In this episode of AI Rising, hosts Leslie D'monte and Jayanth N Kolla take a closer look at the Isreal-Palestine conflict, shining light on the formerly impenetrable Iron Dome, how Hamas managed to penetrate it using low-tech, and how this situation has led to funds drying up for the AI Industry. Referring to the 60-Minute interview with Geoffery Hinton, the godfather of Artificial intelligence, the hosts segue into a discussion of why AGI washing needs to stop and how accepting the opaqueness of deep learning algorithms is healthy but instilling deep-seated anxiety when it comes to AGI is distracting and counterproductive. On a light note, they even contemplate the possibility of humans being treated like animals by AGI and how that could be a blessing in disguise. Having said that, a chatbot has a trillion connections as compared to a human brain which has about 100 trillion connections, and that’s where lies the differentiator. Because, beyond a certain level of complexity, the human brain is incomprehensible to us too. And so, the hosts argue that we don’t give the human brain enough credit; challenging the idea of AI replicating human-like behavior and highlighting the unique capabilities of both humans and machines. While discussing the near future of AI, they speak of how and why RetNets will replace Transformers. And how using stable diffusion, AI is now being used to tap into the intricate patterns of human brainwaves, thoughts and dreams. This episode on Artificial Intelligence is highly recommended by its production team - Tune in Now!
Further Reading:
Geoffrey Hinton's 60-Minutes Interview: https://www.cbsnews.com/news/geoffrey-hinton-ai-dangers-60-minutes-transcript/
Find out more on how AI is used to tap into the human brainwaves: https://www.linkedin.com/posts/sandrookropiridze_ai-techforgood-technology-activity-7114949788780904449-wR09/
Neural networks or deep learning neural networks are a subset of machine learning (ML), and are at the heart of deep learning algorithms. They consist of node layers, containing an input layer, one or more hidden layers, and an output layer. Each node connects to another and has an associated weight and threshold. If the output of any individual node is above the specified threshold value, the node gets activated and sends data to the next layer of the network. Otherwise, no data is passed along to the next layer of the network. While recurrent neural networks (RNNs) are commonly used for natural language processing (NLP) and speech recognition tasks, convolutional neural networks (ConvNets or CNNs) are more often used for classification and computer vision tasks.
CNNs perform better than other neural networks with image, speech, or audio signal inputs. With each layer, the CNN increases in complexity, identifying greater portions of the image. Earlier layers focus on simple features, such as colors and edges. As the image data progresses through the layers of the CNN, it starts to recognize larger elements or shapes of the object until it finally identifies the intended object.
RNNs use sequential data or time series data. These deep learning algorithms are commonly used for problems such as language translation, NLP, speech recognition, and image captioning; they are incorporated into popular applications such as Siri, voice search, and Google Translate. Like feedforward and CNNs, RNNs too use training data to learn. For instance, to make sense of an idiom such as feeling “under the weather”, commonly used when someone is ill, RNNs need to account for the position of each word in the idiom and use that information to predict the next word in the sequence.
RetNet Vs Transformer
Developed by Google Brain researchers, Transformers have become the de facto architecture for large language models (LLMs) since they overcome the sequential training issues of RNNs. Microsoft researchers now say that Retentive Network (RetNet) can become a more efficient foundation architecture for LLMs, simultaneously achieving training parallelism (which speeds up training), low-cost inference (using the model to learn from historical data), and good performance