In January, OpenAI’s language model ChatGPT achieved a record-breaking expansion of its user base, signifying an accelerating intrigue and investment in AI technology. This feat set the tone for a year rich in AI innovations, indicating a strong public and corporate interest in the capabilities and applications of AI ( https://opendatascience.com/meta-ai-releases-new-large-language-model-llama/ ).
Come February 28th, the tech world buzzed with Meta AI’s announcement of their new expansive language model, LLaMA. This model was created as a comprehensive tool for NLP researchers and practitioners around the globe, with an emphasis on making advanced AI technologies more accessible. LLaMA was extensively trained on a vast array of text data, supporting a multitude of languages, particularly those with Latin and Cyrillic scripts, which marked a stride towards democratizing AI research and application. You can read more on https://opendatascience.com/meta-ai-releases-new-large-language-model-llama/ .
However, the journey of LLaMA took a dramatic turn on March 2nd when its weights were unexpectedly leaked online. This incident propelled a wide-ranging conversation about the access to and distribution of AI technologies, emphasizing the complex issues surrounding the management and dissemination of state-of-the-art AI models.
As the year progressed, Meta provided an update on September 27th that illuminated the growth of the LLaMA ecosystem. The report emphasized its broad adoption across major platforms like AWS, Google Cloud, and Microsoft Azure, reflecting the vibrant and evolving nature of AI development and its substantial implications across various sectors. You can read more here: https://ai.meta.com/blog/llama-2-updates-connect-2023/
These events, in unison, depicted 2023 as a year of both exuberance and prudence in the AI domain. While technological advancements unveiled new potential and generated excitement, they also incited pivotal questions about responsible usage, ethical considerations, and the societal impact of AI. Thus, the year stood as a testament to AI’s transformative potential and the continued effort to responsibly harness its capabilities.
Parallel to these advancements was a surge in visual AI technologies, highlighted by innovative tools like DALL-E, which introduced novel methods of creating digital artwork. This confluence of AI and visual arts not only opened new channels for creative expression but also underscored AI’s extensive influence.
The educational sector also felt AI’s impact, with Khan Academy integrating an AI-powered teaching assistant, Khanmigo, into its curriculum. Meanwhile, Zoom’s AI Companion brought AI support to virtual meetings, and Google harnessed AI to enhance productivity and writing within Google Workspace. For developers, GitHub’s Copilot X became an indispensable tool, leveraging the power of GPT-4 to aid in coding.
Adobe’s Firefly model, too, made waves by revolutionizing the creation of high-quality images and text effects, thereby granting creators new powers. Yet, the swift progress in AI wasn’t without its concerns. Tech luminaries like Steve Wozniak and Elon Musk raised alarms, urging a pause in the development of AI systems more powerful than GPT-4 to evaluate the associated risks of such rapid growth.
The business world saw a significant embrace of AI, with Microsoft refreshing Bing with ChatGPT integration. The concept of AI agents, capable of executing tasks autonomously, came to the fore, although it also sparked debates about job displacement and potential biases.
In September 2023, IBM announced the general availability of the watsonx Granite model series, which are a set of generative AI models. These models are engineered to enhance the infusion of generative AI into business applications and workflows. The Granite models are multi-size foundation models that apply generative AI to both language and code, making them versatile for enterprise use. Recognizing that one size does not fit all, IBM has developed the Granite models in various sizes to better align with the unique needs of different business use cases.
The models are designed on a decoder-only architecture, allowing businesses to apply advanced AI capabilities such as retrieval augmented generation for searching enterprise knowledge bases, summarization of long-form content, and insight extraction and classification for tasks like customer sentiment analysis.
IBM’s Granite models have been benchmarked for business relevance and have been trained on datasets from domains such as internet, academic, code, legal, and finance, ensuring that they are curated for business applications. IBM has been vigilant in filtering these datasets for objectionable content and benchmarking them against internal and external models to promote responsible deployment, addressing key issues like governance, risk assessment, privacy concerns, and bias mitigation.
The Granite Foundation Models are part of WatsonX. Which is a comprehensive AI and data platform designed for business applications. It merges the capabilities of IBM Watson Studio, which supports data scientists, developers, and analysts in creating, running, and deploying AI based on machine learning, with cutting-edge generative AI capabilities that leverage foundation models. This platform is geared towards the needs of today’s and tomorrow’s businesses, providing a robust suite of tools to scale and accelerate the impact of AI with trusted data across enterprises.
Granite Foundation Models represent a new family of IBM-built foundation models available within the WatsonX.AI Studio, which is a component of the WatsonX platform. These models apply generative AI to tasks involving language and code, designed to help businesses scale AI and create AI applications that are unique to their specific business needs and use cases. WatsonX provides the infrastructure and tools necessary for businesses to leverage these models effectively and responsibly.
The company is committed to transparency and responsible AI, demonstrated by the publication of details about its training methodologies for the Granite models. These details encompass the architecture, capabilities, data governance, training algorithms, compute infrastructure, energy and carbon footprint, testing and evaluation, socio-technical harms and mitigations, and usage policies.
IBM also assures clients of its standard intellectual property protections for these models, providing an IP indemnity for its foundation models. This step is part of IBM’s broader commitment to earning trust by introducing powerful new technologies responsibly.
