- Creative AI
- Posts
- The superintelligence is coming
Good day, humans
In today's Newsletter, we include:
Let's dive in!
Building a Superintelligence
OpenAI Co-Founder Aims to Build a Safe Superintelligence
The OpenAI saga rivals Game of Thrones in drama.
The Summary:
Chapter 1: Ilya Sutskever, co-founder and chief scientist of OpenAI, was part of the board that fired CEO Sam Altman last year, citing concerns about the safety of the technology being developed.
Chapter 2: Ilya eventually reversed his decision and supported Altman's return.
Chapter 3: Ilya has not made public statements or appearances since, leading to intense speculation about his role, which ended last month when he announced his departure from OpenAI with part of his team.
Today's Chapter: The plot thickens as Ilya announced he is forming a new company called Safe Superintelligence (SSI) with his team, aiming to develop a safe superintelligence.
Why It Matters: This announcement is intriguing because the company does not plan to develop any AI products or services. They are solely focused on building a superintelligence. No one knows why, leading to much speculation. One thing is clear: the SSI team believes superintelligence is within reach and that they can achieve it safely before anyone else.
Sonnet from Anthropic is the New Champion of Chatbots
Anthropic Launches a New Model Surpassing GPT-4o
The competition among chatbots is a marathon, and there is now a new leader.
Anthropic, a three-year-old startup valued much lower than OpenAI, has launched its new model, Claude 3.5 Sonnet, which is outperforming ChatGPTo and other competitors. According to all indications, this latest version works better than both Gemini 1.5 Pro and the highly publicized GPT-4o in several key text and vision metrics.
About Anthropic:
Haiku: More compact and efficiency-oriented, open to the public.
Sonnet: Balanced between Haiku and Opus.
Opus: More powerful, with a cost.
Currently, only a new Sonnet model is being launched, with Haiku and Opus presumably on the way. For now, Sonnet is open to everyone.
Why It Matters: Sonnet also introduces Artifacts, facilitating real-time editing of AI results like code or designs within the chat. Artifacts enable collaborative refinement of prototypes and workflows, evolving Claude from a chat tool to a design/development ally.
Ready for Gen-3 Alpha
Runway Announces Gen-3 Alpha, Their Most Advanced Video Model
Runway has just announced Gen-3 Alpha, its latest and most advanced video generation model. Trained on both videos and images, it promises significant improvements in fidelity, consistency, and movement compared to its predecessor, Gen-2.
Top 5 Video AI Companies Ranking:
DreamMachine from Luma Labs: Top quality, open to the public for free. Current leader pending Runway's release.
Runway: Top quality in Gen-3, not yet available to the public. Gen-2 is available.
Kling: Expected to produce high-quality outputs but currently available by invitation only.
Sora from OpenAI: Expected to produce high-quality outputs but has not been publicly available for over a month.
Pika: No announced model with the quality of this new generation. Awaiting their next move.
Why It Matters: For now, we can only be excited. Gen-3 Alpha is not yet available, but Runway promised it would be open to everyone this week. When launched, Runway will equip the model with reinforced security measures, including a new moderation system and provenance standards. Video AI is reaching a point where people could be misled into thinking it is real.
Hedra Makes Your Characters Talk and Sing
Hedra has announced Character-1, an AI video model that allows users to create synchronized videos from audio and an image, providing greater levels of expression and movement. Although currently limited to 30 seconds and offering low-resolution square format videos, Character-1 stands out for its ability to create characters that talk, sing, and rap, promising to revolutionize multimedia content creation.
How to Use It: Character-1 is free during its Beta phase and allows the creation of videos of any length. You can easily generate characters from photographs, AI images, or text and create synchronized videos with the provided audio.
Go to ElevenLabs and type what you want your character to say. Choose a voice and download the free mp3 file.
Go to Hedra and upload the mp3 file and the image of the character you want to speak.
Download your file.
Midjourney has millions of styles. And now you can mix and blend them too.
Midjourney Allows Style Mixing
Midjourney, the renowned AI art generation platform, has announced new advanced options for its style reference (--sref) and model personalization (--p) functions. These enhancements allow users to mix and personalize their creations in more sophisticated ways, offering greater possibilities for artistic expression.
How to Use It:
Our base prompt is a photographic portrait in 3:4 format for version 6: Photographic Portrait --ar 3:4 --v 6.0
Adding a Single Style:
Photographic Portrait --ar 3:4 --v 6.0 --sref 3319970161
To Mix Two Styles:
Photographic Portrait --ar 3:4 --v 6.0 --sref 3319970161 2667828500
See the results:
Thanks for reading. See you next week!
Hi 👋 I’m Erik Knobl, Product Designer by day and Generative AI explorer on weekends. I share my learnings in this newsletter. Consider subscribing to stay in touch.