OpenAI’s DevDay Announcements ??, Elon Musk’s Grok ??, and more GenAI News
GPT-4? More like GPT-Snore!
— Elon Musk
Hey there,
The world of AI is advancing fast. Keeping up with the news can be daunting, which is why our newsletter is here to help. We've compiled the most significant developments of the week, offering not only relevant updates but also learning opportunities.
This week's highlights range from practical applications, like Sourceful's innovative packaging solutions (also one of our selected sponsors, take a look or spread the word - you’ll also help the newsletter with that), to major announcements made during OpenAI's DevDay, to Amazon investing into building the largest LLM ever built. What is this all about?
The pace is unrelenting. AI adoption via for instance GPTs (OpenAI’s custom ChatGPTs that everyone can build with ease) is taking off, and it's clear we're still just scratching the surface of what's possible.
Enjoy the ride,
Martin P.S.: Happy Diwali! ???
?? GenAI News
Time is money, especially when launching a new product. Traditional design methods can hold you back, but Sourceful's AI-driven tool Spring is a game changer.
High-Quality Packaging Inspiration in Minutes Generate stunning, unique packaging designs in just three simple steps. Love it? Get it delivered. Want changes? Keep iterating until it's perfect.
OpenAI recently hosted its highly anticipated DevDay event, announcing several significant developments including custom GPTs, GPT-4 Turbo, a GPT builder, and a store among others.
Extended Context Window
GPT-4 is now comprehensively pre-trained with global knowledge up to April 2023, finally. It can handle now over 300 pages of text in one prompt due to its 128k token context window.
Attention: research indicates that Large Language Models (LLMs) face challenges in extracting relevant information from very large contexts, leading to decreased answer quality and increased risk of generating incorrect information. As the input context lengthens, the performance of LLMs, even those designed for long-context tasks, declines significantly.
Therefore, we recommend against fully utilizing the maximum context lengths due to two main reasons: a) it is costly, with prices reaching up to $2 per prompt, and b) it often results in diminished performance.
Improved Multimodality
Further, its multimodal approach is upgraded as it now processes images using the Chat Completions API, enabling applications like caption generation, image analysis, and document figure reading.
A notable success story is Be My Eyes, a Danish startup that has integrated GPT-4 into its app to aid people with blindness or low vision. The app's new Virtual Volunteer feature utilizes GPT-4's multimodal capabilities to analyze images and text, providing real-time visual assistance.
The cost-effectiveness of GPT-4 is highlighted by its pricing, at $0.00765 per image processed at 1080x1080 pixels, which could promote broader AI application development across various domains.
Kickstart AI Adoption via GPTs
Additionally, OpenAI has developed customizable versions of ChatGPT, known as GPTs, for specific applications ranging from everyday tasks to professional and personal uses. These GPTs are easy to create and do not require coding expertise. For more information on setting up a custom GPT, see to Martin’s LinkedIn post. (Feel free to give like. ??)
OpenAI's continuous product development and product velocity demonstrates its commitment to enabling (code-free) AI model creation, significantly advancing AI adoption and leading towards an AI-driven world with conversational digital interactions.
Grok, still in its witty beta diapers after just two months, skillfully juggles a wide array of questions, serving up answers with a dash of humor. And, like a fine wine, it's poised to get better with age (and a bit of user Tender Loving Care).
A relatively small Model
The engine powering Grok, known as Grok-1, represents a state-of-the-art language model developed over four months. Its initial iteration, Grok-0, comprised 33 billion parameters and displayed capabilities comparable to LLaMA 2 (70B), but with only half the training resources. Over the past two months, Grok-1 has seen significant enhancements in reasoning and coding abilities, achieving impressive scores of 63.2% on the HumanEval coding task and 73% on MMLU, reflecting its advanced mathematical and reasoning skills.
Small Team, and Data Advantage
Despite having a relatively small engineering team, xAI has developed a language model that stands on par with others and shows potential for continuous improvement. The model's competitiveness could be further enhanced by its access to X/Twitter data, which may not be available to other models, depending on Elon Musk's decisions. This unique access to X data could provide a significant competitive edge.
Grok-1's capabilities suggest its suitability for a variety of applications, including coding assistance, language comprehension, software development, content creation, customer service, education, and providing up-to-date information.
领英推荐
Runway ML's latest update to their video generation model (Gen-2) brings unprecedented accuracy and consistency to AI-generated videos. This advancement utilizes Video Diffusion Models (VDMs), an extension of stable diffusion models like those in Midjourney, Stability AI, and DALL-E, specifically tailored for video content.
What's New?
The Gen-2 model begins by creating a block of frames and then employs a novel 'gradient method' to add more frames. This method significantly outperforms existing ones, ensuring seamless integration in terms of spatial and temporal continuity.
Future Implications
In just five years, we could see remarkable applications of this technology:
The potential extends into Business Operations, Real Estate, and beyond.
We'd love to hear your thoughts and ideas on this exciting development!
?? Upcoming Workshop
Let’s Meet Next February!
I am thrilled to announce our upcoming workshop, which will take place at the Generative AI for Marketing Summit. It is going to be immersive and
?? 26th of February 2024, ??Chelsea Harbour Hotel, London, UK.
Discover and Unlock:
Transition from theory to actionable insights amidst a like-minded tribe. Reserve your seat now with code: SPEAKER10.
P.S: I am looking forward to meet you. ;)
?? Lastly, One Kind Ask
We aim to enhance each edition of our newsletter. While we've concentrated on delivering well-researched content, your feedback is invaluable. Would you prefer more technical details, high-level product insights, or information about our workshops? Let us know what you'd like to gain from this newsletter by simply replying to this email.
Thank you so much.
We possess all the necessary tools to integrate advanced forms of intelligence, such as humanoid robots and AGI, into our world, fostering evolution rather than revolution. Now, let’s hope for the right collective mindset.
Subscribing to, giving feedback about, and sharing the newsletter as well as our renowned online course will be highly appreciated and helps a lot. ??
Our new Generative AI + Marketing Online Course .
Our upcoming book Generative AI: Navigating the Course to the Artificial General Intelligence Future .
If you would like to sponsor an ad to this 30k+ newsletter, please, respond to this email.
Thank you so much for reading,
Martin
Head of Content, IQPC R+D
1 年Great post Martin - Looking forward to working with you!
--
1 年Found it very informative as AI Enthusiast????, looking forward to many more??
Head of Human Resources - Hewlett Packard Enterprise
1 年Great article, looking forward to the future
Managing Director at IQPC - Innovation + Emerging Tech
1 年Martin Musiol great to have you involved again!