Hello innovators and curious minds of AI Society!

Welcome to the 15th Edition of The WatchTower, your quintessential guide to the riveting world of AI. In this edition, we discuss the most recent release of ChatGPT-4o, and a curious similarity between LLMs and the human brain.

📰 Featured in This Edition:

• GPT-4o

• A Curious Link Between Biology and AI: Lessons from Aphasia

GPT-4o

Introduction

Three weeks ago, the tech giant OpenAI announced GPT-4o, a model that can reason across audio, vision, and text in real-time. The online demonstration has showed this latest flagship model can mimic human cadences in its verbal responses and can even try to detect people’s moods.

What can GPT-4o do?

GPT-4o, short for “omni”, is a faster, easier-to-use version of ChatGPT that features the ability to reason across audio, vision and text. GPT-4o enables users to interact with ChatGPT via voice commands. Unlike previous GPT models, which solely work on text-based inputs and outputs, GPT-4o can process and respond to a wide variety of inputs from the surroundings, making interactions more natural and immersive. With the ability to perceive so much information from its voice, vision, and environmental sensors, GPT-4o is incredibly versatile and capable of handling a wide range of situations. The ability to deliver responses in a natural, human-like voice and perform various vocal characterizations upon request also allows GPT-4o to be used as a personal virtual assistant capable of engaging in real-time spoken conversations.

When will it be available?

GPT-4o is now available in the OpenAI API as a text and vision model. OpenAI announced that in the coming weeks, support for GPT-4o's new audio and video capabilities will be provided to a small group of trusted partners in the API. Also, a ChatGPT desktop app with GPT-4o capabilities will be released in the coming months, adding to the current web and mobile versions of the AI technology.

Concluding Thoughts

Once again, OpenAI stunned everyone with another powerful tool that can interact and assist human users. While GPT-4o showcase immense potential, they are not without risks. Privacy and security are a key area of concern as AI agents like GPT-4o gain access to more personal data and environmental information.

_{Published by David Hung, June 04 2024}

A Curious Link Between Biology and AI: Lessons from Aphasia

What is Aphasia?

Aphasia is a language disorder that impairs a person’s ability to communicate. It typically results from brain injury, such as a stroke, affecting areas of the brain responsible for language. Depending on the type of aphasia, individuals may have difficulty producing or comprehending speech, reading, or writing. Fluent aphasia, such as Wernicke's aphasia, allows for grammatically correct speech that often lacks meaningful content, resulting in sentences that are coherent in structure but nonsensical in substance.

Aphasia and AI: A Surprising Connection

How might this relate to AI, you might ask? Fundamentally, the errors observed in aphasia patients can be seen as a form of "error decryption" in human language processing. This phenomenon offers a fascinating parallel to artificial intelligence, particularly Large Language Models (LLMs). When an aphasia patient speaks, their neurons are misfiring in its attempt to decode or produce language. Similarly, LLMs may generate responses that are grammatically correct but contextually irrelevant, especially when they misinterpret the input data.

How AI Forms Coherent Sentences

LLMs are trained on vast datasets that include diverse language patterns and contexts. They function by predicting the next word in a sequence based on the context provided by preceding words. This process involves complex algorithms that adjust internal parameters continuously, ensuring that the generated text remains coherent and contextually appropriate.

The model uses a method called "transformer architecture," which allows it to weigh the importance of different words in a sentence and their relationships to each other. This architecture helps the model understand context and maintain coherence across longer passages of text. By doing so, the AI can simulate a level of understanding and generation that mirrors human-like language processing.

Lessons from Aphasia for AI Development

Studying aphasia provides valuable insights into the potential pitfalls and challenges faced by LLMs. Just as aphasia highlights the complexities of human language processing, it also underscores the sophistication required for AI to achieve similar capabilities. The parallels between human language disorders and AI errors reveal the intricate nature of producing coherent communication, whether in biological or artificial systems.

Researchers can learn from the mechanisms underlying aphasia to improve AI models. For instance, understanding how the brain compensates for language deficits could inspire new techniques in machine learning to handle errors more effectively and enhance the coherence of AI-generated text.

Conclusion

Understanding the coherence of LLMs through the lens of aphasia not only enhances our grasp of artificial intelligence but also deepens our appreciation of the human brain's remarkable language capabilities. This curious link between biology and AI offers valuable lessons for advancing technology and understanding human cognition. By exploring these connections, we can continue to improve AI systems and gain deeper insights into the complexities of human language.

_{Published by Lucy, June 04 2024}

🗣 Sponsors 🗣

Our ambitious projects would not be possible without the support of our GOLD sponsor, UNOVA.

Closing Notes

We welcome any feedback / suggestions for future editions here or email us at [email protected].

Stay curious,

🥫Sauces🥫

Here, you can find all sources used in constructing this edition of WatchTower:

• UNOVA

• GPT-4o

• Decoding Speech for Understanding and Treating Aphasia