Manish Patel’s Post

Wow - this video is terrible. Clearly a major blind spot for AI. As a former USA Gymnastics gymnast, I'm not surprised. How do you accurately train an algorithm to learn what takes a gymnast 6 hours a day, 6 days a week, for more than a decade to learn and perfect. Plus with four different events, equipment changes, and scoring changes, it's one sport that continues to evolve. Plus not enough video data on gymnastics - there should probably be much more. Gymnasts today are also doing new moves and tricks that have never been done before. The point about AI not actually understanding the human body and its motion seems spot on. It's important when adopting a new technology to recognize what it's really strong at, but also be well aware of its weaknesses and adapt accordingly. If you want to see gymnasts doing actual incredible moves, check out our reel on the 2024 US Women's Gymnastics Olympic Team (in the comments). #usagymnastics #youthathletes #ai #gymnastforlife #2024olympics #technologyforgood #healthyhabitsforlife #learnbydoing #springspot

It's just a poor rendering with poor modeling. But the Chopin music is good. Yes. AI can do many thing, but NOT every thing, at least NOT making love nor what we are doing in data analytics. The key element is metadata. Metadata is an enabler. It's a digital content, NOT a technology. It's like a treasure map, curated by human for treasure hunting. Without metadata, NO data can be found/retrieved, even by the most advanced technologies, like AI, GPU, quantum computers, etc. https://lnkd.in/g-aJFnXR Do you or any contacts of yours need our expertise/IP, a copyrighted multilingual metadata, that can do what high technology, like AI, high-end chips, supercomputers, etc., can't do NOW? Thanks.

I am incredibly curious and excited to see just how far gen-AI can take us. Which of current limits will be overcome, and just how far can we push it. How many brilliant new techniques to enhance it will come in the next couple of years? The video showcases the limitations and risks of the technology in a great way.

If you rely on any form of GenAI or think it will takeover expert jobs. Remember you'll get such results. The more multimodal these models get, the less accurate (decreased expertise) they will be...

This an excellent example highlighting how generative AI outputs can seem accurate but be misleading. The gymnastics video looks correct but is far from realistic, similar to how GenAI applications like ChatGPT can provide convincing answers that are actually off. We often don't see these discrepancies in non-visual tasks, emphasizing the need for careful verification. Remember, AI predicts, it doesn't truly understand.

This is the same for all #Artificial_Intelligence (#AI) tools! I have been evaluating some AI tools for a long time. I agree with the 'Garbage in, garbage out' concept, but AI systems just regurgitate training data. It is obvious in some cases (like the video below), but for more complex and analytical cases, you should have the expertise to assess the limitations of AI. We must be cautious when using AI and should consider it as an assistant rather than relying on it to perform tasks independently. Especially in fields where we lack expertise, AI should not be trusted, and critical tasks should not be entrusted to it. What bothers me the most is that since AI became available to the public, most of the posts I see are strikingly similar. Brains are getting lazier day by day, turning more towards artificial intelligence for help. Those lacking enough expertise in a field produce texts using AI and then post them under their names. In social networks, I want to see educational posts written by humans, not by artificial intelligence. When you search for something on the internet, you may encounter junk websites filled with AI-generated texts. I am concerned about the future of scientific research and publication. Today, people use artificial intelligence to produce articles and flood prestigious journals with their junk articles, causing important research conducted by humans to be lost amidst a sea of low-quality submissions. If peer reviewers use AI for conducting reviews, this will be disastrous.

Its truly mindblowing how critics and haters are mouthwatering at the opportunity to bash Ai. its also mindblowing how massive their lack of understanding and interpretation is. Has anybody ever watched a child draw? was it surrreal? Did it lack a sense Physics? Has anyone ever had a Dream? Was it surreal? Did it lack a sense of Physics? Science has basically tried to imitate the human brain. Mechanics we do not fully understand. This resulted in incredible results. But these technological minds lack a physical body. Lack a complete set of senses. They are basically the equivalent of a child in terms of empirical experience, but with the knowledge of our entire species. What we see is the imagination of a child with such a massive wealth of knowledge. AI lacks understanding, not because it's flawed, but because it simply hasn't had the experience yet. The mere fact that we got from Pong in 1972 to a nascent technological mind that has surrealistic interpretation imaginations in 2024 is a scientific accomplishment beyond words. Just imagine what will happen with the next generation of fully developed robots running these Ai.

There are two main things to consider here. Was the model trained with enough data? In this case I would think you need fine tuning on the model and maybe a number of different fine tuning to represent different complex movements.

With not enough data, LLMs make things up. Have this video in mind next time you ask a question to an LLM model. On visual answers it is easy to spot the BS, with textual answers not always. It is convenient to go on copilot mode, easy to ask chatGPT. But is not always reliable. If the question is important to you, validate the answer! Do your research.

It's just a poor rendering with poor modeling. But the Chopin music is good. Yes. AI can do many thing, but NOT every thing, at least NOT making love nor what we are doing in data analytics. The key element is metadata. Metadata is an enabler. It's a digital content, NOT a technology. It's like a treasure map, curated by human for treasure hunting. Without metadata, NO data can be found/retrieved, even by the most advanced technologies, like AI, GPU, quantum computers, etc. https://lnkd.in/g-aJFnXR Do you or any contacts of yours need our expertise/IP, a copyrighted multilingual metadata, that can do what high technology, like AI, high-end chips, supercomputers, etc., can't do NOW? Thanks.