New AI Model Crushes GPT-4o With Shocking Results

So it's been a crazy week in the world of AI, and if you're into this stuff, you're going to want to stick around because some big names just made big moves. We're going to talk about two major updates from Nvidia and Google that you have to know about, so let's jump right in. Alright? First up, Nvidia, the company that pretty much powers most AI systems with their GPUs, well, they just dropped an absolute bombshell. Nvidia quietly released a new AI model. And when I say quietly, I mean no big flashy event, just a stealth drop. But here's the thing. This model absolutely crushes GPT4O. The new model is called Llama 3.1 Nemotron 70B instruct. Bit of a mouthful, I know, but don't let the name fool you. This thing is a beast. It's now up on Hugging Face and it's blowing everyone away with its benchmark scores. On the arena hard test it, it scored an impressive 85.0, which is seriously high. Then on Alpaca Evil 2 LC, it hit 57.6. And for the GPT4 Turbo MT bench, it got 8.98. To give you some context, these numbers totally blow past what OpenAI's GPT4O and even Anthropic's newest model, Claude 3.5 Sonnet, have been putting up. Nvidia's not just catching up, they're taking the lead in a big way. Another key point that sets this new model apart is its focus on something called alignment. Now, in AI, alignment basically refers to how well a model's responses match the needs and preferences of the user. For businesses, this is a game changer. It means fewer errors, more useful responses, and ultimately better customer satisfaction. Nvidia has really dialed into this idea, making the model better at understanding what the user wants, rather than just generating generic or incorrect answers. In the long run, this could save companies time and money by reducing the need for constant oversight or corrections, making their operations smoother and more efficient. Now, Nvidia has traditionally been known for its hardware, right? Their GPUs are what power a lot of the AI models we talk about. But now with the release of this new AI model, they're stepping directly into the language model game. And this move is huge because it puts Nvidia in direct competition with software companies like OpenAI, which have been dominating the large language model space. Now, how did Nvidia pull this off? Well, they took Meta's open source Llama 3.1 model and gave it a makeover using some pretty advanced techniques like reinforcement learning from human feedback or RLHF for short. Essentially, they're making the AI learn from human preferences so its responses feel more natural and honestly, smarter. This model can handle complex queries without needing you to spoon feed it with extra tokens or prompts. A simple question like how many Rs are in Strawberry? Gets a detailed response. It doesn't just answer, it explains. That's next level stuff. Now what does this mean for businesses? Nvidia is making this model available on their platform called build.Nvidia.com and they're offering free hosted inference through an API that's compatible with OpenAI. So if you're running a company and thinking about using AI, you now have a serious alternative to GPT4. And it might even save you some cash, because Nvidia's offering is potentially more cost efficient. But there's more Nvidia's model is super flexible, which is a game changer for businesses. Whether you're using AI for customer service, creating complex reports, or even generating marketing content, this model adapts. And as a cherry on top, Nvidia's been pretty transparent about the fact that while the model isn't tuned for super specialized tasks like high level math or legal reasoning, it's still robust enough for most business applications. Just make sure to use it in the right context and you're good to go. However, the company did warn that the model has its limits, especially when it comes to specific domains like law or mathematics. So while it's an incredibly versatile tool, businesses will still need to implement safeguards to prevent any unintended mistakes. Alright, now let's talk about Google. Google also made some headlines recently with their AI model Imagen 3, which just rolled out worldwide. And get this, it's available for free. But hold up, there's a catch. Before we get into that, let's break down what Imagen 3 is. Imagen 3 is Google's new AI image generator, and according to Google, it's their best model yet for creating high quality images. They claim it's much better at following user instructions and generates fewer visual artifacts than its previous versions. And you know what? Based on what we've seen, they're not lying. You can ask it to generate anything from photorealistic landscapes to oil paintings, even claymation scenes. But here's the annoying part. If you want to generate images of people, you've got to pay for it. That's right. To unlock that feature, you need to subscribe to Gemini advance, which costs $19.99 per month. So yeah, while you can generate all the landscapes, animals and abstract art you want for free, human faces are behind a paywall. And just to put that into context when you subscribe, it's not just the image generator you get. You also get 2 terabytes of Google One storage and access to AI features in Google Docs. Not a bad deal if you're deep into the Google ecosystem. But still, it's a bit of a bummer that the people images aren't available on the free tier. Now I want to talk about how Imagen 3 works, because it's a little different from other AI image generators. While some platforms generate multiple variations of an image for you to choose from, and Imagen 3 gives you just one image per prompt. But here's the cool part. You can ask it to refine that image by requesting changes, and it'll remember what you asked for previously. Every time you request a change, though, it generates a new image from scratch, which is kind of neat. Oh, and by the way, you can download as many images as you like. There's no daily limit. Even on the free tier, you get a 2048 x 2048 resolution JPEG, which is pretty high quality. So what's the big deal? Well, Imagen 3 has some features that really stand out. For example, its text rendering capabilities abilities are solid. If you want to create a comic book panel with a speech bubble, the text should be readable, which isn't always the case with AI image generators. But and here's the caveat, it's not perfect yet. Google is working on it, but we've seen some wonky results, like this 1950s sci fi comic panel where the text just didn't come out quite right. Still, it's a step in the right direction. But even with its limitations, Imagen 3 deserves a spot among the best AI AI image generators out there. And for those of you who need to create images for your projects, this tool is definitely worth exploring. Just know you'll need to upgrade to generate human faces. Now, going back to Nvidia for a second, what's interesting about their new model is that it's not just about benchmarks and numbers. Nvidia's really leaning into the idea of alignment. For those of you who don't know, in AI, alignment means how well a model's outputs match the needs and preferences of its users. Nvidia is emphasizing this because for businesses, it means fewer mistakes, more accurate responses, and ultimately better customer satisfaction. If a model can understand your intent better, you're going to have fewer headaches trying to make it work for you. Nvidia's also made it clear that they're gunning for more than just the text generation game. They recently introduced their NVL M 1.0 family of multimodal models which handle both text and images. Their flagship, the NVLM D72B, boasts a whopping 72 billion parameters. So yeah, Nvidia is going all in on this. They're not just competing, they're setting the pace. What we're seeing here is a massive shift in the AI landscape. Nvidia's entry into software development challenges the dominance of companies like OpenAI and Anthropic, who've held the lead in language model development. And with Google making tools like Imagen 3 more accessible to everyday users, the AI arms race is heating up. In the coming months, everyone's going to be watching how these tools perform in real world scenarios. Nvidia's Llama 3.1 Nemetron 70B instruct is already making waves, but its success will depend on how businesses implement it. So keep your eyes on these two, because we're witnessing the future of AI unfolding right in front of us. Alright, that's it for today's video. If you found this helpful, hit that like button and subscribe. If you haven't already, we've got more AI updates coming your way soon. Thanks for watching and I'll catch you in the next one.