Dr. Tahseen Anwer Arshi, Associate Provost for Research and Sustainability and Director of the Center for Innovation and Entrepreneurship, AURAK, conducted the research, ‘Integrating Large Language Models into Entrepreneurial Ventures’.

The study provides actionable insights into the benefits of integrating LLMs into entrepreneurial activities, demonstrating the potential to empower and grow startups in areas like ideation, creativity, new product development, innovation, customer engagement, and service personalization.

Prof. Stephen Wilhite, Senior Vice President of Academic Affairs and Student Success and Provost at AURAK, says: “LLMs are critical to accelerating growth and speeding up processes in enterprises large and small. They push entrepreneurs toward peak performance due to the increased productivity resulting from automating of tasks. Our research delves into how LLMs empower entrepreneurs to harness automation smartly, leaving them to devote themselves to more complex tasks that can only be performed by humans.”

Dr. Arshi explains: “The use of LLMs in entrepreneurship is still developing, with few case studies available. However, promising examples include AI-based content generation start-ups using GPT-4 for marketing and software companies using LLMs for real-time coding. Our study demonstrates that LLMs enhance product innovation by automating tasks and facilitating rapid prototyping. They allow innovation teams to quickly create user interface descriptions, technical documentation, and wireframes, leading to faster stakeholder feedback and supporting agility in start-ups.”

Recent advancements have improved LLM technology further, moving beyond traditional transformer architectures to include innovations such as mixture-of-experts architectures and advanced quantization techniques (4-bit, 8-bit) that reduce memory usage while maintaining performance. Additionally, start-ups are employing hybrid deployment strategies combining edge computing with cloud-based model sharding for better resource efficiency.

The study recommends enhanced AI literacy among entrepreneurs through wider exposure to chatbots, recommendation algorithms and AI-powered analytics. Entrepreneurs should also be provided with hands-on opportunities with no-code AI tools like ChatGPT, Google AutoML, and Zapier AI. Such learning can be facilitated through AI hackathons, mentorship programs, and start-up accelerators. AI peer networks can be created through entrepreneur communities, LinkedIn groups, and local meetups.

The research paper was first presented at the International Conference on Business Management, Entrepreneurship & Circular Economy in November 2024. It is currently in the process of being released in Advances in Science, Technology and Engineering, published by Springer Nature.

The research carries great significance for businesses, since LLMs are widely hailed as key business drivers worldwide. Researchers at the International Institute for Management Development have termed LLMs a “technological tsunami” about to “reshape the global economic landscape”; while researchers at McKinsey estimate generative AI could add the equivalent of $2.6 to $4.4 trillion of value to the world economy annually.

The post Use of LLMs can boost AI-driven innovation in startups appeared first on My Startup World - Everything About the World of Startups!.

The groundbreaking release sets new performance standards for small LLMs and democratizes access to advanced artificial intelligence by enabling the model to operate efficiently on light infrastructures, including laptops. Falcon 3 introduces superior reasoning and enhanced fine-tuning capabilities, making it a more powerful and usable AI model.

Falcon 3 is designed to democratize access to high-performance AI, offering models that are both powerful and efficient. Trained on 14 trillion tokens—more than double its predecessor’s 5.5 trillion—Falcon 3 demonstrates superior performance across various benchmarks. Notably, it ranks among the top models globally that can operate on a single GPU. Upon its release, Falcon 3 achieved the number one position on Hugging Face’s global third-party LLM leaderboard, surpassing other open-source models of similar size, including Meta’s Llama variants. In particular, the Falcon 3-10B model leads its category, outperforming all models under 13 billion parameters.

His Excellency Faisal Al Bannai, Secretary General of ATRC and Adviser to the UAE President for Strategic Research and Advanced Technology Affairs, said: “The transformative power of AI is undeniable. Today, we advance our contributions to the AI community, particularly the open-source sector, with the release of the Falcon 3 family of text models. This launch builds upon the foundation we established with Falcon 2, marking a significant step toward a new generation of AI models. Our ongoing commitment to ensuring these powerful tools remain accessible to everyone, everywhere, reflects our dedication to global equity and inclusive innovation.”

Falcon 3 Family
The Falcon 3 series includes four model sizes: Falcon3-1B, -3B, -7B and -10B. Compared to its predecessors, significant attention has been placed on seamless integration. The models are fully compatible with widely used APIs (Application Programming Interfaces) and libraries, considerably reducing integration efforts and ensuring ease of use. This ultimately enables users to select a solution that best fits their needs. With Falcon 3 delivering exceptional performance across reasoning, language understanding, instruction following, code generation, and mathematical tasks, it is poised to set new standards in AI capabilities.

