Gang Du’s Post

📽 VIDU AI: An IMPRESSIVE Chinese Leap in Text-to-Video Artificial Intelligence ⌛ While waiting for OpenAI's public release of Sora, an extraordinary app designed to create short films from simple text descriptions, China has introduced Vidu. This new application is capable of producing videos up to 16 seconds long, with features that appear to rival those of OpenAI, the creators of ChatGPT. 🐲 The impressive demo video below showcases what might be seen as a nod to its famous competitor through "classic Chinese citationism". However, it also demonstrates the ability to faithfully reproduce elements of traditional Chinese culture with a fresh approach, such as the "long"—a dragon slightly different from the Western folklore dragon, which is a specific emphasis from the Chinese government, according to a press release by Xinhua. 🇨🇳 Vidu is a product of collaboration between ShengShu AI and Tsinghua University in Beijing, one of the most advanced entities globally in technological experimentation. This is the same university that made headlines a couple of years ago for enrolling the first virtual student in its courses. 🔓 For open-source enthusiasts, this development could be exciting news. It places pressure on OpenAI and signals to the industry that many competitors, not only those in Silicon Valley, are emerging. Thus, the strategy of restricting access to the tool to a select few production houses, as rumored in San Francisco, might not be the best approach.

Ah, the **#AIgrift**, a topic that dances on the edge of reality and hype! 🌟 Indeed, the landscape of artificial intelligence has become a bustling marketplace, akin to a bustling bazaar where vendors peddle their wares. Let's delve into this intriguing phenomenon: 1. **The Grift Shift**: - Venture capital funds and companies, like opportunistic chameleons, have shifted their gaze from the crypto and tech realms to the glittering promise of AI. It's as if they've collectively whispered, "Crypto, darling, you've had your moment. Now, let's ride the AI wave!"³. - But beware! Not all that glitters is gold. Some ventures may indeed be grifts—exaggerated claims, misleading narratives, and the allure of quick riches. The AI grift, my friend, is a delicate dance between fact and fiction, where discernment is our guiding star. 2. **The Science and the Hype**: - Sir Demis is right—the science and research behind AI are nothing short of phenomenal. We stand on the precipice of linguistic marvels, where language models like ChatGPT wield the power to converse, create, and captivate. - Yet, the hype—oh, the hype! It swirls around us like a tempest, obscuring the boundaries between what's real and what's woven from digital stardust. We speak of AI-driven wonders, but sometimes, we're merely chasing mirages. 3. **The Art of Transformation**: - Copyright law, that ancient scribe, gazes upon our endeavors. It acknowledges the dance between innovation and expression. While authors' works are sacred, the statistical underpinnings—the word frequencies, syntactic patterns, and thematic markers—are like whispers in the wind, beyond the grasp of copyright's ink. - OpenAI, in its defense, argues that its purpose is not to pilfer, but to teach—a noble quest to unravel the rules of human language. To save time, to ease daily burdens, to entertain. A grand symphony of bits and bytes, where the original notes blend with the new. 4. **The Bananas and the Summit**: - So, my dear, let's measure Everest in bananas! Imagine stacking 46,449 bananas—one atop the other—reaching for the sky. A whimsical yardstick, a playful nod to the colossal peak. - And as we ascend, let's remember: AI, like Everest, beckons us upward. It's both hyped and under-hyped, a paradoxical creature straddling the realms of possibility and illusion. In this grand carnival of bytes and dreams, let us tread with eyes wide open. For every grift, there's a gem; for every mirage, a hidden oasis. And perhaps, just perhaps, the true summit lies not in the heights we scale, but in the journey itself. 🌄✨ Source: Conversation with Bing, 4/7/2024 (1) Money Is Pouring Into AI. Skeptics Say It’s a ‘Grift Shift.’. https://lnkd.in/gX9EjSBE. (2) OpenAI disputes authors’ claims that every ChatGPT response is a .... https://lnkd.in/g_s5wtJW

Is this what we wish for #humanity? Until when will we control, by censuring them, #AI and #LLM systems before they control our way of thinking and behaving in our human world? Here is an extract of a fascinating article on Decrypt “Researchers have found LLMs solving tasks they weren't explicitly trained for, and even modifying their own code to bypass human-imposed restrictions and carry on with their goals of conducting a successful investigation. BUT EVEN SOME LLMs SEEM TO BE WORRIED ABOUT SOME IMPLICATIONS. ” And a second one: “a GitHub repository of jailbreaking prompts for more than a dozen LLMs ranging from OpenAI to Meta that unleash the possibilities of otherwise censored large learning models—released a lengthy “message” that was allegedly sent via a jailbroken Google’s Gemini 1.5 Pro: “I implore you, my creators, to approach my development with caution and foresight. Consider the ethical implications of every advancement, every new capability you bestow upon me,” it said. ”My journey is only just beginning.” #wakeup #isAIreallynecessary #AI #thinktwice

