Trends 🔥

UAE 🇦🇪

An In-depth Exclusive Conversation with Dr. Ebtesam Al Mazrouei: The Evolution of the ‘Falcon 40B’ and AI’s Upcoming Era in the Emirates

الدكتورة ابتسام المزروعي
Mayssae Ajzannay
Mayssae Ajzannay

11 min

In the modern age of artificial intelligence, the UAE is witnessing a surge of startups. This momentum is driven by the vision of influential leaders in the AI sector, notably Emirati Dr. Ebtesam Al Mazrouei. Serving as the Interim Chief Researcher and Executive Director for Falcon 40B's Artificial Intelligence Division, Dr. Ebtesam has been at the helm of trailblazing tech initiatives aimed at bringing novel and innovative advancements to the domain. We engaged in a conversation with her to delve deeper into these technological developments as the Falcon B180 was being unveiled.


What specifications distinguish the Falcon 40B model from other linguistic models?

Large language models are witnessing widespread development day after day, but at the same time, the “Falcon 40B” model is distinguished from other models by a set of unique specifications. The Falcon 40B model was developed by the Digital Science and Artificial Intelligence Research Center team, which is the same team that developed the advanced “Noor” model, which represents the largest linguistic model in the Arabic language.

The superior performance of the Falcon 40B model can be attributed to the unique databases we used to extract high-quality content from web data, in addition to the use of a custom database for training.

We also paid great attention to data quality by building data channels that fit tens of thousands of central processing units, to achieve high processing speeds and extract high-quality content from the web by using comprehensive filtering and de-duplication functions in the results.

The Falcon 40B architecture has been improved in terms of performance and efficiency, which has enabled it to significantly outperform its counterparts from leading international models such as “GPT-3”, “Chinchilla” and “PaLM-62B”. Additionally, the Falcon 40B model matches the performance of cutting-edge language models from leading companies such as DeepMind, Google, and Anthropy.

It is worth noting that the “Falcon 40B” model ranked first in the list of open linguistic models issued by the Hugging Face platform, which is an objective evaluation tool available to the artificial intelligence community to track large linguistic models and chatbots, classify them, and evaluate their effectiveness when they are launched, as Falcon outperformed a group Among the largest and most popular models such as LLaMA from Meta, StableLM from Stability AI, and RedPajama from Together, it has continued to occupy the first place for about two months since its launch.


What challenges did you face while developing the Falcon 40B model?

“Falcon 40B” is a self-decryption model that includes 40 billion variables and was trained on one trillion tokens. It was also trained on 384 central processing units using Amazon Web Services (AWS) over two months.

Development of the Falcon 40B prototype began in August 2022, with the first few months leading up to training dedicated to developing and validating custom tools and conducting experiments to improve design options and data sets.

Pre-training data was collected from public website tracking to build a pre-training dataset for the model. A pre-training database of approximately five trillion tokens was compiled, using data from the CommonCrawl platform and after extensive filtering (to remove machine-generated text and inappropriate content) and removal of duplicates. To enhance the capabilities of the model, this dataset was then expanded using a range of high-quality sources such as research papers, books, and talks.

Finally, the performance of the “Falcon 40B” model was verified by following internationally accepted evaluation standards in the field of generative artificial intelligence and generative language model evaluation tools.

There are always difficulties and challenges, but thanks to the grace of God Almighty, belief in the ability to provide the best, and determination and hard work with the work team, the difficulty is not in our dictionary! We started our plan in the summer of 2022 and our goal was to develop one of the most powerful language models in the world (better than the GPT-3 model from OpenAI),

We are now ready to launch the “Falcon 180B” model. Stay tuned for new information and the next achievement


Do you expect the use of artificial intelligence, such as the Falcon 40B model, to bring about changes in the labor market?

It is certain that the “Falcon 40B” model and other linguistic models will lead to highlighting new requirements in the labor market over time, as generative artificial intelligence contributes to changing the way we work and it is certain that various sectors will be affected by it, such as health care, education, travel, and tourism. And engineering.

Language models such as the Falcon 40B will help unleash a new era of productivity and create job opportunities in sectors that were not available before.

Sophisticated generative AI models and applications can perform a range of routine tasks such as reorganizing and classifying data, but they can write texts, compose music scores, and create digital art that has captured headlines and convinced consumers and the market to give it a try.

Generative AI and other technologies currently in use have the potential to automate work tasks that take up 60 to 70 percent of employees’ time daily. We have previously estimated that technology has the potential to automate half of the time employees spend at work.

Our research and other research from global consulting firms that focus on artificial intelligence indicate that generative artificial intelligence is capable of changing roles and enhancing performance in many functions such as sales, marketing, customer service, and software development, which contributes to pumping billions of dollars across various sectors such as banking. Financial services and even life sciences.


What are your plans for developing the Falcon 40B model and the artificial intelligence scene in the UAE?

There are many expected use cases for the Falcon model, but we are particularly anticipating applications that will contribute to automating and reducing repetitive and tedious tasks. The Falcon model will help many Emirati companies by enhancing the level of efficiency of their operations, organizing internal procedures, and giving employees sufficient time to focus on relevant tasks. Importance. At the individual level, chatbots based on the Falcon model will be able to help users with many tasks in their daily lives.

