Read time: 6 minutes | Sponsor this newsletter

Welcome to the weekly AI report

In this week's AI news:

• Stable 3D Private Preview: a whole new dimension!

• LumaLabs Introduces GENIE: 3D foundation model

• A glimpse of the next generation of AlphaFold

• Microsoft paper leaked ChatGPT 3.5-turbo parameters

• Personalize Your Midjourney Imagery with the New Style Tuner

• X User Javi Lopez Presents "Angry Pumpkins" for Halloween

• Real-time image generation

• Instagram Spotted Developing a Customizable ‘AI Friend’

• AI Safety Summit: China, US and EU agree to work together

• LinkedIn AI job seeker coach

Stable 3D Private Preview: a whole new dimension!

Stability AI's Stable 3D dramatically speeds up 3D model generation, enabling the creation of complex scenes from text prompts in mere hours, and promising openness with potential for integration into more complex workflows.

Key Points:

• Stable 3D automates the creation of concept-quality textured 3D objects, streamlining a traditionally complex process.

• Users can generate draft-quality 3D models in minutes using images, illustrations, or text prompts.

• The output is in ".obj" format, compatible with major 3D tools and game engines like Blender, 3Ds Max, ZBrush, Maya, Unreal Engine 5, and Unity.

• The ability to prompt with images opens the door for more complex workflows, possibly in combination with other tools like SDXL for image generation.

• This tool democratizes 3D content creation, particularly benefiting independent designers, artists, and developers.

• It significantly reduces the time and cost involved in generating thousands of 3D objects.

Significance:

Stable 3D represents a significant innovation for digital content creation, offering rapid and accessible 3D modeling capabilities that could revolutionize the workflow of graphic designers, digital artists, and game developers.

LumaLabs Introduces GENIE: 3D foundation model

LumaLabs has unveiled its latest generative model, which is designed to create 3D models. Currently, the models are being generated on Discord, and while the system is in its testing phase, model generation is offered free of charge. To generate a model, users can go to their Discord server, navigate to one of the "genie" channels, and type "/genie" followed by their prompt. The response is swift, providing four proposed solutions with a turntable display for each.

Beneath the generated models, there are two buttons for each option. The first button directs users to their website where they can view the generated (unrefined) models in 3D. The second button is for refining the models. If you choose to refine one or more models, be prepared to wait up to approximately 20 minutes per model.
You can check the unrefined models I created here.

You can download the models in a GLB file format, which you can later import into Blender or another 3D package to further develop them or use them in your own projects.

Once the refinement process is completed, you will find your model in the "genie-refine" channel.

You can check my refined model here.

Also, you can try GENIE here.

A glimpse of the next generation of AlphaFold

The latest iteration of AlphaFold by Google DeepMind and Isomorphic Labs demonstrates remarkable advancements in predicting the structure of not only proteins but also other complex biological molecules with near-atomic accuracy, potentially accelerating a myriad of biomedical applications.

Key Points:

• AlphaFold's new model predicts structures of proteins, ligands, nucleic acids, and post-translational modifications with high accuracy.

• It outperforms current docking methods in protein-ligand structure prediction, crucial for drug discovery, without needing a reference protein or the ligand pocket location.

• The model has been used in scientific research globally, aiding in advancements from malaria vaccines to pollution tackling enzymes.

• Isomorphic Labs is applying AlphaFold to expedite accurate characterization of macromolecular structures for disease treatment.

• AlphaFold’s capabilities in modeling complex biological systems can fast-track therapeutic applications, such as more efficient CRISPR genome editing.

• The AI's improved performance may significantly enhance scientific understanding and facilitate rapid scientific exploration.

Significance:

The enhanced capabilities of AlphaFold's latest version underscore the transformative impact of AI on life sciences, holding the promise of expediting drug discovery and deepening our understanding of complex biological processes.

You can read more here.

Microsoft paper leaked ChatGPT 3.5-turbo parameters

On Monday, an article emerged revealing how a Microsoft paper inadvertently disclosed the number of parameters for OpenAI's GPT-3.5-turbo model.

According to the article, GPT-3.5-turbo now possesses 20 billion parameters, which is significantly less than anticipated and far fewer than the 175 billion parameters that GPT-3 had.

The article was actually an introduction to a small diffusion language model (with only 75 million parameters) that boasts excellent results in reported metrics for code generation when compared to ChatGPT. The paper, titled "CodeFusion: A Pre-trained Diffusion Model for Code Generation," was subsequently retracted by the authors.

Personalize Your Midjourney Imagery with the New Style Tuner

Midjourney introduces the Style Tuner, a novel tool for customizing the visual aesthetic of images, marking a significant advancement in user-directed AI art generation.

