I follow the blogs of Allie K. Miller called AI with Allie. She has just released a huge post summarizing the last year of AI achievements and advances. As a PDF it is about 25 pages. At the end of this summary recap, you will be able to receive a copy of her 50-page FREE ChatGPT Guide with 30 expert ChatGPT prompts and examples. All the images for this block have been created with AI.
This article is a recap of the biggest moments and movements in AI during 2024, written by Allie K. Miller. It covers a wide range of topics, including advancements in AI capabilities, hardware failures, and predictions about the future of AI. Here’s a summary of the main points:
Multimodality was a major theme in 2024, with models like GPT-4o and Llama 3.2 being trained as multimodal from the start, leading to the development of AI that can process various forms of input, such as voice, images, and video. The author uses AI tools daily for dictation, and notes the emergence of AI that can see screens and surroundings. Search is also changing, with LLM-based systems becoming vector-first and more multimodal, incorporating voice input for more personalized results.
AI agents are on the horizon, but the author argues that 2024 was more about AI workflows than true AI agents. She predicts that 2025 will be the year of AI agents, and highlights companies like Crew AI, ReWorkd, and Google’s Project Mariner.
AI hardware experienced significant failures in 2024. The Rabbit AI device, the Humane AI pin, and the Apple Vision Pro were among the disappointments. The Meta Raybans were a notable exception, praised for their audio quality, but the author is not convinced that glasses are the future of AI hardware. The author’s bet is on Airpod-like AI hardware.
Slow AI and reasoning became important in 2024. The author notes that ‘test time’ or ‘inference time’ became a key concept, with AIs spending more time ‘thinking’ to produce better answers. She also points out the importance of personalized, predictive, and proactive AI, which requires improved reasoning. The author cites o1 from OpenAI and Claude 3.5 Sonnet v2 as examples of models that exhibit nascent reasoning capabilities.
AI-generated noise and truth questioning are major issues in social media. The author notes the abundance of bots and inauthentic content. She also mentions early signs of truth questioning but noted it was less than expected during the election cycle.
Robotics saw incremental improvements, with companies like Nvidia, Waymo, and Tesla making strides in areas such as self-driving cars and enterprise robots.
AI automation workflows gained traction, with tools like Gumloop, Make dot com, and Zapier being used for productivity. The author’s team relies heavily on AI workflows to search, rank, summarize and discuss top AI news stories each morning.
AI acceptance grew in 2024, with people becoming more comfortable with the technology and its applications, including using ChatGPT for personal insights and casually discussing AGI.
AI and code are a powerful combination. The author highlights tools like Devin AI, Replit, and Cursor AI, noting that AI is well suited for coding tasks. She also mentions Lovable and Claude Artifacts as tools for non-coders. Additionally, the author believes there is a trend of “partial builders” in products like Uizard/Framer AI for websites.
Energy constraints and chip heat are major concerns for the AI industry. GPU shortages persist, and there are growing concerns about AI’s water and electricity consumption. The author notes the large investments into the chip and AI cloud space.
AI advertising had a rough year, with several ads being criticized for dehumanizing portrayals. The author argues that ads should not strip humans of their humanity.
AI gatherings are becoming increasingly popular, and the author encourages people to attend conferences to network and learn. She also hints at potentially launching her own event.
Model generation saw intense competition, with models from Meta, Mistral, Anthropic, and others catching up to OpenAI’s GPT-4. Small models have also made significant advancements. The author also notes that the cost of models has been plummeting, with GPT-4’s cost dropping approximately 90%.
World models were somewhat forgotten in the rush to create AI-generated video. The author notes that the focus should be on creating 3D representations of the world.
The labor market began to shift with a move towards paying for outcomes rather than labor hours. AI SaaS products are now being priced on value rather than work hours.
Early AGI signals are emerging, with experts accelerating their predictions for when AGI will be achieved. The author provides a detailed list of predictions from various AI experts, noting that most of them have moved their timelines forward. She also includes a 5-level AGI guide reportedly from OpenAI.
Have questions about how to utilize AI to your business advantage? Contact Us.
Swift Chip provides managed IT and cybersecurity services for a wide variety of companies in fields with stringent privacy and compliance requirements.
Review us on Google