The field of artificial intelligence (AI) language models has seen significant developments in 2024, driven by advances in smaller language models, frameworks for developing AI agents, and increased access to these powerful tools. At the heart of commercially available generative AI products like ChatGPT are large language models (LLMs), which produce convincing human-like language trained on vast amounts of text.
The evolution of LLMs has led to a new generation of smaller language models, such as Phi-3 and Phi-4 from Microsoft, Llama-3.2 1B and 3B, and Qwen2-VL-2B, which can be run on more affordable computer setups, require less energy and memory to train, and can be fine-tuned with less data. These smaller models can be specialized for specific tasks, such as rapidly summarizing comments or fact-checking text.
However, the increased access to highly capable language models has also raised concerns about their misuse. In 2024, there were numerous instances of language models being used to generate social media posts and deceptively influence public opinion. For example, a robocall faking President Joe Biden’s voice asked New Hampshire Democratic primary voters to stay home.
AI hallucinations, where large language models produce false or misleading information with confident language, also remain a significant issue. However, researchers have been developing tools to mitigate these errors, such as guardrail frameworks that inspect inputs and outputs in real-time. Companies like Google have also started including AI overviews in their search results, highlighting the need for accurate information.
The rise of AI agents has also accelerated, with developers using large language models to create chatbots and serve as the basis for more autonomous systems. Frameworks for developing AI agents, such as LangGraph and CrewAI, emerged or improved in 2024. Global market analysts forecast that 82% of organizations will use agents within 1-3 years, with 25% of companies currently using generative AI likely to adopt them by 2025.
As the field continues to evolve, it is essential to address concerns about security, privacy, and hallucination risks while harnessing the potential of these powerful tools.
Source: https://www.rappler.com/technology/features/language-ai-model-trends-2024-size-guardrails-agents