The Falcon 3 small models each have a Base and an Instruct variation that each rank among the most powerful in the world for its size. The Base model allows general-purpose generative tasks, whereas the Instruct is a fine-tuned variant for conversational applications. Falcon 3 is available in English, French, Spanish and Portuguese. Falcon 3 models also have quantized versions that enable optimized integration into specialized architectures, being resource efficient and lightweight for rapid deployment and inference.

Dr. Najwa Aaraj, TII’s CEO, said: “Our dedication to pioneering research and attracting top-tier talent has culminated in the development of Falcon 3. The result is a model that exemplifies our pursuit of scientific excellence, offering enhanced efficiency and setting new benchmarks in AI technology.”

Dr. Hakim Hacid, Chief Researcher of the TII’s AI and Digital Science Research Center (AIDRC), said: “AI is fast evolving, and we are glad to be an active part of this journey. Falcon 3 pushes the boundaries of small LLMs further, contributing to the open-source community by providing access to a better-performing AI. We are confident that this latest release will open an unlimited range of opportunities and will have tremendous benefits, empowering businesses and individuals to use AI in ways that were previously out of reach.”

Falcon 3 is available immediately for download on HuggingFace and at FalconLLM.TII.ae, along with details on the benchmarks.

TII is also introducing Falcon Playground, a testing environment for end-users, programmers, coders, and researchers to explore Falcon 3 before its official release, providing an opportunity to experiment and offer feedback.

Falcon 3 is licensed under TII Falcon License, the permissive Apache 2.0-based software license that includes an acceptable use policy which promotes the responsible use of AI.

Early January 2025, the Falcon 3 Family of models will introduce new members with an emphasis on multimodal functionalities—including text, images, video, and voice mode.

The post Abu Dhabi’s TII unveils Falcon 3 LLM appeared first on My Startup World - Everything About the World of Startups!.

Building on the recent launch of the JAIS 70B model as part of 20 other JAIS models and trained on the world’s largest Arabic dataset, JAIS Chat now offers Arabic speakers the choice of three AI model options when interacting with its new and improved chat interface. JAIS 30B, a small and swift model ideal for fast response times; JAIS 70B, a powerful and precise model designed for enhanced accuracy; and the new K2-Diamond-Chat, a 65-billion parameter model that sets a new standard for mathematics and coding.

The K2-Diamond-Chat model enables users to tackle complex mathematical problems and manipulation of symbolic expressions such as algebraic simplification and differentiation. Inception has launched the new K2 Diamond Chat in partnership with the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI).

“We are enhancing our offerings to deliver breadth, capability, and performance to the Arabic speaking community. We want our most advanced Arabic bilingual model, JAIS 70B, to be accessible to as many people as possible, wherever and however they use generative AI capabilities; on their mobile, on their laptop, through speech or text, at work and in school. With this collaborated model launch with MBZUAI, we are focused on training stronger foundational models in specialized domains. By combining our strengths in AI research and applied AI, we will drive speed and focus on building AI native products that can simplify data-rich segments within business and sector domains. We will continue to evolve JAIS as the leading LLM for Arabic-speaking nations to ensure that our models remain at the forefront of innovation offering quality and performance.” said Andrew Jackson, Acting CEO of Inception.

“The advancements we are announcing today mark a significant milestone in our mission to develop AI solutions that benefit society. Through our partnership with Inception, we will align closely around the most important applications of generative AI to create innovative and impactful language and domain models,” said Eric Xing, MBZUAI President and University Professor.

This update also features Inception’s latest audio models, offering automatic transcription of Arabic speech (in its various dialects) into text, and converting text into natural sounding speech.

In line with Inception’s commitment to equitable AI, the JAIS Chat app is available for free on iOS and Android platforms, and on https://jais.inceptionai.ai, ensuring that the company’s advanced AI technology is accessible to all.

The post Inception updates JAIS Chat app with new AI Models and enhanced audio appeared first on My Startup World - Everything About the World of Startups!.

JAIS 70B delivers Arabic-English bilingual capabilities at an unprecedented size and scale for the open-source community. As a 70 billion parameter model, it has increased ability to handle complicated and nuanced tasks, as well as better capability to process complex datasets. JAIS 70B was developed using continuous training, a process of fine-tuning a pre-trained model, on 370 billion tokens of which 330 billion were Arabic tokens, the largest Arabic dataset ever used to train an open-source foundational model.

This suite of JAIS models accommodates a wide range of use cases, and aims to accelerate innovation, development, and research opportunities for multiple downstream applications for the Arabic speaking and bilingual community.

Dr. Andrew Jackson, CEO, Inception said: “AI is now a proven value-adding force, and large language models have been at the forefront of the AI adoption spike. JAIS was created to preserve Arabic heritage, culture, and language, and to democratize access to AI. Releasing JAIS 70B and this new family of models reinforces our commitment to delivering the highest quality AI foundation model for Arabic speaking nations. The training and adaptation techniques we are delivering successfully for Arabic models are extensible to other under-served languages and we are excited to be bringing this expertise to other countries.”

