- One more Thing in AI
- Posts
- #11: Uncovering Google's Gemini, Mistral AI Emerges and Insurtech meets AI
#11: Uncovering Google's Gemini, Mistral AI Emerges and Insurtech meets AI
#11: Latest edition of One More Thing in AI Newsletter.
Renjit Philip
December 10, 2023
We are stuck with technology when what we really want is just stuff that works.”
Edition #11: One More Thing in AI
Date: 10-Dec-2023
Gentle reader,
Welcome to the latest edition of the “One More Thing in AI newsletter. For Startup founders and Business Leaders like you who want to get smarter on AI in less than 5 minutes.
Another cracker-jacker of a week with Gemini stealing the hearts of AI enthusiasts and then disappointing them in short order. We also cover several insurtech startups using AI to cover climate risks. Plus learn how to analyse data using power prompts. I hope you enjoy reading this edition as much as I did putting it together for you.
As usual, keep those comments and feedback coming.
Best regards,
Renjit Philip
In this Power-Packed Edition
🌟 Gemini: Unveiling the Truth Behind Google's AI Marvel
The Enigma of Gemini's Demonstration:
Google's recent showcase of Gemini, its latest AI marvel, has stirred quite the buzz in the tech community. A video demonstration featured Gemini(see below) answering queries about various images with what appeared to be remarkable ease. However, a Bloomberg article has peeled back a layer of this seemingly flawless display, revealing that the demo wasn't conducted in real-time or via voice, as it appeared. This revelation has sparked a wave of skepticism, with social media and some media outlets labeling the video as 'fake'.
Performance Under Scrutiny:
At the heart of the debate lies Gemini's performance on the Massive Multitask Language Understanding (MMLU) benchmark, a critical measure of an AI model's knowledge and problem-solving prowess. Google proudly claimed that Gemini was the first AI model to outperform human experts in this test. Yet, this assertion has been critiqued by experts like Brett Winton of ArkInvest and others who have pointed out that Gemini's superior results were partly due to specific prompting techniques used during the test.
So, what was the big move made by Google?
Introducing Gemini: Google's Trailblazing AI Model 🚀
Google has introduced Gemini, their most advanced and comprehensive AI model to date, marking a new era in artificial intelligence. Gemini stands out as Google's largest and most capable AI model, optimized for various sizes: Ultra, Pro, and Nano.
Gemini's Flexibility:
Gemini's flexibility sets it apart, efficiently operating across diverse platforms, from data centers to mobile devices. Its capabilities are set to revolutionize how developers and enterprise customers utilize AI to build and scale solutions.
A Benchmark in AI Performance:
Gemini Ultra has exceeded current state-of-the-art results on numerous academic benchmarks. Notably, it outperformed human experts on the Massive Multitask Language Understanding (MMLU) benchmark, demonstrating superior world knowledge and problem-solving skills.
Gemini beats GPT-4 on many parameters
Natively Multi-modal Approach:
Unlike previous models that stitched together separate components for different modalities, Gemini is said to be natively multi-modal from the outset.
Empowering Developers with Advanced Coding:
Gemini is poised to revolutionize coding, by explaining, and generating high-quality code in various programming languages. It promises to be a valuable tool, accelerating the development of apps and services. [Solopreneur Alert!]
Availability for Developers and Enterprises:
Starting December 13, developers and enterprise customers will gain access to Gemini Pro via the Gemini API in Google AI Studio or Google Cloud Vertex AI.
Checkout the video that went viral:
Lessons Learnt: Don't trust everything that comes from desperate Tech companies trying to catch-up with their competitors in the AI space. Also, Google is a company to watch;they will figure it out eventually!
🌐 Mistral AI: The Rising Star in AI's Galaxy 🌐
💫 A Stellar Ascent to $2 Billion:
Mistral AI has emerged as a formidable force in the AI arena with its rapid rise to a $2 billion valuation. With significant funding from industry giants like Nvidia Corp and Salesforce Inc, Mistral AI has a robust foundation to fuel its growth.
🔍 Distinguishing Features:
What sets Mistral apart is its focus on open-source software for chatbots and other generative AI tools. This approach positions Mistral as a cost-effective alternative by challenging established players with its unique offerings and substantial computing resources.
🌍 European AI Innovation Front and Center:
Mistral's roots trace back to the expertise of DeepMind and Meta Platforms Inc. scientists, harnessing their experience in large language models. Its European origin and successful initial funding of $113 million mark a significant milestone for a burgeoning AI ecosystem outside the US.
