🍎 Apple dropped an AI bomb for Christmas 🎄

Merry Christmas!

It’s that time of the year again and if you are reading this week’s newsletter on Christmas, we know you are a diehard AI fanatic 😎

AI doesn’t care about Christmas so it hasn’t stopped moving at warp speed. let’s dive into this week’s nitty gritty.

Table of contents

• Apple is about to take over the AI landscape 🍎

• Anthropic is really stepping up their game 👀

• The impact of AI on marketplaces 🖥️

• Prompt based programming? 🤯

• AI stepping into the world of law ⚖️

• Google’s Gemini Pro is everywhere

Updates

Apple is about to take over the AI landscape

Apple's latest AI research has introduced new techniques for 3D avatars and efficient language model inference, which could transform iPhones and iPads into more immersive and AI-capable devices. The advancements include:

1. HUGS (Human Gaussian Splats): This method generates animated 3D avatars from short monocular videos, enabling more realistic and efficient representation of humans and their environment. HUGS is up to 100 times faster in training and rendering compared to previous avatar generation methods[1].

2. Inference Cost Model: Apple researchers addressed the challenge of deploying large language models (LLMs) on devices with limited memory by constructing an inference cost model that harmonizes with flash memory behavior. The proposed system minimizes data transfer from flash storage into scarce DRAM during inference, improving inference latency by 4-5x on an Apple M1 Max CPU and 20-25x on a GPU.

These innovations demonstrate Apple's growing leadership in AI research and applications, and could potentially set the stage for a new class of applications and services that leverage the power of LLMs in a way that was previously unfeasible.

Anthropic is really stepping up their game

Anthropic PBC, an AI startup backed by Amazon and Google, is reportedly in talks to raise $750 million in new funding at a valuation of $15 billion, with Menlo Ventures Management expected to lead the round. The deal's ultimate valuation may exceed $18 billion.

This company has their own version of ChatGPT, it’s called Claude and it’s a good alternative to the paid GPT-4. It has a 200k token context window which translates to about 500 pages of information. This is more than GPT-4 which ‘only’ has 128k token context window.

To improve the model's performance when answering questions based on individual sentences in documents, users can try adding a minor prompting edit, such as "Here is the most relevant sentence in the context:" to the start of the model's response. This approach directs the model to look for relevant sentences first and overrides its reluctance to answer based on a single sentence, especially one that appears out of place in a longer document.

The impact of AI on marketplaces

The marketplace business model has evolved over time, with the internet and mobile apps revolutionizing the way transactions happen. Generative AI (GenAI) is now promising to revolutionize marketplaces again, transforming how products and services are sold and made, and creating a better user experience for buyers. The impact of AI on marketplaces can be divided into two main categories:

1. Impact on buyers (demand side):

• Improved search experience: Buyers can use natural language, images, or even vibes to describe what they're looking for, reducing scrolling fatigue and increasing the likelihood of a search converting into a transaction.

• Personalized product recommendations: AI can help match users with suppliers who can bring their vision to life, at the quality level and price they are looking for.

2. Impact on sellers (supply side):

• Efficiency improvements: AI can automate busywork, such as writing listings and answering questions from potential customers, allowing suppliers to focus on the transaction.

• AI-native suppliers: Some marketplaces may pivot to focus on AI-native suppliers, offering their services at much lower prices and trying to make up the difference in volume.

Different types of marketplaces will be impacted by GenAI in varying degrees, with some facing existential challenges from AI, while others will be supercharged by it. For example, marketplaces like Upwork and Fiverr may struggle as buyers generate their own products or services using AI tools, while platforms like Canva, Cameo, and Patreon can leverage AI to enhance their suppliers' offerings and create more engaging content.

In conclusion, the future of marketplaces in the AI era will see more liquid and stable marketplaces, with more listings, higher conversion rates to transactions, and more repeat purchases than ever before. As AI continues to evolve, marketplaces will need to adapt and find ways to leverage AI to improve their businesses and stay competitive.

🛠️ Tool of the week 🗓️

“Rapid prototyping – with speed you haven't seen before” is what GPT Engineer claims. They have developed a tool that programs for you based on prompts. Even though GPT-4 has capabilities to code, GPT Engineer takes this to a whole new level.

Have a look at a demo video here.

More updates

AI stepping in the world of law

Over the past year, Harvey has established itself as a secure generative AI platform for sophisticated professional services and has earned the trust of leading law firms, in-house teams, professional service providers and private equity firms. Engagement on the platform has increased exponentially, revenue has risen more than tenfold since April and we have assembled a world-class, multidisciplinary team.

The company has raised $80 million in Series B funding. The funding was led by Elad Gil and Kleiner Perkins, with participation from OpenAI Startup Fund and Sequoia. The investment brings Harvey’s total funding to over $100 million and values the company at $715 million. The post expresses gratitude to the existing investors and customers, and outlines the company's growth and plans for the future, including expanding custom model building, scaling the team, and building out product features.

Google’s Gemini Pro is everywhere

Google has launched Gemini Pro on Vertex AI, a lightweight version of the more capable Gemini Ultra model. It is now accessible in public preview for Google Cloud customers using Vertex AI.Key points include:

• Gemini Pro API: By default, the Gemini Pro API in Vertex accepts text as input and generates text as output, similar to generative text model APIs like Anthropic's, AI21's, and Cohere's. An additional endpoint, Gemini Pro Vision, can process text and imagery, including photos and videos, and output text.

• Image Processing: Gemini Pro Vision addresses one of the major criticisms of Gemini, which is its inability to accept images despite being "multimodal" (i.e., trained on a range of data including text, images, videos, and audio).

• Customization: Within Vertex AI, developers can customize Gemini Pro to specific contexts and use cases leveraging the same fine-tuning tools available for other Vertex-hosted models, like Google's PaLM 2.

• Pricing: Input for Gemini Pro on Vertex AI will cost $0.0025 per character while output will cost $0.00005 per character. This is reduced 4x and 2x, respectively, from the pricing for Gemini Pro's predecessor, and Gemini Pro is free to try for Vertex AI customers until early next year.

• Future Features: Google plans to introduce Gemini Pro-powered conversational voice and chat agents, as well as search summarization, recommendation, and answer generation features in Vertex AI, drawing on documents across modalities and sources.