😳 Alibaba's EMO, NVIDIA launching RTX, Apple Car scrapped and Claude 3 by Anthropic

A lot has happened in the world of AI from “Space Karen attempting to sue AI’s wonderchild SamA” to new powerhouses like Alibaba entering the game with their “EMO” (make static images sing with raw expressive emotions shown in eyes etc) and NVIDIA releasing their new AI model, RTX; even Apple scrapped their EV ‘Apple Car’ program to focus on their AI program which should begin truly presenting with their next iOS software release/update and especially in the next release of the newest iPhone.

But last week in the intro here for AI Logs I dug into the magic BEHIND OpenAI’s Sora, which is their “large world model”, as well as the new “large action model” revealed by phenom Rabbit and their R1 handheld AI device. Well this week, Google has revealed their preliminary Large World Model offering as well, “Gemini”, a text-to-2D-game tools that effectively creates worlds from prompts, and the outputs, while currently very simple, are also super cool. Additionally, a Large Nature Model was revealed here that also holds massive potential for biomimicry, training robots, and much more.

Although I’ve been bullish on “Large X Models” here recently, LLMs keep moving forward ahead — it was just announced that Anthropic just released their Claude 3 which they claim outperforms ChatGPT 4. I haven’t fully validated this, but it’s compelling to hear, and I always welcome healthy competition, particularly is industries moving as quickly into the future as the AI industry is. The new Claude 3 has three tiers, in a fashion Mistral is also validating — and notably they claim the top tier, Opus, is effectively AGI (more on that next week once I’ve corroborated and hopefully fully tested).

Lastly, speaking of training AI, last week I touched on a text-to-code platform called Magic AI that may be close to “active reasoning” (a big step toward AGI), and also Grow, and their new LPU chips designers specifically for AI in contrast to current GPUs which aren’t. These LPUs are much cheaper and substantially faster than GPUs by orders of magnitude, which will make training new models and fine tuning them exponentially faster and more affordable.

Did a friend forward this e-mail to you?

IE₊ SUPPORT INTERESTING ENGINEERING
_{Invest In Science And Engineering}

Enjoy exclusive access to the forefront of AI content, highlighting trends and news that shape the future. Join a community passionate about AI, delve into the latest AI breakthroughs, and be informed with our AI-focused weekly premium newsletters. With IE+, AI reporting goes beyond the ordinary - and it is Ad-Free.

NEWS

• 🌌 Debris killer: AI-powered solution to track tiny floating space junk

  There are millions of pieces of space junk flying in Low Earth Orbit, says NASA.
  

• 👩🏽‍💻 Apple unveils new M3 chip-powered MacBook Airs boasting AI superpowers

  Available in 13- and 15-inch sizes, the MacBook Air has a sleek and lightweight design, with up to 18 hours of battery life.
  

• 👂🏽 New AI-powered phone app detects ear infections with 93% accuracy

  This could help decrease unnecessary antibiotic use in young children.

MUST READ

Claude 3: Anthropic unveils its most powerful ChatGPT-challenger chatbot

Anthropic, backed by tech giants Google and Amazon, introduced its latest series of artificial intelligence (AI) models – ‘Claude 3.’ This unveiling represents another move in the ongoing rivalry among Silicon Valley companies, each vying to offer increasingly potent AI solutions.

Claude 3 entails three refined models: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. Their computational prowess escalates from Haiku to Opus. Standing against competitors like GPT-4, GPT-3.5, Gemini Ultra, and Pro, Anthropic’s flagship model, Claude 3 Opus, claims superiority in various performance metrics.

OTHER IMPORTANT UPDATES

• 🪱 Morris II: AI worms could compromise ChatGPT, Gemini, scientists show

  Researchers have created a worm that could potentially steal data and deploy malware from generative AI-powered applications.
  

• ☢️ Scientists use AI to stabilize nuclear fusion for the first time
  An AI controller promises to predict what can go wrong in a nuclear fusion reaction 300 milliseconds before it actually happens.

PROMPT OF THE WEEK

After studying a number of test results as well as conducting my own, here are some cool concepts and myths disproven that can help you step up your promoting game:

The first is that, after extensive testing, it evidently does NOT yield superior results when you’re overly “nice” or cordial to your AI; in other words, no need to even say “please” and “thank you”, to praise it etc., and in some tests doing so actually makes it perform slightly worse.

Secondly, after many tests, it does appear that telling AI in the prompts they the better it does, the more money you will tip it or donate to charity in its name does actually get it to perform better. As weird as this is, to augment Shakira’s lyrics, “the data don’t lie”.

Thirdly, something that’s been highly effective for me is asking in my prompts for the AI to onl the ask me one question at a time, rather than asking a laundry list of questions. This way I can just answer them one at a time, and we get to the results much quicker.

Speaking of getting to results much quicker, the last two prompt tips for this week are to not accept the first results from your prompts, as it’s an iterative process, as well as asking the AI within your prompt for 5 things it needs to know before delivering its response in the optimal fashion; this allows it to help you help it.

I hope these are helpful on your promoting journey. Ultimately, it’s always productive to ask your AI how it can best help you and let it as the expert help guide you, the user. Consider your AI to be an extension of yourself and your team.

AI PICTURE OF THE WEEK

The most notable piece of AI-generated art to date is "Portrait of Edmond de Belamy," which was sold at Christie's auction house for a staggering $432,000. This artwork was created by Obvious Art, a Paris-based collective of artists, using a Generative Adversarial Network (GAN). GANs involve two neural networks—the Generator and the Discriminator — competing against each other to create new, original images based on a dataset.

TUTORIAL

Back in issue 7 several months ago (we’re currently in issue 21) I showcased how to use Suno to create amazing AI music, but it’s been a while, they’ve got even better, and I truly believe that generating music using generative AI right now is one of the easiest, most compelling, and most fun ways to use and get value from AI. He’s it for jingles, original music for videos and social media, creating a fun song for a loved one, or even making new music that’s exactly what you’d want to hear to enjoy personally.

So Suno is still available for FREE using their V2 which is incredible, but they’ve just rolled out V3, which is consequently better, although the V2 is still amazing. Notably, you can create instrumentals in V3, it can create longer songs, the vocals and range of genre seem even more tuned in, and the pricing starts as low as $10/mo ($8/mo if paid annually). I believe it’s the most powerful, versatile, easy-to-use music generation tool out there. To upgrade, go to the hamburger menu, select “subscription”, sign up for Pro or Premium, and then when you go to “create” to create a song, toggle to V3.

Regardless of what version you’re using, you’ll then either tell it the theme and things you want to include and have it generate the lyrics and song for you, or “change to custom mode” and input your own lyrics and tell it what genre to emulate. It’s able to create bilingual songs, remix songs, and create. Once you’ve inputted what it needs simply hit the “Create 🎵” button, and within a minute or so (less if you’re Pro/Premium) you’ll have an amazing song (two versions to compare/contrast). If you want to see the lyrics, just toggle “show lyrics”.

Now that we can make our Pika and soon Alibaba EMO images sing, a bonus tutorial is to create a short song on Suno, a video using yourself or your favorite character inside of Pika, and add your original song to it and have it sing. Please share your best songs and videos here with us!