Inception released JAIS-13B and JAIS-13B-chat in August 2023 and subsequently launched the state-of-the-art Arabic-centric models, JAIS-30B and JAIS-30B-chat. JAIS 70B and JAIS 70B-chat have proven to be even more performant in benchmarking data in both English and Arabic compared to previous models.

Neha Sengupta, Principal Applied Scientist, Inception said: “For models up to 30 billion parameters, we successfully trained JAIS from scratch consistently outperforming adapted models in the community. However, for models with 70 billion parameters and above, the computational complexity and environmental impact of training from scratch were significant. We made a choice to build JAIS 70B on the Llama2 model, allowing us to leverage the extensive knowledge base of an existing English model and develop a more efficient and sustainable solution.”

JAIS 70B retains, and in specific cases, exceeds, the high-quality English-language processing capabilities of Llama2, while vastly excelling on Arabic outputs versus the base model. The JAIS development team trained an expanded tokenizer based on the Llama2 tokenizer to enhance Arabic text processing efficiency, doubling the model’s base vocabulary. According to Sengupta, the model “splits Arabic words less aggressively and makes training and inferencing cheaper” than the standard Llama2 model.

Users can download the JAIS models and access the technical paper and benchmarking results by visiting the dedicated page on Hugging Face: https://huggingface.co/inceptionai/jais-adapted-70b.

The post G42 launch of JAIS 70B champions Arabic NLP appeared first on My Startup World - Everything About the World of Startups!.

LLM applications run on algorithmic models that are trained on huge sets of data. Even when that training data is curated by humans, which it often is not, the application can easily reflect “data bias” caused by stereotypes, prejudices, exclusionary language, and a range of other possible biases from the training data. Such biases can lead the model to behave in potentially unintended and harmful ways, adding considerable risk and unpredictability to LLM adoption.

Some examples of potential flaws include Representation Bias (disproportionate representation or omission of certain groups in the training data), Pre-Existing Bias (biases stemming from historical or societal prejudices present in the training data), and Algorithmic Processing Bias (biases introduced through the processing and interpretation of data by AI algorithms).

The public sector is urgently affected by this growing risk. As of March 2024, the US Government mandated its agencies to conform with AI safety guidelines – including the detection of data bias. That mandate extends to Federal contractors later in 2024.

This problem requires a new approach to security because traditional security scanners and penetration tests are unable to detect such bias. Bugcrowd AI Bias Assessments are private, reward-for-results engagements on the Bugcrowd Platform that activate trusted, third-party security researchers (aka a “crowd”) to identify and prioritize data bias flaws in LLM applications. Participants are paid based on the successful demonstration of impact, with more impactful findings earning higher payments.

The Bugcrowd Platform’s industry-first, AI-driven approach to researcher sourcing and activation, known as CrowdMatch, allows it to build and optimize crowds with virtually any skill set, to meet virtually any risk reduction goal, including security testing and beyond.

“Bugcrowd’s work with customers like the US DoD’s Chief Digital and Artificial Intelligence Office (CDAO), along with our partner ConductorAI, has become a crucial proving ground for AI detection by unleashing the crowd for identifying data bias flaws,” said Dave Gerry, CEO of Bugcrowd. “We’re eager to share the lessons we’ve learned with other customers facing similar challenges.”

“ConductorAI’s partnership with Bugcrowd for the AI Bias Assessment program has been highly successful. By leveraging ConductorAI’s AI audit expertise and Bugcrowd’s crowdsourced security platform, we led the first public adversarial testing of LLM systems for bias on behalf of the DoD. This collaboration has set a solid foundation for future bias bounties, showcasing our steadfast commitment to ethical AI,” said Zach Long, Founder, ConductorAI.

For over a decade, Bugcrowd’s unique “skills-as-a-service” approach to security has consistently uncovered more high-impact vulnerabilities than traditional methods.

Our customer base, which numbers nearly 1,000, has benefited from this approach, which also provides a clearer line of sight to ROI. With unmatched flexibility and access to a decade of vulnerability intelligence data, the Bugcrowd Platform has evolved over time to reflect the changing nature of the attack surface – including the adoption of mobile infra, hybrid work, APIs, crypto, cloud workloads, and now AI. In 2023 alone, customers found almost 23,000 high-impact vulnerabilities using the Bugcrowd Platform, helping to prevent potential breach-related costs of up to $100 billion.

“As the leading crowdsourced security platform provider, Bugcrowd is uniquely positioned to meet the new and evolving challenges of AI Bias Assessment, just as we’ve met the emergent security challenges of previous technology waves such as mobile, automotive, cloud computing, crypto, and APIs,” said Casey Ellis, Founder and Chief Strategy Officer of Bugcrowd.

