Mark Ghuneim

• Hugging Face’s website that allows users to explore Llama v2, a large language model, with TGI (Transformers for Interactive Generation) demos. Try running it on your MacBook!
• Demo for LLaMA-2-70B is hosted on HuggingFace or chat with it on perplexity.ai
  
• Google’s image generation model available on Googlecloud API for external users.
• Blog post by Stephen Wolfram discussing generative AI, space, and the mental imagery of alien minds.
• Stability AI’s website that describes their research on “Minds Eye,” a project related to AI image generation.
  
• Islamov’s website that discusses stable diffusion web GPU, which is likely related to AI image generation.
  
• The Atlantic discussing Hollywood’s strategy for dealing with the pandemic.
• An announcement that Llama-v2 is now open source and available for commercial use, with links to the model’s website and paper, as well as a list of industry and academic personalities who have endorsed the open-source approach..
• A number of personalities from industry and academia have endorsed our open source approach
• Digging the visuals of this simulation of a simple biome

In another lifetime, I edited video on 3/4 inch media on Sony RM-440 editing machines.  If we wanted something like this prompt generates we had to do a table top shoot that had manual physical panning and tilting.  The next step would be the editing – this took hours if not days.   Here we have it in a paragraph of text.

I have a high-resolution panoramic image that I’d like to convert into a video. The video should have a 3:2 aspect ratio and the image should fill the entire video, even if some of the sides are cut off. The video should start at the center of the image, pan smoothly to the right until it reaches the edge, then return to the center. From there, it should pan smoothly to the left until it reaches the other edge, and then return to the center again. The video should be created using the imageio library in Python, saving the frames directly to a video file instead of storing them in a list. The frame step should be 8 pixels. If necessary, the edges of the image can be cropped so that the size of the image is divisible by the frame step. Please find the image attached image.

 META’s take on Twitter

Meta has a long history of mimicking products and product features of competitors.
Most if not all of the efforts have been met with meh.  Will this time be different?

We are in a brief moment of time between the products announcement and its debut.
What are the expectations?  There are some clear wins that could help insure success.  

• – Keeping the Core Functionality: To make Threads a game-changer, it must retain Twitter’s essential features that made it successful: real-time, open, and free. These three elements formed the backbone of Twitter’s survival until it fell into the wrong hands, eroding its legacy value and driving away intelligent users. 
• – Embracing Real-Time Openness: Streamlining Onboarding and Data Usage: To make transitioning to Threads seamless, an import function to import social graphs and a user-friendly API are crucial. Users should be able to effortlessly bring their connections from other platforms to Threads. 
• – Privacy Opportunity – Be there, be square, don’t be a data pig. Prioritize user privacy and avoid excessive data collection,  EU regulations and others are scoping to new data realities for platforms.  Why collect unnecessary information. Be smart about retention and use cases.  Ban individual targeted advetisting from the start.
• – Utility Fork for Governments and Organizations: To cater to different user segments, Threads can create a separate utility fork designed specifically for governments and organizations. This fork can include features such as user agent identification, media authentication, and dedicated tools tailored to their unique needs, making Threads a versatile platform for a wider user base. 
• – Empowering Users with Curation and Tuning Features:  Give users the ability to curate their content and fine-tune their feeds. Allowing users to filter out unwanted content and customize their preferences will enhance their sense of ownership and personalization, leading to a more enjoyable and tailored experience.   Twitter tried to this a platform level which was not the optimal path. 
• – Enhancing Functionality: Bookmarks for saving and revisiting important posts will greatly enhance the user experience. Improved video uploading capabilities to ensure a smooth and hassle-free process, along with realistic durations, 
•  – Gradual Monetization Approach:  The platform should be ad-free for a initial period of time, allowing users to fully appreciate its value. After this initial period, a commercial layer can be introduced, ensuring ads are relevant but without invasive targeting, striking a balance between revenue generation and preserving the organic user experience.