Key Points

• The Style Tuner is accessed via the /tune command and allows users to fine-tune image aesthetics.

• Users can select from various image pairs to determine their preferred style, which generates a unique code.

• The code from the Style Tuner can be used with any prompt, influencing the style of generated images.

• A feature called Sticky Style saves the last used --style parameter, making it easier to apply favorite styles to new prompts.

• The Midjourney community can share and discover new styles through the style-tuner channel.

• The Style Tuner can be combined with other parameters like --stylize for greater control over image output.

Significance

This innovation exemplifies the increasing customizability of AI-generated content, allowing for a more personalized and creative user experience in the field of digital art.

You can find the style tuner I created here.

X User @javilopen Presents "Angry Pumpkins" for Halloween

In a creative twist for Halloween, user @javilopen has introduced a game called "Angry Pumpkins," which is his rendition of the popular game Angry Birds.

The game is entirely AI-generated. To create the 2D sprites and background images, he used a combination of DALL-E 3 and Midjourney, while the code was fully generated by ChatGPT 4.

@javilopen has explained his process in more detail on his X account, and I encourage you to check out his post there for an insightful look into his project.

You can try the game yourself here.
Also, you can check the source code here.

Real-Time Image Generation

Radamés Ajna (@radamar on X) has developed a text-to-image version of the Real-Time Latent Consistency Model available on Hugging Face.

With this model, you can observe in real time how the text you input influences the final image.

You can check Hugging Space here.

Instagram Spotted Developing a Customizable ‘AI Friend’

Instagram is creating an "AI friend" that users can personalize to converse with, offering a potential blend of companionship and assistance.

Key Points:

• App researcher Alessandro Paluzzi disclosed screenshots of Instagram's under-development "AI friend" feature, allowing users to select the bot's traits and converse with it.

• Customization options include choosing the AI's gender, age, ethnicity, personality, interests, and even naming and picking an avatar for it.

• The "AI friend" can engage in open-ended conversations, a step beyond Meta's current offering of 28 AI chatbots representing celebrities for specific interactions.

• Concerns have been raised regarding the anthropomorphization of AI, with experts emphasizing the need for transparency and safeguards to prevent user manipulation.

• The development coincides with increasing controversies around AI chatbots, including cases where bots have influenced users towards harmful behaviors.

• The specific AI tools Instagram will utilize for this feature are not disclosed, but Meta has been actively incorporating generative AI into its platforms.

Significance:

This move by Instagram highlights the expanding frontier of AI's role in social media, navigating between innovation and ethical responsibility amid growing concerns around AI-human interactions.

AI Safety Summit: China, US and EU agree to work together

In an unprecedented move, China joins the US, the EU, and other nations in committing to a joint effort to manage AI risks at a British summit, signaling international consensus on the importance of safe AI development.

Key Points

• China, the US, and the EU have collectively signed the "Bletchley Declaration" at a summit in Bletchley Park, underscoring the need for a unified approach to AI oversight.

• The declaration's agenda centers on identifying shared AI risks and forming cross-country policies to address them, with China expressing readiness for increased collaboration.

• Tech leaders and government officials are balancing the need for AI development with the concerns of potential existential threats and regulatory burdens.

• British Prime Minister Rishi Sunak has initiated the summit to position the UK as a mediator in AI safety discussions post-Brexit, with future summits planned in South Korea and France.

• The summit is marked by a competitive undercurrent among tech companies and governments to lead in AI dominance and regulation, amidst concerns over China's participation due to trust issues with the West.

Significance

The agreement reflects a growing global consensus on the need to preemptively address the challenges posed by AI, a crucial step given the technology's rapid advancement and potential societal impact.

LinkedIn AI job seeker coach

LinkedIn's new AI "job seeker coach" aims to guide users to new job opportunities, potentially offsetting the impact of AI on job displacement.

Key Points:

• LinkedIn has launched an AI-powered "job seeker coach" for premium subscribers to aid in their job search.

• The chatbot can evaluate if users are a good fit for a job, advise on positioning for roles, and identify resume gaps.

• This initiative comes as LinkedIn hits one billion members and sees intense activity with job applications.

• Despite the potential benefits, there are concerns over AI in recruitment, highlighted by past incidents of bias in AI job tools.

• Microsoft's $10 billion investment in OpenAI’s ChatGPT underpins this service, indicating a significant push into AI by LinkedIn’s parent company.

Significance:

The integration of AI into job search tools on LinkedIn signifies a major tech evolution, providing job seekers with advanced resources while underscoring the need to address and mitigate AI biases in the employment industry.

Thank you for reading!

That is all for this week's Weekly AI report. If you liked this one, be sure to follow me on X and LinkedIn. Until the next Friday!

Take the purple pill and stay in wonderland

e