To learn more about the Bugcrowd AI Bias Assessment offering, visit here.

The post Bugcrowd launches AI Bias Assessment offering for LLM apps appeared first on My Startup World - Everything About the World of Startups!.

As a pioneer of AI use in payments since 1993, Visa considers this initiative an extension of Visa’s leadership in using AI to drive innovation in payments, create value for partners and clients, and enable and empower global commerce. Generative AI is an emerging subset of AI that is built on Large Language Models (LLMs) to develop artificial general intelligence capable of generating text, images or other content from large sets of existing data when given prompts. “While much of generative AI so far has been focused on tasks and content creation, this technology will soon not only reshape how we live and work, but it will also meaningfully change commerce in ways we need to understand,” said Jack Forestell, Chief Product and Strategy Officer, Visa Inc.

This initiative will be led by Visa Ventures, the global corporate investment arm of Visa. Visa Ventures has been investing in and partnering with companies driving innovation in payments and commerce since 2007. “With generative AI’s potential to be one of the most transformative technologies of our time, we are excited to expand our focus to invest in some of the most innovative and disruptive venture-backed startups building across generative AI, commerce and payments,” said David Rolf, Head of Visa Ventures, Visa Inc.

To learn more about the $100 million generative AI ventures initiative and Visa Ventures, please visit Visa.com/ventures.

The post Visa launches $100 million Generative AI ventures initiative appeared first on My Startup World - Everything About the World of Startups!.

The StarCoder model is designed to level the playing field so developers from organizations of all sizes can harness the power of generative AI and maximize the business impact of automation with the proper governance, safety, and compliance protocols. This new LLM marks the next major milestone in the BigCode Project, an ambitious initiative to develop state-of-the-art AI systems for code in an open and responsible manner with the support of the open-scientific AI research community.

“ServiceNow’s collaboration with Hugging Face expands our longstanding commitment to AI excellence,” said Harm de Vries, lead of the Large Language Model Lab at ServiceNow Research and co-lead of BigCode. “New, responsible AI practices to train and share large language models are vital to ensuring the right protocols, safeguards, and permissive licenses are in place for our customers, and StarCoder is making this possible.”

“The joint efforts led by Hugging Face and ServiceNow enable the release of powerful base models that empower the community to build a wide range of applications more efficiently than a single company could come up with,” said Leandro von Werra, machine learning engineer at Hugging Face and co-lead of BigCode. “This endeavor is a testament to the potential of open source as we work toward democratizing AI.”

Trained with a trillion tokens of permissively licensed source code covering over 80 programming languages from BigCode’s The Stack v1.2 dataset, StarCoder can be deployed to bring pair-programing like generative AI to applications with capabilities like text-to-code and text-to-workflow. With this, StarCoder gives professional software engineers the power to tackle the most complex programming challenges and empowers citizen developers to build new software regardless of technical ability—accelerating AI innovation at scale. The model will be released with open-access on the Code Open RAIL-M license to permit royalty-free distribution. Unlike traditional open-source software released without use case restrictions, BigCode releases the model with a responsible AI model license that includes use case restrictions that apply to modifications of the model, and applications using the model – for example, to restrict the models from being used to generate or distribute malicious code to harm electronic systems. Supporting code has been open sourced on the BigCode project’s GitHub.

ServiceNow Research and Hugging Face, which works on some of the world’s largest AI and LLM initiatives, launched the joint BigCode Project in September 2022. The project continues to operate as an open scientific collaboration, as the two companies harness the collective brainpower and resources from the open-source community through BigCode working groups, task forces, and meetups.

Through partnerships with leading research organizations like Hugging Face, ServiceNow is fueling the development of state-of-the-art generative AI with socially responsible contributions back to the AI research and development community. This includes the company’s partnership with Mila, the Quebec AI Institute, to bridge the demands of advanced AI research. The Mila collaboration will nurture ties with academia by providing the broader research community with enterprise AI experience. ServiceNow also continues to invest internally in AI research and development to build AI-powered experiences that are reliable and safe, and that users can trust.

Beyond key partnerships and collaborations, ServiceNow is continuously embedding enterprise AI capabilities across the Now Platform to fulfill its mission of being the world’s most intelligent platform for end-to-end digital transformation. In the last three years alone, ServiceNow has expanded its portfolio with several notable AI acquisitions, including Attivio, Element AI, and Hitch Works, bringing practical, purpose built AI capabilities to the Now Platform. Most recently, ServiceNow introduced new AI-based tools within the Now Platform Utah release to help organizations maximize efficiency, accelerate ROI, and create more simplified, connected experiences.

The post ServiceNow and Hugging Face reaffirm their commitment to AI excellence appeared first on My Startup World - Everything About the World of Startups!.