Open Letter from Security and Privacy Researchers in relation to the Online Safety Bill

The economic potential of generative AI: The next productivity frontier

Stability AI CEO Emad Mostaque, in an interview with Peter H. Diamandis for the Moonshots and Mindsets Podcast, said that in the next five years, there will be no human programmers. He also said that by the end of next year (2024), ChatGPT will be available on phones and won’t require an internet connection to run. The CEO also highlighted how AI is growing at a rapid pace.

Google the AI  scraping game is on    publicly accessible sources  For example, we may collect information that’s publicly available online or from other public sources to help train Google’s languageAI models and build products and features like Google Translate, Bard, and Cloud AI capabilities. Or, if your business’s information appears on a website, we may index and display it on Google services.

Open-source “Davids” are taking on GPT-4 and other Goliaths Open-source LLMs are a possible antidote to Microsoft and Google’s control of chatbots like GPT-4

We’re starting to see the first open-source *video* generation models 

Links from today…

Voice actors are discovering that decades-old contracts have given companies the right to use their voices to train AI models that compete with them

Platform-agnostic AI models. Inflection AI just got $1.3 billion from Microsoft, Nvidia, Bill Gates, and other investors

Segment Anything Meets Point Tracking
https://github.com/SysCV/sam-pt
https://arxiv.org/abs/2307.01197

AI and the automation of work  [ben evans]

Uncensored Chatbots Provoke a Fracas Over Free Speech  [nyt]

Accurately predicting hit songs using neurophysiology and machine learning

MusicGen is state-of-the-art model for music generation by Meta AI 
MusicGen in 🤗 Transformers

https://github.com/Stability-AI/generative-models/blob/main/assets/sdxl_report.pdf

To say the pace of this sector is moving fast is an understatement :  Links to key advances,  moves and developments.

Microsoft has integrated AI-powered shopping tools into Bing search and Edge, including AI-curated buying guides and AI-powered review summaries.

Salesforce AI Research has released the XGen-7B, an open-source 7B LLM (Large Language Model) trained with 1.5 trillion tokens over an 8K input sequence length.

DreamDiffusion is a groundbreaking technique that generates high-quality images directly from brain EEG signals, bypassing the need to translate thoughts into text.

Microsoft has announced an AI Skills Initiative, offering free coursework in partnership with LinkedIn, a global grant challenge, and increased access to free digital learning events and resources for AI education.

Stability AI has released OpenFlamingo V2, an open-source reproduction of OpenAI’s Flamingo model with over 80% of the original performance.

Unity has introduced two AI-powered tools: Unity Muse, which generates animations and sprites using text and sketches, and Unity Sentis, which embeds an AI model into Unity Runtime for games and applications.

ElevenLabs is launching Voice Library, a platform for sharing AI-generated voices created with their voice design tool.

Merlyn Mind has released three open-source, education-specific LLMs (Large Language Models) aimed at revolutionizing the learning experience.

Amazon’s AWS has launched the Generative AI Innovation Center, a $100 million initiative leveraging AWS machine learning and AI expertise to develop and implement generative AI solutions for businesses.

Zeroscope_v2 XL is an open-source text-to-video AI model that delivers high-quality videos without watermarks.

MotionGPT is a new motion-language model designed for various motion-centric tasks.

Databricks is set to acquire the open-source startup MosaicML for $1.3 billion. MosaicML has released the MPT-30B, an open-source model licensed for commercial use.

According to Indeed’s data, generative AI-related job postings in the US increased by around 20% in May.

The DragGAN algorithm source code has been released, enabling interactive point-based manipulation on the generative image manifold.

Baidu’s ERNIE 3.5, a foundation model, outperforms ChatGPT (3.5) and even GPT-4 in several Chinese language capabilities.

Google is organizing the first-ever Machine Unlearning Challenge on Kaggle.

Adobe is offering to cover claims in lawsuits related to the use of content generated by Adobe Firefly, their generative AI image tool.

Google is launching generative AI coding features exclusively for Pro+ subscribers in the US on Google Colab.