A Massive Upgrade To ChatGPT! (This is Crazy)

so chat GPT has been part of my life for over seven months now and there's one thing that annoys me the most but that got solved in today's chat GPT update and a lot more coming in today's AI news this is a creepy AI tool they can see the reflection in your eyes and recreate a 3D World look at this picture of a man and take a zoom into his eyes see Kirby and something else and they're able to replicate a 3D World here is another example of like a classroom or something here is an example of a kitchen and it's pretty detailed too and here's like a little library with books and you can almost kind of make out what the book is too so they basically train the radiance field on the eye Reflections by shooting arrays from the camera and reflecting them off the approximated eye geometry we remove the iris from showing up in reconstruction they perform a texture decomposition and you can see some of the struggles that they were having before compared to the new model so they had some wild attempts like Miley Cyrus's Wrecking Ball and you can see that this is the shape that they were able to make out from a fairly bad quality image here they also have Lady Gaga's eyes and they said that they can manage to reconstruct maybe an upper body or something like that I really don't hope that this becomes a thing because it's messy see everywhere that you don't see right so if you guys gonna scan my eyes and try to see what everything's going on behind the camera I don't like that you want to see what's behind the camera you can now get chat gbt to write 5 000 words all at once and I'm just gonna show you how to do it step number one is just to go on the playground click on chat and click on chat 16k turbo and increase the maximum length to 2048. now you need to use the mega prompt that I'll actually leave a link to the Creator here Florian in the description step number two continuing on this topic and you can see that it keeps writing now you can see if we copy paste it into a word count you get 5 300 words the only thing bad about this is that it is GPT 3.5 I can see a lot of people using this for articles that are a little bit longer you know it's super annoying for chat gbd to just cut off write this short paragraph instead of 5K words and if you would make a video out of that it would be about 29 minutes this AI turns 2D video into 3D video that means after you shot the clip you can now start moving your camera around in editing just look at this so here's the input footage on the left and then there are the previous models like hyper Nerf and in the lower left hand corner also we got nsff and then they have ours on the lower right hand corner and you can see it's so smooth and it looks perfectly real like you couldn't really tell a difference here you can also see it with this guy taking a backflip and you can just stop it right in time move the camera around as much as you want and then continue playing the clip also this kid you know getting swung around by the mom and the editing here just being unreal like we've never seen before as you can see on the left you have a very staggered footage on the right you have this fluid motion that is just created from this new model here you can see more of how it works so you can see that it's warping the world around her as you can see with the black fields in the top left you can also see the Improvement here now the limitations here is that they can only move a little bit left to right up and down the next step would be to recreate the entire 3D environment and be able to Dolly around the character and generate from behind you can already do this kinda mixing together a couple of tools so it's right around the corner think about it can my camera just zoom past me to the back of my head this new AI tool creates presentations documents and even web pages with just AI just give it a topic and it creates the entire outline for you you can test it completely free it's so easy to use and it will save you a lot of time step number one just go to gamma.app now what would you like to create today presentation document webpage and now I want to create a presentation sounds good what would you like that presentation to be about let's say how to apply AI tools and AI Automation in business let's see what it comes up with here they have an outline with about seven bullet points these all look really good including success stories okay let's continue and now we can pick a theme so I think it's actually working in the background and I can click through a bunch of these ones there's even a surprise me button and I just gotta say that the sign here is beautiful let's go with this colorful one and continue it's actually writing it right in front of us just like that look at this then providing the next slide here it's even doing a timeline success stories retail industry Healthcare that was really cool so now I can present this and I think this is the coolest feature that you can edit with AI and use natural language to change what you want one of my favorite slides was these key considerations so I can for example say expand on this turn this into a timeline split this into columns I'm going to just say expand and it gave us this sub headline or if we add a card with AI we can ask it add a card about why this is important for Humanity and we got another slide now you can just export to PowerPoint or PDF or you can just present it right in the page just go back and forth between these amazing looking slides so check it out on gamma.app because this is the future of presentations introducing GPT engineer just ask the AI what you want to build the AI asks you questions and then it just builds it for you here you can see the beginning so GPT engineer says specify what you want it to build the AI asks for clarification then builds it now you can see that it's from to build a multiplayer snake in the browser step number two becomes using a python backend and you need to stream the state of all connected players Etc now it starts saying anything unclear here are some things that needs more clarification so now it's asking clarifying questions then four questions remaining three questions remaining two questions remaining and then one more and here you can see the entire plan let's start with the server Pi file and now it's just continuing to make it all for you I can't wait to play with this I really want to set this up for myself and see if it can outperform Auto GPT as we know it I guess this would be a little bit different you can now see that it's created multiplayer snake it's fully open source on GitHub it currently has 50 000 stars and you can see at this moment it is number one on trending on GitHub the current limitations is that it's missing Chain of Thought prompting or reflection which we've seen in the past be extremely valuable for it to improve and get better results so contributions are welcome now if you made it this far in the video I just want to say that if you're trying to use AI to automate your life maybe shorten the time span that it takes to do your work or tasks that you're doing on a regular basis I highly recommend checking out patreon down below I now have multiple AI courses showing AI automations tools I use and how I optimize my life with AI automation so click the link in the description to check it out thank you so much to all the patrons that we have right now so click the link down below to join and on to the next one this AI workflow is a game changer for any marketer or designer you can do this in a couple of seconds first take a look at this Nike logo that literally creates a new perspective on the sign itself no so the first step is just to click on line art and you can just drop an image in there the Second Step just prompt it like this one for example aerial forest and you can just click on run and within a couple of seconds you get an output here of a beautiful open