2024 was a huge year for AI; recording significant growth, integrations, and adoption.
It all started with the NVIDIA GTC 2024 AI conference, held in March. At the conference, the NVIDIA CEO declared that AI has closed the technology divide: “You don’t have to be a C++ programmer to be successful.”
Then came the AI Summit in Washington, D.C., held in October. Again, Nvidia showed off its AI software platforms, dedicated to helping businesses launch their AI solutions.
AI continues to gain a lot of attention, with talking points emerging in privacy, security, and compliance.
That said, here are hot topics expected to shape AI in 2025.
- Multimodal AI
Unlike large language models, which process only data, multimodal AI can process diverse data types, including audio, videos, and images.
If you are already familiar with OpenAI’s DALL·E and Google’s Gemini, 2024 saw JP Morgan launch DocLLM, and Meta also flaunted its multimodal solution Chameleon.
Several firms have introduced multimodal solutions in healthcare, and applications have emerged in finance and customer service.
- AI and Robotics
Tesla combines advanced AI with inference hardware to achieve autonomy in humanoid, self-driving robots.
Last week, Elon Musk showcased Tesla’s general-purpose humanoid robot, Optimus, which he claimed has infinite potential and can do anything.
During Tesla’s ‘We, Robot’ showcase, we saw robots mixing drinks, dancing, and socializing with guests.
Whether they were controlled by autonomous AI, remote-controlled, or teleoperated, time will tell.
Meanwhile, other firms like Toyota and Boston Dynamics are partnering to develop AI-powered robots.
- AI Regulations and Ethical Concerns
The EU AI Act was finalized in 2024 and is scheduled to take effect in 2025. The major talking point is the regulation of the providers of high-risk AI systems, i.e., those used in critical infrastructure or law enforcement.
The US hasn’t got a clear regulatory framework but AI, but expect to see some changes once the election year is over.
In September, the Cyberspace Administration of China released a draft regulation aimed at regulating AI-generated content. i.e., the key talking point is the labeling of AI-generated content.
- Shadow AI
Refers to employees using AI within an organization without adequate oversight or approval.
Microsoft reported that 75% of global knowledge workers use AI.
Currently, organizations have slow IT approval processes and no clear AI policies.
- Virtual Agents
2024 saw firms like Zoom update its Zoom Virtual Agent (ZVA), its AI-powered chatbot.
With a GenAI makeover, Zoom plans to extend its capabilities to voice (AI Virtual Voice Agent) and multi-intent detection for complex issue resolution.
AWS in September launched AWS Re: Post Agent, a generative AI-powered assistant that allows cloud developers to get faster technical guidance.
ROG gaming laptop manufacturer ASUS updated the virtual assistant software in the ROG Zephyrus G16.
The VA software is also available for AMD Ryzen™ AI 300 Series processor models as an update.
The update brings a Q&A interface, document summarization, and voice transcription capabilities.
- Open Source AI
Mark Zuckerberg is advocating for Opensource to be the new industry standard by launching Llama 3.1 405B, an improved open-source large language model (LLM) released in June this year.
Developers can experiment, fine-tune with their data, and build smaller models with it.
Zuckerberg thinks that open-source AI is good for the world as they are more transparent and can be subject to scrutiny than closed systems.
China’s Alibaba also launched as many as 100 new open-source AI models (Qwen 2.5) in September to compete favourably with rivals. These models are applicable in gaming, automobiles, and science research.
- Retrieval-augmented generation (RAG)
At the NVIDIA GTC 2024, NVIDIA CEO Jensen Huang said that the solution to the problem of hallucination in AI systems is a process known as RAG.
RAG is a technique employed to improve the accuracy and reliability of AI models to avoid hallucinations.
While hallucinations lead to incorrect output or responses, RAG works by linking LLMs to external sources.
Google is developing speculative RAG. While AWS, IBM, and Microsoft have either adopted or are working on RAG solutions.
That’s a wrap. If you think we’ve missed anything, do let us know in the comments.
