PreIPO INTELLI™ Weekly Insights | Series 54
Friday, October 20th 2023 | Volume 1 Series 54 | AI Innovations, Convoy Closure Update, and more recent funding rounds.
Happy Friday, everyone! Yet another week has zoomed past, and with it comes another set of news and updates, courtesy of the INTELLI™ Newsletter. In this week's newsletter, we'll discuss independentz AI developers' endeavors, including models like LLaVA-1.5 and Adept's offerings, which are contributing to the evolution of multimodal AI capabilities. Additionally, we explore the closure of Convoy, the Seattle-based trucking startup once valued at $3.8 billion, revealing the challenges it faced in finding a buyer and the consequences for its employees and investors.
Introducing two open source contenders challenging OpenAI's 'multimodal' GPT-4V
OpenAI's GPT-4V is being celebrated as a groundbreaking AI model known for its "multimodal" capabilities, enabling it to understand both text and images. However, while it offers valuable features, it also comes with potential downsides that are challenging to address. Here's a comparative analysis of this technology.
Multimodal models possess capabilities that surpass those of models specialized in either text or image analysis. For instance, GPT-4V can provide instructions that are more easily demonstrated visually, such as how to fix a bicycle. Moreover, these models can not only identify the objects in an image but also understand and interpret the contents to some extent. This goes beyond the obvious, like suggesting recipes based on the items found in a fridge pictured in an image.
Nonetheless, multimodal models introduce new risks. OpenAI initially delayed the release of GPT-4V due to concerns about potential misuse for identifying individuals in images without their consent. Even now, GPT-4V, available only to OpenAI's ChatGPT Plus subscribers, exhibits worrisome issues, including an inability to recognize hate symbols and a propensity to exhibit biases against certain genders, demographics, and body types, as acknowledged by OpenAI itself.
Despite these risks, companies and independent developers are actively pursuing open source multimodal models that, while not as capable as GPT-4V, can still perform many similar tasks. One recent release is LLaVA-1.5, developed by researchers from the University of Wisconsin-Madison, Microsoft Research, and Columbia University. This model can answer questions about images based on prompts, such as "What's unusual about this picture?" and "What should I be cautious about when visiting here?"
Adept, a startup specializing in AI models for autonomous navigation of software and the web, also introduced a GPT-4V-like multimodal text-and-image model. Adept's model stands out due to its ability to understand "knowledge worker" data, including charts, graphs, and screens, enabling it to manipulate and reason with this data.
LLaVA-1.5, an improved version of its predecessor LLaVA, employs a "visual encoder" and Vicuna, an open source chatbot based on Meta's Llama model, to analyze images and text and their relationships. The LLaVA-1.5 team enhanced the model's training data by increasing image resolution and incorporating data from ShareGPT, a platform for sharing conversations with ChatGPT.
This larger LLaVA-1.5 model with 13 billion parameters can be trained in a day using eight Nvidia A100 GPUs, offering a more cost-effective alternative to GPT-4's expensive training process. However, performance remains a critical factor for its success.
Tests conducted on LLaVA-1.5 revealed strengths and weaknesses. It demonstrated an impressive ability for "zero-shot" object detection, but struggled when analyzing images with multiple objects. Additionally, it had difficulties with text recognition compared to GPT-4V, which recognized text more reliably. While this might seem like a drawback, it could also be seen as a security advantage, as GPT-4V could potentially be manipulated into bypassing safety measures with malicious text-containing images.
Adept's Fuyu-8B, on the other hand, is an open source multimodal model designed to demonstrate the company's capabilities and invite feedback. It's not intended for commercial use but aims to showcase Adept's work with knowledge worker data. This model stands out for its ability to understand unstructured data, locate specific elements on screens, extract relevant information from software UIs, and answer questions about charts and diagrams.
While Fuyu-8B has potential, its open source nature raises concerns about potential misuse, especially if it exhibits the same issues present in GPT-4V. Developers are increasingly embracing open sourcing multimodal models, but the consequences of doing so without appropriate safeguards remain uncertain.
