Read time: 10 minutes

Welcome back, Fellow Futurist !

It’s been a busy week with multiple news on AI front coming each day. From funding of strategically important startups in GenAI space to Google to Meta to IBM. Therefore, this week’s newsletter will be tad bit on the lengthier side.

Down to business. After being on the sidelines for a while, Google is back with a boom and a bang. The current pace of transformative advancements in AI is just insane. After numerous delays and a rumored 2024 launch, Google unveiled Gemini, a new AI model surpassing the capabilities of not just OpenAI's GPT-4 and ChatGPT but every other model on almost every parameter.

In other news, EU marathons over landmark AI regulations, Meta launched Purple Llama for Responsible AI development and IBM launched watsonx.governance for business to govern their Generative AI deployments.

Responsible AI is getting hot 🔥 

Future of AI, Today

♨️ Fresh off the Press !

👀 Hot Goss - Google Gemini’s capabilities surpass all LLMs and FMs

🧠 AI Wisdom - How AI sees wealth and poverty

💸 Funding News - Mine, RightBot, Kyron, Leonardo.ai raise funding; Musk looking to raise $1 billion for xAI

📚️ Micro Learning - Security threats in LLMs

😎 Cool Tools - GPTs and AI tools for you

_Fresh off the Press !!_♨️ 

• Google launches Gemini, the most advanced AI so far

• EU close to finalizing landmark rules governing artificial intelligence

• Google’s Gemini won’t be available in Europe to start with

• Meta launches Purple Llama for Responsible AI; and an AI powered image generator

• IBM launches watsonx.governance to help businesses govern their generative AI deployments

 HOT GOSS 👀  👀 

Google Gemini’s capabilities surpass all LLMs and FMs 

Google recently unveiled Gemini, a groundbreaking AI model that surpasses all its predecessors in capability. This remarkable innovation comes in three distinct versions:

1. Gemini Ultra: The most advanced and powerful version, designed for complex tasks. Its new, most sophisticated AI model, Gemini Ultra, surpasses OpenAI's GPT-4 in various areas including text, images, coding, and analytical tasks. Starting early next year, Gemini Ultra will be accessible via a new AI chat feature named Bard Advanced. Currently, it is undergoing enhancements and comprehensive safety evaluations, which include rigorous testing by external experts as per the company's recent statement.

2. Gemini Pro: This is smaller model that is versatile and ideal for a broad spectrum of tasks. Gemini Pro is already available to users through Google's Bard chat platform. Starting December 13, developers will gain access to Gemini Pro via an API.

3. Gemini Nano: This is the most compact model perfect for tasks executed on devices. Starting December 13, Android developers will have the opportunity to utilize Gemini Nano in their creations. Gemini Nano is slated for integration in Google's upcoming Pixel 8 Pro smartphone

All three versions are capable of processing and producing outputs in text and image form, as well as handling audio and video. Google plans to incorporate the Gemini models into its various products and services, including online search and advertising.

This development is significant because numerous companies have invested heavily in trying to outdo OpenAI's GPT-4, yet none have succeeded until now. Google's Gemini Ultra has set a new standard by outperforming GPT-4 in many essential benchmarks, achieving superior results in 30 of the 32 major academic evaluations used in the field:

But wait, Gemini didn’t just beat ChatGPT. Google DeepMind revealed that they benchmarked Gemini Ultra against various competing models, such as OpenAI's GPT-4, Anthropic's Claude 2, Inflection's Inflection-2, Meta's Llama 2, and xAI's Grok 1. The results showed that Gemini Ultra surpasses these rivals in several tests, including professional and academic multiple-choice questions as well as Python programming tasks.

The versatility of Gemini is one of its most impressive features. As a multimodal AI, it can handle various inputs and tasks, including text, code, images, audio, and video. Its applications are diverse, ranging from text and code analysis and generation to creating new images and understanding video and audio content to respond to queries.

Here’s a video that shows Gemini’s powerful capabilities:

According to Google,

“Starting on December 13, developers and enterprise customers can access Gemini Pro via the Gemini API in Google AI Studio or Google Cloud Vertex AI.

Google AI Studio is a free, web-based developer tool to prototype and launch apps quickly with an API key. When it's time for a fully-managed AI platform, Vertex AI allows customization of Gemini with full data control and benefits from additional Google Cloud features for enterprise security, safety, privacy and data governance and compliance.

Android developers will also be able to build with Gemini Nano, our most efficient model for on-device tasks, via AICore, a new system capability available in Android 14, starting on Pixel 8 Pro devices. Sign up for an early preview of AICore.”

You can read the full official announcement here and full technical report here.