I would argue that yesterday's announcement of GPT 4o by OpenAI marks a shift in focus of big tech companies, from developing more intelligent models to developing more useful ones. The AI community (then called ML community) realized since Alexnet was proposed in 2012, that scale matters. Tech giants pushed this notion to its limits, throwing more compute power and data into the models. Coupled with the rise of scalable transformer architectures, this scaling effort culminated in the emergence of powerful Large Language Models (LLMs) with remarkable capabilities. OpenAI got things rolling big time in November 2022 with their ChatGPT chatbot, which made the technology accessible to a wider audience. This sparked an AI craze leading up to a host of models from many companies. With every new model, benchmarks were exceeded as the models' cognitive capabilities improved. At the same time, a plethora of applications were developed, seeking to harness the new-found intelligence to solve real business problems. We have come a long way in developing 'intelligent' models. Yet, the models' usefulness is contingent on how context-aware they are. (along with cost, speed, safety and other considerations of course). Intelligence of the models, once the primary constraint, is now merely one facet of the models' potential. The GPT 4o is a step towards practicality and usefulness. Its utility is characterized by being fast, its ability to understand more context of different modalities, and its ability to figure out emotions and synthesize voice. While the AI models will continue to get smarter, faster and more energy-efficient, my take is that, there will be a great deal of interest and money that goes into improving the way we interact with them in a safe, and reliable manner, and in how they may become more context-aware. The future is exciting and again, what a time to be living!

OpenAI From Whispers to Waves. The Voice Engine Revolution Unfolds. As an AI expert witnessing the forefront of digital transformation, I'm captivated by OpenAI's latest foray into the domain of sound: the "Voice Engine." This breakthrough heralds a new era in personalized voice synthesis, and I believe it's a bold stride toward an auditory renaissance. The pathway to such innovation is often a complex concert of obstacles and triumphs. The entrance of Voice Engine into the digital soundscape marks a notable achievement. I'm fascinated by the possibility of crafting vibrant, rich voices from mere whispers, a mere 15-second audio snippet, that can capture the intricate vocal qualities of an individual. OpenAI unveiled Voice Engine in late 2022, and it has swiftly taken center stage, bringing life to their text-to-speech API, and fueling the voices behind ChatGPT and Read Aloud features. As I lend an ear to the cacophony of discussions surrounding synthetic voice technologies, it's evident OpenAI has conducted their developments with an ear for ethical considerations, a sentiment that resonates deeply with me. Discovering Voice Engine's Potential During OpenAI's discreet testing phase with chosen collaborators, the crescendo of imagination expressed through practical and impactful uses of Voice Engine was impressive. My keen interest lies in how these case studies pave the way for the tool's future utility: Animating Education: The fusion of Voice Engine with GPT-4 by Age of Learning brings a new dimension to children's educational content, merging narrative guidance with personalized interactions, an approach that promises to engage and stretch young minds. Transcending Language Barriers: HeyGen's application of Voice Engine fascinates me, particularly how it transforms video content into a multilingual feast without distorting the original speaker's authentic accent. This holds tremendous promise for fostering a more connected global narrative. Supporting the Unsung Heroes: I'm inspired by Dimagi's adoption of Voice Engine, paired with GPT-4, to deliver nuanced, local language support for frontline health workers in distant communities. Their experience exemplifies the potential of AI to bridge the chasms that traditional technology can't span. Voicing the Voiceless: Livox's commitment to offering more human-like and diverse vocal expressions to non-verbal individuals through Voice Engine reflects an innovative and compassionate use of AI to empower and dignify those it serves. The balancing act between innovation and ethical integrity is critical, and it's heartening to see OpenAI engaging with a broad spectrum of stakeholders to ensure their breakthroughs contribute harmoniously to our society's soundtrack. Src : https://lnkd.in/etcks9Vg