Nvidia continued dominance in the AI GPU market was clear, as they continued to harness their robust software ecosystem and hardware prowess to stay ahead. Simultaneously, the advent of Small Language Models (SLMs) brought a nuanced shift in AI’s application, pivoting towards targeted, specialized tasks that cater to specific industry needs. This nuanced approach signaled a broader industry trend towards specialized, efficient AI solutions. The competitive zeal in the AI arena was palpable, with the debut of platform X’s Grok chatbot, Google’s versatile Gemini model, and AMD’s aggressive entry into the GPU competition, each contributing to a diverse and bustling AI ecosystem.
The AI landscape in 2023 witnessed not just competition but also significant collaboration, as seen in Pika’s innovative venture into the AI video app domain and the strategic alliance between tech giants like Meta and IBM. These moves underlined a shared dedication to fostering a cooperative environment for AI progress. Meanwhile, Intel’s “AI Everywhere” initiative illuminated their extensive array of AI products, underlining the tech industry’s resolve to weave AI seamlessly into the fabric of daily technological applications, from business to consumer levels.
Despite the specter of an AI investment bubble, the stock market’s rally, driven in part by AI innovations, suggested robust optimism for AI’s future. Moreover, the year saw substantial investments in AI, with companies like Mistral.ai advocating for open-source AI. Meanwhile, corporate events like the brief ousting and subsequent reinstatement of OpenAI’s CEO, Sam Altman, highlighted the challenges and power dynamics in the swiftly evolving AI landscape.
The year 2023 was a cornerstone in AI regulation both in the US and the EU, which played a significant role in shaping the open-source ecosystem and guiding responsible AI development.
In the US, President Biden issued an Executive Order laying out a comprehensive strategy for responsible AI innovation. This directive established new standards for AI safety and security, promoted equitable and civil rights, and took extensive measures to protect American privacy and workers. It also highlighted the importance of fostering innovation and competition, reinforcing America’s leadership in AI both domestically and globally – The White House – https://www.whitehouse.gov/briefing-room/statements-releases/2023/10/30/fact-sheet-president-biden-issues-executive-order-on-safe-secure-and-trustworthy-artificial-intelligence/ ).
Moreover, the US took steps toward AI governance with the introduction of the Artificial Intelligence Research, Innovation, and Accountability Act of 2023 (AIRIA). This act highlighted the need for transparency in generative AI, recommending sector-specific oversight for High-Impact AI Systems, including those affecting housing, employment, credit, education, healthcare, or insurance. It also mandated risk management assessments and transparency reports for critical-impact AI systems, ensuring they align with the Constitution’s rights and safety regulations as DLA Piper states: https://www.dlapiper.com/en-us/insights/publications/ai-outlook/2023/us-senators-introduce-bill-to-establish-ai-governance-framework.
The EU AI Act is a proposed regulation by the European Union designed to ensure the safe and lawful development and use of artificial intelligence within its member states. It introduces a legal framework for AI systems, categorizing them based on their risk to citizens’ rights and safety. High-risk AI systems face more stringent requirements. The Act aims to foster innovation while protecting fundamental rights and preventing discrimination, ensuring that AI systems are transparent and have human oversight. Slated for implementation in 2025, the EU AI Act seeks to foster innovation while protecting citizens’ rights. ( https://www.europarl.europa.eu/news/en/press-room/20231206IPR15699/artificial-intelligence-act-deal-on-comprehensive-rules-for-trustworthy-ai )
The AI Alliance, formed by Meta and IBM among others, illustrates the growing trend of collaboration in the AI industry, aiming to promote open-source AI development. It brings together major tech companies, startups, nonprofits, and public institutions to develop open foundation models, provide free benchmarks, and advocate for responsible AI development. This alliance underscores the importance of open collaboration in advancing AI technologies while ensuring ethical standards and broad. The AI Alliance aims to advance open, safe, and responsible AI by fostering collaboration among a diverse range of organizations. It focuses on developing benchmarks, standards, and educational resources to promote responsible innovation and address safety concerns in AI. IBM views this as a pivotal moment for defining the future of AI and is committed to ensuring that the open ecosystem drives an innovative agenda underpinned by safety and scientific rigor. Details on AI Alliance: https://thealliance.ai/ .
These regulatory efforts and collaborative initiatives reflect a commitment to balance AI’s transformative potential with ethical considerations and societal well-being, setting a framework for the future of AI development and integration.
Ultimately, 2023 in AI was not solely about technological leaps forward; it was equally about grasping the broader ramifications of these advancements for society, ethical considerations, and the future interplay between humans and machines. It was a year that shaped the trajectory for AI’s future, blending innovation with a conscientious approach to development and usage. From ChatGPT’s explosive growth to the democratizing force of Meta’s LLaMA, AI’s reach expanded across industries. IBM’s WatsonX and Granite models stood as testaments to innovation tailored for business precision. Amidst this technological renaissance, the US and EU laid regulatory frameworks, ensuring AI’s safe trajectory. The AI Alliance, a testament to collective ambition, showcased a unified stride toward an AI future that’s open, ethical, and inclusive, reflecting a year where AI’s potential was matched by a conscientious approach to its integration into the fabric of society.
