Bot Eat Brain
Posts
Dolly 2.0, The world's first truly open LLM

Dolly 2.0, The world's first truly open LLM

PLUS: OpenAI's Latest Game Changer

Anthony Castrio
April 13, 2023

Good morning human brains, welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

OpenAI's new text-to-image breakthrough 🖼️
The world's first truly open LLM 🐑
Why did Elon Musk just buy 10,000 GPUs? 🤔
AI clones teen girl’s voice in $1M kidnapping scam. 🦹

Follow @BotEatBrain on Twitter for breaking news.

P.S. We are hiring researchers and writers, click here for more info.

APPETIZER

OpenAI's Latest Game Changer

Say hello to the future of AI image generation.

OpenAI has just released the codebase for Consistency Models, a new text-to-image generation technique that’s much faster than diffusion.

They've ditched the slow, iterative process of diffusion models for a speedy one-step approach. No more waiting for ages to get mind-blowing results, these bad boys deliver quality FAST.

Consistency models completely skip over several intermediary steps making them much faster than previous models.

Alright, let's break it down:

1/ They’re fast. Consistency models will make creating images, audio, and video much faster.

2/ They’re versatile. Much like diffusion-based models, they can handle tasks like image inpainting, colorization, and super-resolution without needing extra training.

3/ They’re still nascent. The example images included in OpenAI’s research paper show that the model is not yet tuned for commercial use.

Images generated with OpenAI’s Consistency Model.

BUZZWORD OF THE DAY

Stable Diffusion

A deep learning tool from 2022 that turns text descriptions into images. Created by Stability AI and partners, it's a type of deep generative neural network that works on standard computers with a decent GPU. Unlike earlier models like DALL-E and Midjourney, Stable Diffusion is available for public use, not just on cloud services.

FROM OUR PARTNERS

Build MVPs without Dev Headaches

Planning is the most important part of product development. You have a clear product vision, but how can you deliver the right software on time and within budget?

Get your product playbook.

Enter stage left: The first product workshop to build better MVPs.

Choose only features that matter and add value
Create a roadmap for all development phases
Receive a professional document at the end of the workshop.

Save hundreds of hours in development costs with fully fleshed-out plans, templates, and SOPs crafted for non-techies.

MAIN COURSE

Free Dolly: the world's first truly open LLM.

Databricks spent the last two weeks co-opting their entire 5,000-person workforce to manually train an open-source LLM that can compete head-to-head with models like GPT and LLaMA.

The result: Dolly 2.0

Announced just yesterday, the dataset, dataset licensing, training code, and model weights are all open-source and suitable for commercial use, enabling organizations to develop and customize their LLMs without depending on third-party API access or sharing data.

What’s more, Databricks has released smaller models that can fit on consumer-grade GPUs you can comfortably run on your laptop.

@vagabondjack, Applied AI at Databricks

Why it’s a big deal:

1/ Power to the people. Dolly 2.0 is open-source, so more people can play with AI and create cool stuff without paying for license fees or API access.

2/ Keeps secrets safe. Companies can make their own AI without spilling their valuable data to third parties.

3/ DIY domain expertise. You can fine-tune an AI for specific applications without relying on one-size-fits-all models.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

AI clones teen girl’s voice in $1M kidnapping scam: ‘I’ve got your daughter’

Replacing my best friends with an LLM trained on 500,000 group chat messages

Artificial neural nets are good enough. Neural Networks seem to be good enough at simulating human neurons for consciousness.

Startup News

Elon Musk is building a generative AI project at Twitter powered by at least 10,000 GPU’s.

Alphabet leads $100 million investment in AI startup AlphaSense.

Betaworks announces camp for early-stage AI startups, offers $500,000 in funding.

Research

Stanford releases a leaderboard for LLMs across 42 scenarios and 57 metrics.

From Deep to Long Learning: Prompt lengths that can be used in LLMs are going to get orders of magnitudes longer very soon, giving LLMs the ability to make conclusions across entire books or even libraries without fine-tuning.

Tools

Edit Anything by Segment-Anything: Re-imagine specific sections of an image, like the sky, or a background character. (open-source)

Dexa: AI-Powered Podcast Assistants. Explore, search, and ask questions using AI bots powered by your favorite podcasts.

Landing AI. Make beautiful websites in seconds. Explain your product & branding and get a unique landing page generated by AI.

TWEET OF THE DAY

AI ART-SHOW

“untitled” - @EliLeno

Until next time 🤖😋🧠