Convoy, Valued at 3.8 Billion Last Year is Shutting Down
Convoy Inc., a trucking startup based in Seattle and backed by high-profile investors like Jeff Bezos and Bill Gates, is shuttering its operations following a failed four-month quest to find a buyer, according to a memo sent to employees by the company's founder and CEO, Dan Lewis. "Today marks your final day with the company," he stated in the memo.
Lewis explained, "Over the course of numerous months, we undertook a comprehensive process where we examined every possible strategic avenue for our business, and unfortunately, we've arrived at the decision to cease our primary business operations. We're currently in the process of reviewing various strategic options for what may lie ahead."
As per last year's valuation, Convoy was worth $3.8 billion, and it had downsized its workforce from 1,500 to around 500 people, with the remaining staff laid off on Thursday. Lewis cited a "massive freight recession" and a "contraction in the capital markets" as hurdles that deterred potential buyers, resulting in the company's closure.
Convoy had secured over $1 billion in funding from various investors, as reported by PitchBook. With the company's closure, many investors might see their stakes rendered worthless, while Allen & Co., a New York-based investment bank, stands to gain significantly. Allen & Co. had invested in Convoy back in 2015 when the company was valued at less than $60 million. They sold all their shares last year when Convoy was valued at $3.8 billion in a funding round that included other investors such as Baillie Gifford and T. Rowe Price.
The downturn in the logistics industry has also affected other startups like Flexport Inc. and Flexe Inc., both of which had to lay off workers due to a decline in demand post-pandemic and a challenging venture capital market.
Recent Funding Recap 🤝
Petfolk - $40 Million Series B
Petfolk, a contemporary veterinary care company committed to delivering top-tier services to pets, pet owners, and veterinarians within a connected care community, is thrilled to announce a $40 million Series B funding round. The investment was spearheaded by Movendo Capital, a growth equity investor supported by a family, with a focus on the food and beverage, retail, and branded consumer sectors across Europe and North America. Existing investors, including White Star Capital, Idea Farm Ventures, and others, also participated in this funding round.
Petfolk is at the forefront of a transformation in the veterinary field, addressing its fragmentation and staffing challenges by offering services both in-person at cutting-edge pet care centers and virtually through the Petfolk mobile app. Founded by CEO Dr. Audrey Wystrach, DVM, Petfolk's unique approach is a culmination of her extensive 29+ years of experience as a practicing veterinarian. Since 2021, Petfolk has expanded from a single location in Charlotte, North Carolina, to an anticipated 11 locations by the end of 2023, with an additional 20 locations slated for 2024. Petfolk currently ranks as the fastest-growing and highest-rated veterinary service provider in the Southeast, with existing locations in North Carolina, South Carolina, Georgia, and Florida.
iink - $12 Million Series A
When your home sustains damage from weather, fire, or other disasters, your primary desire is to have it restored to its original state. However, the response time for insurance claims can be protracted, leaving you and your contractor in a state of waiting for weeks or even longer.
This is where iink steps in – a digital payments network dedicated to simplifying the disbursement of funds related to multi-party property insurance claims. Based in Tampa, the company is collaborating with mortgage servicing banks and insurance providers to establish a completely digital and automated workflow that streamlines the property restoration process.
This initiative has received a significant boost with the infusion of $12 million in Series A funding, led by Headline. Joining them in the funding round are Motley Fool Ventures, Chartline Capital Partners, Silver Circle Ventures, and a consortium of existing investors.
Fingerprint - $33 Million Series C
Fingerprint, a device intelligence API aimed at enhancing security solutions by identifying hardware accessing websites, has secured a $33 million Series C investment. Nexus Venture Partners led the funding round, with participation from Uncorrelated Ventures. Fingerprint's technology, based on the open source library Fingerprint.js, uniquely identifies devices through attributes like installed fonts and screen resolution, offering fraud prevention services to high-scale businesses and mitigating issues like unauthorized access and credit card fraud. The company's CEO, Dan Pinto, underscores the importance of device identification as a means to prevent illicit activities facilitated by internet anonymity.
As always, stay tuned for more updates and insights from the team at INTELLI™.