Anna Fellet’s Post

Some of my takeaways after the first day of #JIAMCATT2024: - The AI brain fog is lifting, we're talking concrete use cases now. Abstract comparisons between human and presumed artificial intelligence have become pointless - The developments since last year's JIAMCATT confirm that IOs (I've recently discovered this handy abbreviation for international organizations) had actually been technologically pretty well prepared; skilled and motivated people doing interesting things everywhere - When asked about the most relevant application of AI, it still is... MT, where of course AI is at best only an incremental improvement over previous technology (the now old hat NMT), especially when considering languages that get less the limelights than, say, English (had you noticed?). And the operational model of MT remains postediting, even if that's still not being done consistently. Not much disruption to be seen there after all - The last mile, as Adam Bittlingmayer has called it, seems, strangely, to get longer than ever. There we were confronted with a completely unheard-of situation where a UNESCO colleague dared give her speech in French, in Paris, and part of the audience was outraged -- where was the AI magic then? - The density of lucid people is definitely well above average in IOs' language services (at least it was until I entered the room); especially the wisdom that the logical effect of technological progress is that the value of human capital *increases* (dixit James Phillips) should be spread more widely - IOs should develop shared guidelines about reasonable use of AI for multilingualism so we don't all have to figure out individually how to explain to our stakeholders that "computer-assisted translation" remains a valid depiction of today's state of the art

Reading and thinking about "The AI Revolution Is Crushing Thousands of Languages," by Matteo Wong. The dominance of English in the digital world has long disadvantaged non-English speakers, limiting their access to essential online services and tools. Despite the internet's potential to bridge language and cultural divides, most websites and platforms cater primarily to English users. This linguistic barrier is further amplified by the rise of generative AI, which, instead of fostering inclusivity, risks solidifying English's dominance online and beyond. When everyday technologies fail to recognize or accommodate something as fundamental as a person's name, language, or cultural background, it undermines their very sense of identity and dignity. https://lnkd.in/eXw4DtNw

The Language Gap in AI: A Call for Inclusion As we navigate the rapid advancements in artificial intelligence, we face a critical challenge: the potential exclusion of thousands of languages. Recent studies reveal a significant oversight within prominent AI models, such as the Fon language—spoken by millions in Benin and neighbouring regions—being incorrectly identified as "fictional." This issue highlights a larger problem: the prevalence of English online and in digital tools, from search engines to voice assistants, which sidelines a vast portion of the global population. Although English is a key language for business and science, it's important to remember that over 80% of the world does not speak it. As technology and communication professionals, we have both an opportunity and a duty to support and create AI technologies that are genuinely inclusive. #AILanguageInclusion #TechForGood #DigitalEquality #InclusiveAI Read more at The Atlantic: 'The Language Gap in AI: A Call for Inclusion': https://lnkd.in/eiMvBd_M

✨ HyperCLOV A X: Bridging Language Barriers with a Multilingual AI Powerhouse ✨ 💡 Introduction: Introducing HyperCLOV A X, an advanced AI language model designed to master the nuances of the Korean language, while also excelling in English, math, and coding tasks. Developed by NA VER Cloud Platform, this model is a leap forward in multilingual AI, leveraging a mix of Korean, English, and code data, fine-tuned with human-annotated datasets for superior performance across various domains. ⚙️ Main Features: HyperCLOV A X stands out with its exceptional reasoning, commonsense, and factuality in both Korean and English. It showcases robust cross-lingual proficiency, generalization abilities, and has been developed with a focus on safety, undergoing systematic red teaming and safety data collection processes. With its tokenizer vocabulary size tailored for optimal performance, it demonstrates state-of-the-art machine translation capabilities and a commitment to ethical AI development. 📖 Case Study or Example: HyperCLOV A X has been instrumental for regions seeking to develop sovereign Large Language Models (LLMs). By providing guidance and demonstrating its prowess in real-world applications, such as chatbots and translation services, it has become a valuable asset for companies like Naver, setting a precedent recognized by international bodies like the United Nations. ❤️ Importance and Benefits: The significance of HyperCLOV A X lies in its ability to understand and generate language with a high degree of cultural and contextual relevance. It benefits the industry by offering a scalable solution for multilingual communication, facilitating more inclusive technology development, and ensuring safer AI interactions by prioritizing the removal of harmful content. 🚀 Future Directions: Looking ahead, HyperCLOV A X is poised to explore multimodal capabilities, improve model efficiency through quantization, and enhance integration with external tools. The ongoing commitment to advancing AI technologies promises to open new avenues for innovation and societal benefit. 📢 Call to Action: Discover the full potential of HyperCLOV A X and how it's transforming the AI landscape. Dive deeper into the world of multilingual AI and join the conversation on the future of ethical, inclusive technology. Visit [NA VER Cloud Platform](https://lnkd.in/eTX9cyz3) for more details. Read the full paper here: [https://lnkd.in/ewAs4xwP] #HyperCLOVAX #MultilingualAI #LanguageModel #EthicalAI #MachineTranslation #ArtificialIntelligence #TechInnovation #NAVERCloud

