• AI Times
  • Posts
  • Google Unleashes the Future: 'Gemini' AI Software Set to Revolutionize!

Google Unleashes the Future: 'Gemini' AI Software Set to Revolutionize!

Author

AI Times
15th septembre 2023

🤖 Hello, AI Enthusiasts! 🤖

Brace yourselves: Google's Game-Changer: AI Software 'Gemini' is about to launch! Dive in now.

In today’s AI Times:

  • ⌛Google's Game-Changer: AI Software 'Gemini' Soon to Launch!

  • 🎥 Today’s capsule: Redefining Boundaries: Generative AI's Leap into Mastering Social Nuances!

  • 🏆 Today’s trick: Make ultra-realistic photos with Midjourney

  • 🔎 AI recent research

  • 🚀 3 Best AI tools for students

📰AI NEWS📰

Scientists have created an AI tool, RETFound, that diagnoses and predicts multiple health conditions using retinal images. Unlike traditional methods, RETFound employs self-supervised learning, eliminating the need to label each training image. The tool's foundation model approach, similar to ChatGPT's training, allows it to adapt to various tasks. While highly effective for ocular diseases, its performance for systemic diseases surpasses other AI models, paving the way for broader medical imaging applications.

Google is nearing the launch of its conversational AI software, Gemini, designed to rival OpenAI's GPT-4. A select group of companies has been granted access to an early version of Gemini. The software encompasses large-language models for various applications, from chatbots to generating text like email drafts and news stories. While developers currently have access to a sizable version of Gemini, an even larger version is in development. Google plans to offer Gemini through its Google Cloud Vertex AI service.

News Corp's CEO, Robert Thomson, warned against the potential job losses and "rubbish" content resulting from AI in journalism. Despite this, News Corp Australia is at the forefront of AI-produced news content, leading to concerns from its staff. The company uses AI for hyper-local news stories, while ensuring human oversight. Other media companies, like Gannett and Nine, are exploring AI's potential, with varying degrees of caution. As the media industry navigates AI's role, the Australian government is examining AI's impact on copyright laws.

In a study published in Scientific Reports, AI chatbots and humans were tested on the Alternate Uses Task (AUT) to measure divergent thinking. The chatbots demonstrated creativity on par with the average human, but top-performing humans still outshined the best chatbot results. The AUT evaluates creativity by asking participants to list alternative uses for common items. While chatbots scored higher on average for originality and creativity, the range of human scores was broader, with the highest human scores surpassing chatbot scores in most categories. The study suggests potential for AI to enhance human creative processes.

🎥 TODAY’S CAPSULE 🎥

Generative AI is a type of artificial intelligence that can create new content, such as text, images, or music. It is trained on a large dataset of existing content and learns to identify patterns and generate new content that is similar to the data it was trained on.

Social nuances are the subtle and often unspoken rules that govern social interactions. They include things like body language, tone of voice, and cultural norms. Learning social nuances is essential for humans to function effectively in society.

Recent research has shown that generative AI can be trained to learn social nuances. One study, published in the journal Nature, trained a language model on a dataset of human conversations. The model was able to learn to identify and generate social cues, such as sarcasm and humour.

Another study, published in the journal Science, trained a computer vision model on a dataset of human facial expressions. The model was able to learn to identify and generate different facial expressions, such as happiness, sadness, and anger.

These studies suggest that generative AI has the potential to learn social nuances. However, there are still some challenges that need to be addressed. For example, generative AI models can sometimes generate content that is offensive or harmful. It is important to develop methods for training generative AI models to be socially responsible.

Here are some of the recent research that explores the future of social learning in generative AI:

  • "Training Socially Aligned Language Models in Simulated Human Society" (2020) This paper proposes a new way to train language models to be socially aligned. The authors argue that by allowing language models to learn from simulated social interactions, we can better align their behavior with human values and societal norms.

  • "Socially Situated Artificial Intelligence" (2021) This paper argues that the importance of social context in AI development. The authors suggest that AI agents can substantially improve their performance and societal alignment through ongoing, real-world interactions with humans.

  • "Towards a General Theory of Social Learning in Artificial Intelligence" (2022) This paper provides a comprehensive overview of the field of social learning in AI. The authors discuss the different approaches to social learning, the challenges that need to be addressed, and the potential applications of social learning in AI.

