OpenAI Co-founder: "We Were Wrong"

"ClosedAI", DALL-E 3, Midjourney's Hand Problem, Microsoft 365 Copilot and More

Nicely done! You are about to get caught up on all the AI related tech and news from the last few days.

Here's what we got for you today:

  • šŸ” OpenAI Behaving More Like ā€œClosedAIā€

  • šŸ–ļø Midjourney v5 Alpha Still Canā€™t Do Hands

  • šŸ‘©ā€āœˆļø Microsoft 365 Copilot

  • šŸ‘ OpenAIā€™s DALL-E 3 Coming Soon?

  • šŸ¤Æ Cool A.I. Tools and News 

  • šŸ–¼ļø  A.I. Generated Image of the Day

  • āš”ļø A.I. Meme of the Day

    OpenAI Behaving More Like ā€œClosedAIā€

Before its release, there was a lot of talk about the size of GPT-4's training data. The general expectation was that GPT-4 would outperform GPT-3 with its impressive 100 trillion parameters, compared to GPT-3's 175 billion parameters. The idea was that training on a larger and more diverse dataset would make the model even smarter and more accurate for various types of tasks.

Never confirmed by OpenAI

OpenAI has officially confirmed what many already suspected - they will no longer be sharing detailed information about their AI technology for two main reasons. The first reason is to maintain their competitive edge, while the second reason is related to safety implications.

Given both the competitive landscape and the safety implications of large-scale models like GPT-4, this report contains no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar.

ā€œWe were wrong. Flat out, we were wrong. If you believe, as we do, that at some point, AI ā€” AGI ā€” is going to be extremely, unbelievably potent, then it just does not make sense to open-source. It is a bad idea... I fully expect that in a few years itā€™s going to be completely obvious to everyone that open-sourcing AI is just not wise.ā€

Midjourney v5 Alpha Still Canā€™t Do Hands

If you're curious about whether an image was created by AI, pay attention to the hands and eyes. Although Midjourney v5 Alpha is an impressive text-to-image generator, it still struggles with getting the hands right for more complex gestures.

Check out the results for the prompt "girl making heart with fingers --v 5" - it's impressive, but the hands still look creepy, right? So, artists, don't worry, your jobs are safe for a few more months!

girl making heart with fingers --v 5

Having worked with a variety of AI models for the past year, I can confidently say that Midjourney outperforms the competition when it comes to generating high-quality images. That being said, there are a couple of areas where it could improve:

  • Dedicated website for image generation. Although the current Discord server does the job, it's not the most user-friendly platform out there.

  • An API for developers would be a welcome addition.

According to the Midjourney team, this is not the final version of v5.

DO NOT rely on this exact model being available in the future. It will be significantly modified as we take v5 to full release.

ā€” Midjourney

Microsoft 365 Copilot

Playing around with AI-generated images and using ChatGPT for writing has been a blast, but it's becoming increasingly clear why Microsoft invested in OpenAI and the potential benefits of integrating AI with the tools we use on a daily basis. Check out the summary below, and if you want to learn more, click here.

Copilot is integrated into Microsoft 365 in two ways. It works alongside you, embedded in the Microsoft 365 apps you use every day ā€” Word, Excel, PowerPoint, Outlook, Teams and more ā€” to unleash creativity, unlock productivity and uplevel skills. Today weā€™re also announcing an entirely new experience: Business Chat. Business Chat works across the LLM, the Microsoft 365 apps, and your data ā€” your calendar, emails, chats, documents, meetings and contacts ā€” to do things youā€™ve never been able to do before. You can give it natural language prompts like ā€œTell my team how we updated the product strategy,ā€ and it will generate a status update based on the morningā€™s meetings, emails and chat threads.

ā€” Official Microsoft Blog

OpenAIā€™s DALL-E 3 Coming Soon?

There are a lot of tech giants out there with massive resources and funding, but what sets OpenAI apart is their commitment to shipping and doing it frequently. It's pretty impressive when you consider that they've managed to accomplish so much with a team of only about 375 employees.

In the last few months alone, OpenAI has released a number of notable advancements in the field of AI. However, there's one product that's fallen a little behind - DALL-E. Text-to-image apps like Midjourney and Stable Diffusion have actually surpassed it in many ways. But don't fret, the DALL-E team released an experimental version of DALL-E 2 back in early March and we're all eagerly waiting for the full release of DALL-E 3, especially now that GPT-4 is out.

So, props to the OpenAI team for their hard work and dedication to pushing the boundaries of AI.

Hi from the DALLĀ·E team!

Since the early days of DALLĀ·E (almost a year ago!), your feedback has been key to helping us develop the DALLĀ·E model, product and policies into what the world uses today.

Today weā€™re sharing an experimental model with you and a small group of other Labs users for early feedback on how useful the model is, the quality of the generations, and any safety issues or instances of bias you notice. This new model is work in progress as part of the next update to DALLĀ·E. Youā€™ll notice some improvements, such as sharper image quality and increased photorealism for faces.

Weā€™ll only have this model up and running for about 10 days while we collect feedback. After that point, weā€™ll take it down, work on it internally and use our learnings to improve the next version of DALLĀ·E.

Youā€™re welcome to share your creations publicly ā€” please just keep in mind that most users do not have access to the preview.

Thank you for being a part of the DALLĀ·E community! Weā€™re excited to hear what you think.

- the DALLĀ·E team

Hey, today's newsletter is brought to you by Konjer!

With Konjer, you can ask a book to tell you about itself, allowing you to revisit old ideas or discover new ones. This app lets you have a conversation with 50 of the best-recommended books on topics like business, startups, mindfulness, and more. Start absorbing knowledge in a whole new way with Konjer!

https://www.konjer.xyz/

Cool A.I. Tools, News and Demos

The world of AI and machine learning is moving incredibly fast, and it's tough to keep up with all the cool new developments! Here are some of the most exciting AI-powered tools and news.

100% A.I. Generated Image of the Day

Midjourney Prompt - medium-full shot of an elderly chinese woman with deep wrinkles and a warm smile, sitting in a charming soho cafe filled with plants, looking out the window as people walk by, wearing a bright pastel linen blazer and floral print silk blouse, natural afternoon light shining through the windows & reflecting off her eyeglasses, shot on Agfa Vista 200, side-angle view, 4k --ar 16:9 --stylize 1000 --v 5

A.I. Meme of the Day

Let's give credit where credit is due - AI-generated imagery is still pretty darn cool. It's just a tad bit creepy when those bizarre-looking hands start reaching out from the screen towards you.

Peace out,

Daily Zap Team

How much did you enjoy this email?

Login or Subscribe to participate in polls.