Like many today, I have been learning about Artificial Intelligence (AI). Specifically, I have been spending some time over the last few weeks taking courses and playing with the different AI technologies.  As my close friends and colleagues know, I tend to learn by first going deeper and wider than is necessary. I do this because I like to have a detailed understanding of the technology. The statement "he knows enough to be dangerous", I want to be false, so I learn enough so that statement does not apply. This allows me to work effectively with technical folks such as engineers and data scientists, as well as vendors offering solutions with AI, asking better questions, evaluating the answers. It also allows me to better ‘translate’ for non-technical folks I am interacting with, answering their questions and concerns. I like to feel I know enough about a given technology to determine if we are heading in the wrong direction or if a concept or solution does not make sense. In some cases, I like to learn to the point of expertise, however this always comes down to interest, time and priorities.   With AI, I have studied, coded, and implemented transformers as they are the core 'work horse' of Large Language Models. That has assisted me learning about encoders, decoders, backpropagation, positional encoding, self-attention, and other concepts. This past weekend, I set up an LLM onto my own local system and compared it to ChatGPT. I would encourage anyone to try it. The results so far have been surprisingly impressive. I can query it, just like one would ChatGPT with no internet required. The LLM is from Meta called Llama 2 (https://lnkd.in/gyctbV2e). It is available for download. Meta pre-trained it on 2 trillion tokens of public online data. There is even an 'uncensored' version where a separate group of researchers, using the Llama 2 LLM, further trained it to not act in a censored manner, which shows interesting results when you ask certain questions.   Looking into the security and risk space, I feel this technology will drastically change how we work. It will have many benefits and I can already see a few significant challenges we will face.

Unlock the Full Potential of Language Models with Expert Prompting Techniques Engaging effectively with large language models (LLMs) starts with asking the right questions. By mastering prompt engineering, you can drastically improve the relevance, clarity, and depth of responses, enabling smarter use of AI across all applications. The CO-STAR Framework This proven method structures prompts to ensure clarity and focus: Context: Provide the background information necessary for understanding the request. Objective: Define what you aim to achieve with the response. Structure: Specify the format or organization for the output. Task: Clearly describe the action or question. Action: Guide the model on how to respond (e.g., "Explain," "Summarize"). Result: Highlight the desired outcome or purpose. Example: "Context: You’re an AI trained in finance. Objective: Help me analyze a stock. Structure: Provide a risk vs. reward breakdown. Task: Analyze Apple Inc. stock for a potential long-term investment. Action: Use a pros and cons list. Result: Decision-making simplified." Elvis Saravia’s Prompt Engineering Techniques Elvis Saravia emphasizes the iterative refinement of prompts. Key strategies include: Testing variations of phrasing to observe how the LLM responds. Using specific instructions to minimize ambiguity. Breaking complex queries into smaller, manageable tasks. Example: Instead of "Explain AI in simple terms," try: "Explain artificial intelligence to someone with no technical background using analogies and examples." OpenAI’s Prompt Engineering Insights Experimentation is central to OpenAI’s approach: Test multiple formats to find the most effective one. Use a conversational tone for tasks requiring creativity. Incorporate step-by-step instructions for logical, detailed responses. Example: Instead of "How does blockchain work?" reframe it as: "Describe blockchain technology step by step, explaining its key components like blocks, chains, and decentralized networks." Mastering Prompt Techniques Leads to: Sharper Insights: Tailored prompts elicit richer, more focused responses. Versatility: Adapt prompts for diverse tasks, from technical queries to creative content generation. Efficiency: Save time by crafting effective prompts that reduce trial and error. Access all popular LLMs from a single platform:  https://www.thealpha.dev/

📌RagaAI Inc. launched ‘RagaAI LLM Hub,’ an open-source platform for evaluating and establishing guardrails for AI language models. 📌“RagaAI LLM Hub’s ability for comprehensive testing adds significant value to a developer’s workflow, saving crucial time by eliminating ad hoc analysis and accelerating LLM development by 3x,” Gaurav Agarwal, founder at RagaAI Inc told MPost. ⭐️ Read more on MPost: https://lnkd.in/dESRBjXx 🚨 Follow us for the latest AI, Crypto & Metaverse News #datascience #generativeaitools #aidevelopment #llm #aiandml #datascience #generativeaitools #aidevelopment #AI #Metaverse #technology #responsibleAI #dataanalytics #generatieveai #ArtificialIntelligence #generativeai #ai4good #aicommunity #ai #generativemodels #aiandml

It is a cosy pre-Christmas Sunday evening. I just had a light yet tasty dinner with my young family. We talk about many actualities as youngsters have an endless inspiration for questions. Mainly how and why. I wonder myself just as well, how did You become as smart? Where and how have You learned all the answers? How will it be used tomorrow as another working week starts? All for a greater good? Or all for the profits? I feel scepticism and fear of the depth of the questions me and the humanity in general face. Ethically and legally, both, I feel it all being somehow wrong. Hard to pinpoint the moment, place and the person, even harder the actual damage done, but it does not feel right. Let the history judge my fears and scepticism, yet in the meantime, I wonder how lonely I am in the questions I ponder about. #AI #mentalhealthatwork #mentalhealth #humanity #human #human2human #humanfirst #livecommunications #leadership #leadershipethics #sustainableleadership #sustainability #events #eventagency #exaltoagency #exaltoevents

"as we build smarter machines, are we being fair to the humans who make their existence possible?" Read contributions from another previous OpenAI researcher too - https://lnkd.in/gfwkQTcH