🤝 A Diverse Investor Portfolio:
The funding round saw participation from notable firms like General Catalyst, Lightspeed Venture Partners, and Bpifrance.
Takeaways: Mistral AI's meteoric rise and substantial valuation underscore the promising future of AI technology outside of the US. In the competitive AI landscape, Mistral symbolizes the global reach of AI innovation.
🌟 Revolutionizing Game Development with AI: Arcweave's Leap Forward 🌟
Empowering Creators with AI:
Arcweave, a narrative design platform, has raised over $850k in its seed funding round to enhance its game creation tools. They aim to introduce a multi-modal, AI-powered assistant to generate game stories from prompts. This advancement promises a new frontier in game development, where AI aids and elevates the creative process.
Bridging the Narrative Gap:
Arcweave targets a crucial need in the gaming industry: sophisticated scriptwriting tools for complex, player-driven narratives. Their low-code platform, with its user-friendly interface, is designed to seamlessly blend into existing game development workflows, enhancing the efficiency and scope of narrative design.
Their exploration into generative AI is set to provide game creators with tools for consistent and engaging game narratives. Thinking of creating a game? This could be your moment!
A New Era of Interactive Storytelling:
With the support of investors like Galaxy Interactive, Arcweave is poised to redefine interactive storytelling. Integrating generative AI into their platform promises an extra layer of innovation, aiding developers in maintaining narrative consistency and enriching game worlds with new content.
Takeaway:
Arcweave's merges AI's capabilities with human creativity to forge immersive and complex virtual worlds. AI will enable several indie game shops to be empowered to create viral video games that challenge the Activisions / Segas / Nintendos of the world!
🚀 AI Meets Biotech: A New Era of Therapeutic Discovery
The Dawn of Collaborative Innovation:
AbbVie Inc. and BigHat Biosciences have announced a groundbreaking collaboration to harness artificial intelligence and machine learning (AI/ML) to discover next-generation therapeutic antibodies. This venture combines AbbVie's oncology and neuroscience expertise with BigHat's Milliner™ platform, a fusion of machine learning technologies, and a high-speed wet lab. This alliance is not just a business transaction; it's a fusion of two distinct worlds - biotech and AI - aiming to revolutionize drug development and patient care.
A Financial and Technological Powerhouse:
In this deal, BigHat will receive an upfront payment of $30 million, with the potential to earn up to $325 million in research and development milestones. It's a beacon of hope, signaling the potential of AI to address the most pressing medical needs of our time.
Redefining Biologic Therapies:
BigHat Biosciences aims to create safer, more effective biologic therapies. By integrating machine learning with synthetic biology, they are pushing the boundaries of antibody design. Imagine a future where treatments are effective and tailored to individual patient needs, thanks to the precise and rapid optimization capabilities of AI platforms like Milliner.
Rising Tides, Rising Innovations: How AI and Insurtechs Are Reshaping Risk Management
Imagine a world where insurance payouts are as predictable as the rising tide. In Freeport, Texas, this isn't just imagination—it's reality, thanks to a groundbreaking approach to insurance.
Picture a warehouse owned by Postlane Partners in Freeport. Here, a mere 8 inches of floodwater triggers an automatic $3 million insurance payout from FloodFlash, a pioneering startup. Should the water reach 16 inches, the payout escalates to $5 million. This isn't your typical insurance policy; it's a parametric policy. The insurtech in question, FloodFlash, uses on-the-ground sensors that enable an automatic payout when a catastrophe occurs.
Owners of homes and businesses have watched with alarm as major insurance companies have stopped offering coverage in California, Florida, and other parts of the country prone to natural disasters.Traditional insurance models, strained under the unpredictable nature of climate-related events, often find such risks too hot to handle. Enter insurtechs backed by cutting-edge tech, offering a solution where others see insurmountable challenges.
Here is another one: Kettle, an insurance startup that uses a type of artificial intelligence to understand how climate change affects risk and to sell property insurance accordingly.
Lessons Learnt: The secret sauce of these startups is technology. They are using better data science and incorporating artificial intelligence. In an era where climate change is rewriting the rulebook on risk, AI + parametric + insurtechs stands as a beacon of adaptability and innovation. It's not just about managing risks—it's about redefining them.
