers before being exposed as a misleading representation of the model’s capabilities.
Greg voiced over GPT-4V generating live reactions to his drawings, emoji signals and games, achieving a responsive back-and-forth.
He published the code and acknowledged the remake wasn't as polished as Google's production, but demonstrated ChatGPT's prowess in real-time.

The relevance: GPT-4 looks to still set the gold standard, excelling at voice tasks even without edits. Videos like this make the Gemini misstep look even worse — presenting serious questions heading into a critical launch for Google.

TRENDING TOOLS

🤖 Tars Copilot- Stop worrying about project management for good - let Tars (inspired by Interstellar) handle your tasks, deadlines and project status (link)*

🎥 Dubverse- Easiest (and magical) way to dub videos (link)

🔥 Krater.ai- Your one-stop AI powerhouse for a multitude of tasks. (link)

🌄 Freepik Pikaso- Real-time AI-powered sketch-to-image conversion (link)

💼 BoldDesk- An affordable Zendesk & Freshdesk alternative at half the cost (link)

📖 Depth Tale- Craft interactive stories easily with a modern, user-friendly platform (link)

🎧 Xound- AI-powered tool ensuring pristine audio and pitch for content creators (link)

📊 Sheet AI- Put AI in your Google Sheets (link)

🕵️‍♂️ GPT Zero- Detect whether an essay is ChatGPT or human-written (link)

Browse the most popular tools ever featured with our tool database.

* = This is sponsored content

AI TRAINING

📓 How to level up your notes with NotebookLM

Google’s AI note-taking app just got a major upgrade with Gemini — and its available for free. Here’s how to take advantage of this productivity hack:

Get access: Sign up for NotebookLM here. If you're outside the U.S., just use a VPN or Opera Browser.

Add your docs: Click on "New source" and select the Docs you want to add to your Notebook (as many as you want). NotebookLM will automatically give you a summary or "Source Guide" for the Docs you upload.

Ask questions: In your Source Guide box, you'll also see auto-generated questions to try — it’s the ultimate ChatPDF. You can also opt to ask your own questions.

Pin quotes: NotebookLM automatically shares citations from your sources whenever it answers a question — so you can quickly jump from a citation to the source, letting you see the quote in its original context.

Change formats: NotebookLM also helps organize your notes into structured documents — simply select a set of notes you’ve collected and ask NotebookLM to create something new, like a study guide or email newsletter.

Take your notes to the next level with this amazing AI organizer!

QUICK HITS

🎙️ Meta debuts demo for AI voice, sound effects

Meta just launched a demo of Audiobox, its foundation model for generating highly realistic, voices, sound effects, and music — created from a combination of natural text prompts and/or voice inputs. Try the model for yourself here.

QUESTION OF THE DAY

What do you think is the most impactful potential use case of AI audio tools like Voicebox?

Vote below (click for live results)

📖 Google explores using AI for your life story

Demo drama aside, Google has reportedly floated "Project Ellmann" to leverage Gemini AI and generate users' life events from just data from your phone. Envisioning AI that knows "everything about your life," features could include chatbots detailing your bio and travel history from photos.

🎥 Stanford researchers release W.A.L.T. for AI video

Stanford researchers just introduced W.A.L.T., an AI system that generates photorealistic and consistent video from text prompts or still images — while also capable of outputs with consistent 3D camera motion. Check out some of the the wild demo generations here.