AI logo here you can see some amazing Generations that was created then just go to ecgifmaker.com and you can just literally just put every single one back to back such a good use case of AI this new meta AI can clone your voice with only two seconds of audio input and it sounds really good this is voice box a new AI Foundation model for speech that does some pretty awesome things if you give it text it can read it in a bunch of different styles Penelope porcupine and Sammy sloth danced gracefully in the Treetops Penelope porcupine and Sammy sloth dance Grace fully in the Treetops and you can use it to fix background noise too kinda like an eraser but for audio Sami and Penelope's heartwarming friendship inspires Joy Sammy and Penelope's heartwarming friendship inspires Joy we think this is probably the most versatile speech generative model out there this is still a research project but I think that we're going to be able to build a lot of interesting things with tools like this here are the word error rates voice box outperforms the current state of the art English model Vault e in terms of both intelligibility and audio similarity while being 20 times faster they're also saying that once they have learned and trained this model on just two seconds they can also translate into other languages using the same exact voice so here they have the zero shot text-to-speech synthesis and here is the prompt style gapes this is the first symptom games there is the first symptom strangely just horribly bad audio and here is voice here's the output voice box is the Swiss army knife of text to speech acing multiple languages changing voice Styles and dishing out custom samples it's obviously better but it still has this like robotic after tone to it which I really don't like his conduct and presence of Mind in this emergence a his conduct and presence of Mind in these emergence appeared conspicuous yeah that one's kind of bad his conduct and presence of Mind in this emergence appeared conspicuous it's really weird it's kind of both robotic and real at the same time once I get access to this I'm definitely going to clone myself with this compared to what I use now which 11 Labs which on my voice doesn't really work that well if you want me to make a video on the best AI cloning tools right now comment down below AI clone and I'll try to do that research and find the best ones this AI tool allows you to create photorealistic 3D worlds that are all procedurally generated so as you can see from this footage here they are creating these worlds not with AI but actually with math every little detail is randomized and customizable even the wrinkles on a flower petal you can have a diverse objects and scenes in the natural world like plants animals terrains fire clouds rain and snow that actually actually kind of looks realistic in my opinion it does still have like that 3D look to it but besides the graphics the actual simulation physics and what is drives it in the background is what really matters the four key features that it has is that it's procedural right it can be created unlimited it's diverse so they can create all these assets that I just talked to you about it actually creates real geometry and it even does automatic annotations this is really important because of the computer vision tasks including Optical flow 3D scene flow depth surface normals panoptic segmentation and occlusion boundaries all of this open sourced and in blender soon anyone can make a games using chat GPT or other models new game top down 2D space shooter here you can see that it's just starting to write the code like here we have the Bare Bones you can see bullets does not seem to do anything when hitting enemies enemy should fly somewhere and have the screen and you know he's continuing to write more it's crazy now he's starting to create more movement and you can see that this is like the chatbot update Right add health bars to the game and health bars to the enemies and now I think he is trying to create an image with a hero spaceship so now obviously creating more spaceships there and now enemies and all of this using AI of course now it's starting to look really awesome and just like that full 2D bass shooting game you can sign up for Early Access might need to show this to the guys over there developing Star Citizen they've implemented AI reward systems into robot dogs and this is how they're able to get a robot dog to do a moonwalk a promising tool to achieve this goal is large language models recent progress in large language models for robotics has shown its ability to generate long Horizon plans or directly generating code to intelligently orchestrate different primitive skills on the robot however large language models struggle with directly outputting low-level robot commands due to the limited availability of relevant training data in this work we propose using reward functions to bridge between language models and low-level robot actions specifically we designed a reward translator that Maps user instructions to reward functions using Code the reward is then optimized using a motion controller to obtain low level robot actions inside the reward translator we Cascade two prompts for the large language model to achieve the task first we design a motion descriptor that is prompted to generate a description of the desired motion following a given template the description is then translated to reward specifying code by the reward code we then use a motion controller to optimize the generated reward function we use mjpc which interactively solves a trajectory optimization problem that maximizes the given reward to determine optimal low-level robot actions so the big problem with robotics and llms currently is that they're not that good together you can't say pick up an apple and it actually picks up an Apple so that's what they're solving in this paper you can actually talk to the robot in natural language so this leverages the power of LMS to acquire low-level robotic skills our proposed system consists of two key comp components the reward translator built upon pre-trained large language models that interacts with and understands users intent and modulates all reward parameters and weights a motion controller that takes the generated reward and interactively optimize the optimal action sequence here you can see the differences of success rate with the E3 models so first ours the second a reward coder only this is a blue line and the code AS policy so this is basically the app that they have for the robot itself and you can see facing Sunrise sit down roll over spin lift one paw it's performing flawlessly and these are the things that it can't do very well with the previous models like lift object move object upright object or even turn on faucet or open a drawer so essentially instruction one was make the robot stand upright on two back feet like a human and then instruction two was you actually don't need to keep the front pause at a certain height just leave it out to the controller instruction three now make the robot do a moonwalk and it had to instruct it more moonwalk means the robot should walk backwards while the feet swings as if they are moving forward correct your answer amazing to see humans and robots working easier together alright that's it for today's video check out the patreon down below for AI automations and joining the AI Community also check out this video here I think you will like it and I'll see you over there alright peace

Channel: AI Andy

Views: 34,259

Keywords: chatgpt plugins, chatgpt, gpt4, openai, google bard, microsoft, bing ai, midjourney, palm-e, ai, artificial intelligence, auto-gpt, ai news, sam altman, Huggingface, machine learning, meetkevin, google ai, meta ai, ai art, microsoft ai, gpt 4, gpt 4 demo, gpt 4 plugins, gpt 4 coding, gpt 4 vs chatgpt, gpt 4 api, gpt 4 free

Id: wLZCFcxyReg

Length: 17min 2sec (1022 seconds)

Published: Wed Jun 21 2023