Technology is profoundly interwoven with our everyday lives, with new innovations in natural language processing, machine learning, and AI causing a ripple affect through many sectors and industries. We wanted to share some interesting developments in the industry from 2023 that could impact our future.
How Large Language Models Could Assist with Teaching
Stanford researchers have developed a number of LLM and AI tools to help teachers, which they hope to one day integrate into a single platform.
NLP has limitations when it comes to teaching, but there are many ways it could improve processes and experiences for teachers, amplifying their strengths.
LLMs can analyze the comments in online lectures to identify when students are most engaged and categorized student comments to identify the greatest source of confusion.
Researchers found that models can reframe teacher’s feedback using growth mindset language to improve student outcomes.
New Chatbots Focus on Personalities
This year, Meta AI launched its Personas platform, a series of chatbots modeled after famous individuals.
Currently, the platform features 28 characters. While each character has a fictional name and personality, they utilize the voice and images of partnering celebrities.
The chatbots interact with users using natural language processing to provide advice on specific interests and converse as if they were real people.
The platform was created using a custom NLP model based on Meta's LLaMA 2 LLM.
Flagging for Depression and Anxiety in Healthcare Workers
New York University researchers have shown that NLP can detect concerning indicators of psychological distress in healthcare workers.
The NLP model analyzed anonymized psychotherapy transcripts from 820 healthcare workers and the same number of those from other industries as a control.
The tool was able to flag certain phrases as evidence of anxiety and depression, and detect topics related to common healthcare issues. Healthcare workers who discussed topics related to their work with their mental health professional were more likely to be diagnosed with anxiety or depression.
In the future, this tool could be used to catch early warning signs of mental health concerns in healthcare workers and provide recommendations for them to seek help from a mental health professional.
Innovations Saving Rainforests from Deforestation
Conservationists in Brazil are using AI tools to predict areas of the Amazon rainforest where deforestation is likely to occur next, allowing them to take preventative measures more efficiently.
The tool analyzes numerous factors including those that make deforestation less likely, such as inhospitable terrain and the presence of Indigenous communities, and factors that make deforestation more likely, such as population density and road infrastructure.
After its analysis, the tool is able to provide automated mapping and updates on areas of concern that are far more frequent than what conservationists could achieve on their own.
It is the hope of the conservationists that further use of this tool will allow a greater focus on preventing deforestation rather than merely penalizing it.
Generative AI Sees New Capabilities
Last year, there was an increasing interest in generative AI as new technologies were released starting with OpenAI’s ChatGPT. Since then, generative AI has grown in its abilities and reach.
While originally the training datasets were from 2021, training datasets now cover up until mid-2023 and allow the AI to perform searches for more current information.
While some generative AI are for general purposes that cannot speak as closely to intensive fields, there has been some progress in the creation of specific generative AI engines for specialized fields such as law, medicine, and climate adaptation.
Generative AI has been extremely helpful with programming, streamlining processes for experienced developers and providing new developers with avenues for exploration.
Anticipating Brain Tumor Progression with Machine Learning
Researchers from the University of Toronto and the University of Waterloo are investigating the use of machine learning to predict the progression of brain tumors.
The researchers began by analyzing a small number of MRIs that show the cancer progression where no interventions were used, serving as a baseline.
The machine learning model provided specific growth estimates specific to each patient.
Future research will expand the dataset to analyze and predict how tumors will respond to various treatments, providing a valuable, life-saving tool.
We believe that natural language processing has an infinite range of possibilities to make the world a better place, and we’re excited to see that dream take shape. We have been working on our own advancements, such as collaborating with the UBC Master of Data Science program to improve Unigrams, our qualitative analysis tool. There is a wealth of knowledge and insights hidden in qualitative data, and our goal is to help others discover them.