These are just a few of the many recent research papers that explore the future of social learning in generative AI. This is a rapidly growing field, and I am excited to see what new developments are made in the years to come.

Here are some of the potential applications of generative AI in social learning:

The potential applications of generative AI in social learning are vast. As the technology continues to develop, we can expect to see even more innovative and beneficial applications.

🏆 TODAY’S TRICK 🏆

Make ultra-realistic photos with Midjourney

Achieving realistic images with Midjourney can be challenging due to the intricacy of crafting precise prompts.

Solution: Photorealistic, a chatGPT plugin

Photorealistic ?

The Photorealistic ChatGPT Plugin, a third-party app, empowers GPT to craft photorealistic prompts for Midjourney. This tool enhances GPT's understanding by providing visually engaging prompts, suitable for chatbots, virtual assistants, and more. It offers customization, integration with other tools, and access to pre-built templates. Overall, it amplifies GPT systems, ensuring improved accuracy and user experiences.

To get the perfect image on Midjourney you have to specify :

  • Photo type: Specify image type using keywords like portrait, macro shot, candid, close-up, etc.

  • Camera model: Experiment with camera brands for desired results, e.g., Sony a7R IV, Nikon D850, Canon EOS R5.

  • Focal length & lens type: Define perspective with settings like 18mm or 100mm. Lenses below 85mm suit landscapes; above are for portraits.

  • Depth of field: Describe focus distance for prominence, using terms like narrow depth, wide depth, or bokeh.

  • ISO value: Indicate brightness with ISO values, e.g., ISO 50, 400, 12,800.

  • Aperture: Define focus range with values like f/1.4 or f/16. Consider lens compatibility.

  • Lighting type: Use terms like natural light, dreamlike, dramatic, or neon lighting.

  • Shutter speed: Determine motion appearance with speeds like 5s or 1/120s. Low speeds blur motion; high speeds freeze it.

    NO NEED with Photorealistic

    WITHOUT PHOTOREALISTIC

    ChatGPT

    WITH PHOTOREALISTIC

    ChatGPT

🔎RECENT RESEARCH🔎

The study presents "UniHSI," a novel framework for Human-Scene Interaction (HSI) that facilitates versatile interaction control using language commands. Recognizing the importance of HSI in applications like embodied AI and virtual reality, the authors aim to bridge the gap between motion quality and user-friendly interfaces. UniHSI consists of two main components: the LLM Planner, which translates language prompts into task plans, and the Unified Controller, which executes these plans. A unique feature of this framework is its interaction annotation-free training, leveraging Large Language Models (LLMs) to generate interaction plans. To support the framework, the authors introduced a new dataset, "ScenePlan," containing thousands of task plans based on diverse scenarios. Comprehensive experiments showcased the framework's effectiveness in versatile task execution and its adaptability to real scanned scenes.

The study delves into the realm of Explainable AI, emphasizing the importance of human-interpretable representation learning (HRL). Recognizing the challenges in ensuring that learned AI concepts are genuinely interpretable, the authors propose viewing interpretability as the machine's ability to communicate with a specific human-in-the-loop. They introduce a conceptual and mathematical model for HRL that explicitly incorporates the human element, leveraging techniques from causal representation learning. A significant contribution is the development of an "alignment" notion, ensuring that the machine's conceptual representation aligns with that of the human observer. This alignment is linked to the property of disentanglement, often associated with interpretability. The study further explores various settings of human concept interactions, from simple disentangled scenarios to more complex, unrestricted settings. The research aims to bridge the gap between machine learning representations and human interpretability, providing a foundation for future work in the field.


🚀 BEST AI TOOLS FOR STUDENTS🚀

  • The VoxScript ChatGPT Plugin integrates with GPT-4, offering instant YouTube video transcripts and topic-based video searches. It also provides financial data searches, beneficial for investors and analysts. Additionally, it enhances Google searches with its advanced language capabilities.

    gpstore.ai for more features

  • The Wolfram plugin enhances ChatGPT by providing it with powerful computations, precise mathematics, deep knowledge, real-time data, and visualization through Wolfram|Alpha and Wolfram Language.

wolfram.com for more features

  • The experimental Wikipedia plugin for ChatGPT searches and summarizes Wikipedia articles in response to general knowledge queries

📰 THANKS FOR READING 📰

📢 YOUR FEEDBACK IS VALUABLE 📢