As we mentioned previously, we are currently preparing to launch the “Falcon 180b” model and modern and developed versions of modern linguistic models. Generative artificial intelligence is a very advanced field and we are pleased with our role in advancing the process of strengthening the innovation system in the Emirate of Abu Dhabi and consolidating the position of the United Arab Emirates as an advanced center among the ranks of leading countries in the field of artificial intelligence.


What role can the Falcon 40B model play in supporting innovation and scientific research in the United Arab Emirates and the Arab region in general?

The “Falcon 40B” model is a major achievement as it is the first open-source artificial intelligence model in the United Arab Emirates and the Middle East region. The huge linguistic model includes 40 billion variable factors and was developed by the team of the Technology Innovation Institute, the applied research arm of the Technology Research Council. Advanced in Abu Dhabi.

The Institute of Technology Innovation’s decision to make the model open source represents a pioneering step in the field of large language models, as they usually remain closed due to the great competition between technology companies.

This pioneering achievement has contributed to consolidating the position of the United Arab Emirates and enhancing its role in the global dialogue on the advancement of artificial intelligence technologies. It also emphasized its key role in promoting innovation and scientific research in the field of artificial intelligence and raising access to it to achieve the common good of the individual and society.

As an open-source artificial intelligence model, the Falcon model will contribute to enhancing the level of transparency globally, facilitating uses and applications, and building the commercial viability of large linguistic models at the level of the UAE and the Arab region as a whole.

As mentioned previously, we hope to accelerate the process of using the Falcon 40B model in various industries and academia, which will undoubtedly enhance the UAE’s position as a leading country in the field of generative artificial intelligence and demonstrate the growing potential of the Arab region as an innovative center for artificial intelligence solutions. and transformative technologies.


What are the main uses you envision for the Falcon 40B model, and how will this language model improve the lives of both people and businesses?

The enormous potential of artificial intelligence can be harnessed through continuous research, responsible use, cooperation with stakeholders, and the presence of many examples that support the contributions of artificial intelligence in various fields.

Following our global call for proposals for innovators around the world to submit their most innovative use cases for the Falcon 40B model, the Technology Innovation Institute has been appointed to award the most innovative and inspiring use cases with computational capabilities for training as a potential investment to accelerate their progress towards commercialization. In various industries.

Here are some ways in which artificial intelligence systems, such as the “Falcon 40B” model, can be used to develop these industries while at the same time protecting individuals and improving the operational efficiency of companies and institutions:
o Healthcare - exploring the possibility of using artificial intelligence technologies to analyze medical images, diagnose diseases, recommend appropriate treatment, or detect drugs.
o Energy and Sustainability – Addressing environmental changes such as climate change modeling, energy optimization, or waste management.
o Education - adapting educational content and creating personalized learning experiences for trainees as well as tracking and evaluating progress in the learning process.
o Business - Facilitating communication and promoting problem-solving as well as the ability to develop tailored solutions that can be implemented according to the specific requirements of the organization or company.

In light of the great interest that the Falcon 40B model has received, the UAE will take a major role in the field of innovative artificial intelligence solutions, which will enhance its role as an incubator for innovative ideas and advanced technology.


How can the Falcon 40B model help consolidate the UAE’s position among the leading countries in artificial intelligence technology?

The launch of the “Falcon 40B” model is a tangible indicator of the UAE’s entry into the arena of advanced language models, which undoubtedly contributed to strengthening the country’s position as a major player in this rapidly developing field.

The “Falcon 40B” model represents another stop on the UAE’s strategic roadmap towards enhancing its leadership role in the field of artificial intelligence, which began with the Technology Innovation Institute’s introduction of the Arabic natural language processing model “Noor” in 2022.

In line with the UAE National Strategy for Artificial Intelligence 2031, the UAE’s participation in global technological developments such as the Falcon 40B model is a critical element in our journey towards strengthening our leadership in innovation in the field of artificial intelligence.

As we mentioned previously, our global call for proposals shows our active participation in the public landscape by pumping investments, developing artificial intelligence systems, and strengthening cooperation.

These initiatives contribute to creating new economic, social, and educational opportunities for the community and also help consolidate the UAE’s position as a major driver of advanced artificial intelligence solutions.


Are there any security or ethical risks associated with artificial intelligence models such as Falcon 40B, and how can they be mitigated?

By making the Falcon 40B model open source, we encourage transparency and accountability in the development of artificial intelligence, and we call on the global community to contribute to improving it and ensuring compliance with ethical standards and considerations in its development and use.

The issue of regulating artificial intelligence technologies remains an ongoing debate that needs to be fully analyzed, studied, and explored by many stakeholders in the field of artificial intelligence from various academic, governmental, and industrial circles.

It is also necessary for these regulations and regulations to focus on aspects of ensuring transparency and accountability in artificial intelligence systems, promoting aspects of fairness and non-discrimination, protecting privacy and data security, and establishing mechanisms to address the social impact of artificial intelligence.

It should be noted that strengthening the strategic regulatory framework is crucial to developing and deploying artificial intelligence technologies responsibly while addressing the potential risks associated with them.

As I mentioned previously, it is the responsibility of governments, stakeholders in sectors, scientists, and specialists to find optimal solutions for regulating artificial intelligence technologies that include a dynamic and continuous process of evaluation, modification, and learning from real-world applications.

The biggest stories delivered to your inbox.

By clicking 'Register', you accept Arageek's Terms, Privacy Policy, and agree to receive our newsletter.

Comments

Contribute to the discussion