Cohere for AI's Aya Surpasses Language Barriers, Here's How... The Game Changer... Cohere for AI's Aya, an open-source multilingual large language model, offers unprecedented language coverage. It understands instructions in over 100 languages, including less common ones like Azerbaijani and Welsh. Behind the Scenes... Aya's superiority stems from a unique dataset of prompt/completion pairs, enhanced by both machine translation and human curation. This ensures cultural nuance is captured, setting a new standard for multilingual LLMs. The Impact... Aya's release marks a significant leap in language research, preservation, and AI inclusivity. It not only surpasses existing open-source multilingual LLMs but also opens new avenues for global communication and understanding. The Future Awaits... This breakthrough has the potential to democratize AI language models, making them more accessible and inclusive. Aya's advancement could lead to broader adoption and innovation in AI technologies across diverse linguistic landscapes.

ALMOST 50% OF WORLD LANGUAGES MIGHT BECOME EXTINCT SOON, BUT AI MIGHT HELP Did you know that approximately every two weeks, a minority language dies? According to UNESCO, there are about 6,000 languages spoken in the world today, and aproximately 3,000 of them might go extinct by the end of the 21st century. This represents a significant loss to the diversity and richness of human culture and heritage. However, there is hope – artificial intelligence. How can it be helpful? AI can process and store large amounts of linguistic data, identify patterns, create new ones, and assist with translation and transcription. It can also help in creating digital platforms and resources for learning and teaching endangered languages, as well as raising awareness and interest among speakers and learners. For example, in Australia, a team of linguists and computer scientists has created an AI system that can automatically analyse and transcribe recordings of the Aboriginal language Jingulu, which is considered an endangered language with only several elderly speakers remaining. Even though this language is relatively simple, it demonstrates how this idea can be used in the future. This application just illustrates how vast the potential for the use of AI is. At InoCloud, we are excited about every piece of information like this, and we hope our solutions will contribute to their widespread use in the future. Which application of AI are you most looking forward to in your business? #InoCloud #artificialintelligence #language #technology #applications

Nigeria launches first multilingual large language model for inclusive AI development Nigeria introduces its first Multilingual Large Language Model (LLM) to promote inclusive AI development and strengthen national computing capacity.

Silo AI is releasing Poro, our take on open source LLMs. Our goal is to make state-of-the-art LLMs available for all official EU languages, supporting not just technological advancement but also cultural and linguistic inclusivity and European digital sovereignty. Poro represents a breakthrough in training LLMs for low-resource languages, outperforming existing models in Finnish and setting a new standard for AI development aligned with European values. By integrating high-resource languages in its training, Poro has achieved superior performance in Finnish without compromising English proficiency, showcasing an effective approach to enhancing linguistic diversity in AI. Happy to discuss further on how to use LLMs to create value with AI in your business. Read more about Poro and its implications for European AI and language diversity in our blog. Link in comments. #LargeLanguageModels #GenerativeAI #AIforPeople

by Joel Nwankwo  November 6, 2023 "Artificial intelligence (AI) is still changing the world in many ways, offering humans both opportunities and risks. Given its large language model, AI requires a huge amount of data, an area in which Africa is underrepresented... Given the ongoing demand for artificial intelligence systems for language translation, it is clear from the discussion of these programs that Africa cannot avoid this technology. Economic, social, and cultural localization is increasingly required as a result of globalization, and this need for efficient and successful outcomes necessitates the use of technology. The preservation and promotion of Africa’s history, culture, and native languages through artificial intelligence algorithms can be extremely beneficial. Governments in Africa need to know how to move forward with implementing strong AI regulations and creating thriving AI ecosystems that sustainably support startup growth and the advancement of research and engineering talent, given that AI development is advancing quickly and has the potential to boost economic growth The rise of AI has also given rise to chances to design more inclusive AI systems, democratize existing AI practices, and boost the involvement of underrepresented communities in AI research. Research projects that artificial intelligence will boost the world economy by $15.7 trillion by 2030. However, there hasn’t been a particularly equitable distribution of power and finance made possible by AI technology up to this point. The benefits of these technologies have not been felt throughout the African continent, despite the global reach of the AI supply chain. Instead, they are mostly enjoyed by Big Tech businesses in China and North America. However, the African continent has a special situation that makes it possible to use AI technologies to promote local socio-economic growth. This includes the potential to bypass the Third Industrial Revolution’s technological infrastructure and a youthful population that is expanding and dynamic and prepared to adopt new digital jobs and entrepreneurial ventures. Adopting AI solutions that facilitate the achievement of national developmental agendas and foster inclusive and successful societies needs to be a top priority for African governments and developers." #metaglossia_mundus: https://lnkd.in/eX9HtbGq  

"In addressing the English language dominance, the Spanish LLM project “seeks to provide an open, public, and accessible language infrastructure for companies, so that they can have a large-scale language model that facilitates services that allow them to develop advanced applications for their economic exploitation, such as intelligent assistants, conversational and dialogue systems or content generation models”." #llms #largelanguagemodels #languagemodels #generativeai #ai #artificialintelligence #automation #contentgeneration