Happy New Year everyone, 2023 has certainly been an incredible year for AI, it is remarkable how quickly things are moving. One of my favorite quotes this year was from Michal Kosinski, computational psychologist and professor of organizational behavior at Stanford Graduate School of Business - “The differences between GPT-3 and GPT-4 is like the difference between a horse cart and a 737 – and it happened in a year.”
Here are a few notable achievements/research reported in 2023 (by no means an exhaustive list)
YouTube started working on an AI music tool that will let you use the voices of famous musicians…meaning that you could remake old hits using current artists, think Maroon 5’s Adam Levine singing the Beatles “I Want to Hold Your Hand.”
Elon Musk announced Grok AI bot, stating that it will be better and “spicier” than ChatGPT (I love this use of "spicy", what he means is less censorship).
A research study at Stanford found that LLMs may be developing a humanlike capability known as theory of mind. Theory of mind is a human’s ability to predict another person’s thoughts and emotions (I know this sounds crazy, it's too complicated to explain here, but I have included a link at the end of this article).
Researchers at Google DeepMind developed a type of AI system that demonstrated the capability of social learning like how humans can learn from limited exposure to elements in an environment. This would be a significant breakthrough in how deep learning models are trained.
A research study at the psychosocial science department at the University of Bergan, Norway found that AI has reached at least the same or even surpassed the average human’s ability to generate ideas, known as creative divergent thinking. The AI researchers used the Alternative Uses Test (AUT), used by psychologists on theories of human intelligence.
New York University and Pompeu Fabra University in Spain developed Meta-Learning for Compositionality (MLC), the capacity of artificial neural networks to have compositional generalizations, like how humans grasp and expand new concepts—matching and even surpassing human performance.
Siemens and Microsoft have started using Copilot to write programming code for factory machinery to track down and resolve software bugs. This ChatGPT approach has reduced tasks that had previously taken weeks to complete to a matter of minutes. So, imagine your robotic arm on an auto assembly line has a software bug, it notifies one of your developers that it has discovered a software bug, and has already attempted to fix it. Next, your developer evaluates and modifies the generated code, or starts to interact with Copilot to resolve the bug.
Anthropic, MIT, and Stanford started working on a method, which they call “generative active elicitation” (GATE), using LLMs to help convert human preference into automated decision-making systems. So, the LLMs will ask questions and then learn on the fly.
The rise of the use of robots in warehouses brings up safety concerns. Amazon now has 250K robots and a $1B innovation fund. The total warehouse robot market is estimated to reach $14B by 2030. Today Autonomous Mobile Robots (AMRs) are becoming common in factories and warehouses, for instance, it is estimated that roughly 5% of the US forklift fleet is self-driving. The University of Michigan developed a new approach that would enhance the ability of robots to explore open-world environments and navigate them in personalized ways using LLMs, Zero-shot interactive Personalized Object Navigation (ZIPON) while engaging in conversations with users.
This is merely a snapshot of the rapid evolution and adoption of AI. I believe that we are at a pivotal point in AI and we will only see things accelerate from here. 2024 will be a year where we see much more efficient LLMs (significantly fewer parameters needed), leading to greater capabilities. Many of you may be wondering why I didn't speak of Artificial General Intelligence (AGI), or human-level intelligence. It would appear that we are seeing "sparks of AGI" as Microsoft stated in their March 2023 paper on GPT-4, but I believe that we are not quite there yet. Perhaps in 2024, we'll start to see more evidence of humanlike capabilities, but it is unlikely (if ever) these models will gain agency or self-awareness or any kind of consciousness.
The last thing I will leave you with is the crazier side of today’s evolving AI and honestly, something that feels like sci-fi - Researchers at Indiana University were able to develop what they refer to as “Brainoware”, creating a living brain organoid from human stem cells mounted on a silicon chip to improve traditional machine learning.
References:
Comments