 AI WISDOM 🧠 

How AI sees Wealth and Poverty

For the most effective poverty relief, it's crucial to identify the locations where it's needed the most. Traditionally, this is done through infrequent and geographically limited household surveys in many countries.

However, recent breakthroughs in artificial intelligence (AI) have revolutionized the way we assess poverty and other human development metrics. Few researchers in Sweden have utilized deep convolutional neural networks (DCNNs) to analyze satellite imagery. This method can detect certain types of poverty with an accuracy comparable to household surveys.

This AI approach is particularly useful in developing countries experiencing rapid land-use changes. It allows for satellite monitoring to quickly identify areas requiring aid, a process much faster than traditional ground surveys.

Additionally, the visually captivating images produced by their deep learning model offer a novel perspective on how AI perceives the world.

The Swedish researchers discovered that a DCNN, initially trained on ImageNet's extensive image repository, could be fine-tuned with satellite images to accurately assess poverty levels. This training enabled the DCNN to outperform humans in detecting poverty from low-resolution satellite images—a task humans typically do with high-resolution images. This achievement is equivalent with AI's superhuman capabilities in areas like Chess and Go.

To understand how the DCNN discerns "high wealth" in satellite imagery, we started with a "blank slate" image filled with random noise. Through iterative adjustments, the DCNN transformed this image into what it recognized as an area of high wealth. This process revealed the model's learned indicators of wealth, such as road density and urban layout, offering insights into the AI's internal learning mechanisms.

Comprehending poverty, especially within its geographical or regional framework, is a challenging task. Traditional approaches have primarily concentrated on the individual elements of poverty. However, AI's use of satellite imagery has greatly advanced our understanding of the geographical distribution of regional poverty.

The true worth of AI in evaluating poverty lies in its ability to provide a detailed spatial analysis. This enhances current poverty research and contributes to the development of more focused and efficient strategies for intervention.

You may read the full article in detail here

 FUNDING NEWS 💸 💸 

Mine, RightBot, Kyron, Leonardo.ai etc raise funding; Musk looking to raise $1 billion for xAI

• Mine, a startup disrupting the data privacy market, announced today that it has raised $30 million in Series B funding, co-led by Battery Ventures and PayPal Ventures

• Rightbot, a startup developing suction-based robots that can unload truck-transported freight in a range of sizes, has raised $6.25 million in a funding round led by Amazon’s Industrial Innovation Fund (IIF) with participation from SOSV and Entrepreneur First.

• Kyron Learning, an AI-based learning startup, announced today its $14.6 million Series A funding round plus an $850,000 grant from the Bill & Melinda Gates Foundation. The new capital will further develop the platform’s generative AI capabilities and build out its K-12 math curriculum

• AssemblyAI, an “AI-as-a-service” startup which focuses specifically on speech-to-text and text analysis services, lands $50M to build and serve AI speech models

• Sarvam.ai has come out of stealth mode and announced it has raised $41 million as the five-month-old Indian startup races to build a suite of full-stack generative AI offerings in the world’s most populous nation

• Vast Data which provides a scale-out, unstructured data storage solution designed to eliminate tiered storage (i.e. setups that move data between high cost - low cost storage hardware), today announced that it secured $118 million in a Series E round

• Leonardo.ai, the AI art production platform for consumers and enterprise users announced a $31 million round. The startup lets users save, edit and build multiple assets in the same style so they can be reused. They can also build and train their own models for image generation.

• EnCharge AI raises $22.6M to commercialize its AI-accelerating chips. EnCharge’s charge-based, in-memory computing technology was initially developed under DARPA and Department of Defense funded programs

• Elon Musk is looking to raise $1 billion in funding for his next venture, xAI. The company has debuted one product, a chatbot called Grok, trained on data from the X social network, which Musk also owns. It is “designed to answer questions with a bit of wit and has a rebellious streak,” according to the company website.

 MICRO LEARNING 📚️ 

Security Threats in LLMs

Prompt injection attack is a new technique for manipulating large language models (LLMs) using carefully crafted prompts to make them ignore instructions or perform unintended actions, potentially revealing sensitive data or executing unauthorized functions. This video takes you through type of LLM attacks

 COOL TOOLS 😎 🛠️

• This Person Does Not Exist: Create non-existent people images easily with one click.

• Murf AI: Generate lifelike, studio-quality voiceovers swiftly for various professional uses.

• Visily AI: Convert screenshots, templates, or text into wireframes.

• SummarAIze: Repurpose your audio and video content into captivating social posts, emails, summaries, quotes.

• Til;dv: Record, Transcribe & Summarize Google Meet, Zoom and MS Teams calls.