Learn AI terms: Fine-Tuning
Just as a well-tailored suit fits perfectly, fine-tuning allows the LLM to be tailored for specific tasks or domains, such as customer service chatbots, legal or medical text, and brand-specific interactions.
Fine-tuning a Large Language Model can significantly improve its performance in several ways:
1. Improved Accuracy: Fine-tuning LLMs with task-specific or domain-specific data can lead to higher accuracy levels, ensuring that the model's outputs closely align with expectations.
2. Domain-Specific Knowledge: Fine-tuning enables LLMs to understand and generate highly relevant content for a particular business or industry.
3. Reduced Data Requirement: Fine-tuning can reduce the data required to train an LLM, making it more efficient for specific tasks. This is beneficial when the availability of task-specific data is limited.
4. Customized Interactions: For applications like chatbots, fine-tuning helps tailor responses to match a brand's voice, tone, and guidelines, ensuring a consistent and on-brand user experience. However, it may be less reliable for factual recall or training the model about entirely new knowledge.
Prompting- Power your Data Analysis
One of the valuable features that AI-powered Chatbots like ChatGPT, Claude, and Perplexity can do is analyze data files. For example, here are the formats that ChatGPT can process:
Here are a few data analysis prompts that I picked up from Superannotate:
"Highlight the most important arguments and statistics from the attached powerpoint deck."
"Find interesting and unusual correlations in the attached dataset in the first sheet of the excel sheet."
"Identify trending themes, overlaps, and contradictions in the attached excel files."
"Create a line chart from the attached data in weekly increments for 2020 upto 2023.”
"Prepare a graph that shows organic traffic on a per-blog basis for the previous month."
These last two examples point to a valuable use case for content marketers: Making graphs and charts from data to create reports for your customers or managers. I think you can open up a fiverr profile right now to process content marketing data and make a side-hustle out of it.
Digitize handwritten notes. ChatGPT's file handling ability also includes image formats.This is helpful if you take notes on paper or whiteboards and for any other physical deliverables that you encounter during customer research or other primary research.
Here is a hack that I recently used: I got access to a list of VCs and angel investors (on Airtable), recently published on LinkedIn. I tried copying the email-ids, but the table was set to view-only, which prevented me from doing so. Enter ChatGPT. I uploaded the screenshots of the list and asked ChatGPT to create an Excel file for me. I found that I could extract an editable Excel file from the screenshots of that Airtable list. Try this out when you want to extract data from images.
Please look at the bottom of the newsletter for sponsorship opportunities to reach over 1000+ business and startup professionals. If you subscribe to our email edition, you get to read this edition before non-subscribers, and it stays in your mailbox as a future reference. Subscription has its value!
A word from our sponsor…
Yadget
“The name Yadget stands for ‘Yet another Data Generator'’ Cheeky, right?
You can produce realistic, non-identifiable datasets to help you test and validate your products. Ideal for software developers for data testing and validation processes. Among the data that can be generated are financial Data like Bank Account Numbers, Credit Card numbers, and Crypto information such as BTC and ETH addresses.
Why Yadget?
1. Generate synthetic data without the high cost of real-world data.
2. Faster than traditional methods of collecting and processing real-world data.
3. Flexible: Yadget provides tailor-made data to meet your specific needs.
Give Yadget a try, and let us know what you think. Check out Yadget here: Link.”
References / Deep Dive:
Feedback
How did you like this edition?Your feedback helps us to improve. |
That is it for this edition of the newsletter. If you subscribe to our email edition, you get to read this edition before non-subscribers, and it stays in your mailbox as a future reference. Subscription has its value!
Until next time, keep learning, applying, and experimenting with AI!
Support & Sponsor
Forwarded this email by someone, and you want to subscribe?
You can help us spread the word by sharing this newsletter with a friend or colleague.
To sponsor this newsletter, subscribed to by hundreds of startup founders and business leaders, write to sponsor@onemorethinginai.com
This newsletter is distributed on LinkedIn and by email. Circulation is counted on both platforms.
To get the newsletter as soon as it is published, hit the subscribe button. If you want to interact with me on social media, follow me on LinkedIn and Twitter.
Disclosure: Some of the links in this newsletter are affiliate links, which means we may earn a small referral fee at no extra cost to you if you purchase from them. I personally test out the AI tools during my review process. Your support helps keep this newsletter running—thank you!
Join the conversation