😺 Amazon enters the AI war

PLUS: OpenAI hype is getting a little too hot...

Welcome, humans.

The AI hype train is full steam ahead today…for some reason, OpenAI employees are releasing cryptic posts like “OpenAI is unbelievably back” and “iykyk.”

Full disclosure: OpenAI’s rivals are closing in, so this could be all hype, no gas. That said, we do know what OpenAI does whenever it feels threatened (drop a new model).

One thing’s for sure, there’s a lot of AI launches on the horizon as we approach the end of 2024. A potential Gemini-2, a Grok-3, a full o1 drop, and maybe even SORA (after last week’s PR debacle, looks like the SORA testing is up and running again).

The thing is: do we even need SORA? I mean, check out this genAI Batman movie a team of creators just made with Midjourney to Kling / MiniMax / Runway (probably alternating between them), with Eleven Labs for voices and Topaz for upscaling:

As the comments say, “This so just unbelievably real”…

Here’s what you need to know about AI today:

Amazon launched Nova, a family of 5 AI new AI models.
Tencent launched HunyuanVideo, a new SOTA text-to-video AI model.
HuggingFace CEO warned of China's growing AI influence.
Google expanded its rollout of new video generator Veo.

Advertise in The Neuron here.

Watch out, AI giants—Amazon is coming for your market-cap with new chips and AI models.

AWS announced a flurry of AI-related news at its RE:Invent 2024 event. We stayed up late to watch the Monday Night Live (not to be confused with SNL, there were no sketches), where the company unveiled “Project Rainier”, an “ultra-cluster” of 100K+ Trainium 2 chips that Anthropic will use to train the next generation of Claude.

As it turns out, Apple’s going to use them, too.

But that was just the opening act. On Tuesday, AWS announced a slew of additional AI announcements. The most important? It’s launching its own model called Nova to rival ChatGPT and Claude.

Here’s what is going on: Amazon wants to compete—with NVIDIA in the chips department, and with Microsoft (who invests in ChatGPT) and Google (who owns Gemini) in the model department.

“Today, there’s really only one choice on the GPU side, and it’s just NVIDIA…we think that customers would appreciate having multiple choices.”

Matt Garman, AWS CEO

There’s practical reasons for building its own chips (besides NVIDIA’s $3T market cap)—with Trainium, AWS can own its own supply chain …AND not have to rely on NVIDIA (here’s a great read from the WSJ all about that).

But really, today’s news is all about Nova: AWS is launching 5 models out of the gate, with more coming in 2025. Here's the lineup:

Nova Micro is the speedster—text-only, lowest latency, very low cost with a 128k token window.
Nova Lite and Pro are the multimodal heavyweights, both handling text, images and video with 300k token windows.
1. They can process 30-minute videos and 15,000+ lines of code in one shot.
2. And in early 2025, AWS plans to expand these windows to over 2M tokens.
Nova Canvas handles image generation (already integrated into Shutterstock), while Nova Reel makes 6-second videos in about 3 minutes.
1. With Reel, you can control camera movements (360-degree rotations, zooms) just by describing what you want.
2. Oh, and a 2-minute version is coming soon, too.

Now as for benchmarks, here’s a great breakdown.

…But on LiveBench, it’s less impressive. For the full specs, read this.

The fine print: Nova works in 200+ languages (only optimized for 15), requires specific access through AWS Bedrock, includes built-in watermarking and moderation, and comes with indemnification against copyright issues. Yaay, commercial applications!

To try them yourself, check it out here (you’ll need an AWS account), request access to Nova models, and make sure you're in the US East (N. Virginia) Region. You can upload documents up to 4.5MB, images up to 20MB, and videos up to 25MB (or 1GB via S3). Pick your model, select apply, enter your prompt, and click Run.

One more thing: Amazon dropped a new tool called Automated Reasoning Checks to automatically combat hallucinations.

FROM OUR PARTNERS

Learn prompt engineering best practices

As generative AI accelerates, businesses seek solutions to generate desired, high-quality outputs and avoid misinterpretation by AI.

Effective prompt engineering enables responsible, cost-effective, accurate, and relevant content.

Join AWS’ upcoming webinar to discover how to create a strategic prompt framework that can be reused and optimized to maximize AI investments.

Learn innovative, prompt engineering strategies from leaders at AWS and Rackspace Technology that allow developers to improve generative AI output.

Gain insights on how prompt engineering makes it easy for you to obtain relevant results right from the first prompt.

Learn More

Treats To Try.

*Glambase helps you build monetizable AI influencers that create content and engage with audiences autonomously. Early adopters get priority access. Check it out here.
Stackfix lets you compare software prices and features side-by-side to quickly find the best tool that matches your requirements.
Roster connects creators with experienced talent like video editors and thumbnail designers who've worked with top YouTubers like MrBeast.
Kyan Health helps companies support their employees' mental health with tools, multilingual counseling, and data insights that track well-being trends across locations (raised $16M).
Intellectia analyzes stocks and crypto investments with personalized technical analysis, trend ratings, and recommendations—we tried the free version, and it’s pretty comprehensive!

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

CRAZY video, and it’s open source?! Requires 45GB VRAM though…

HunyuanVideo from Tencent (above) is a 13B parameter open-source text-to-video (and video to video) model focused on high quality video generation with physical accuracy and scene consistency (code here).
230M people in China use genAI (that’s 1 out of every 6), with the company Baidu’s Ernie Bot the most popular at 11.5%. AI companies in China face a delicate balancing act between pursuing AGI and not threatening the government’s control over the people.
With the rise of popular Chinese open-source AI models like Qwen and DeepSeek, the CEO of open-source AI platform HuggingFace is now worried—he predicted in a separate post that China would lead the AI race by 2025.
Google announced it will let Google Cloud customers who use its Vertex AI platform use its Veo video generator (expanding from Youtube + Google Labs).

Under the Hood.

Fileserver is a tool that interfaces with your computer's file system to perform common operations like reading, writing, moving, and searching files on your desktop from within Claude.

Researchers released a new multilingual LLM evaluation benchmark called INCLUDE covering 44 languages with a focus on regional knowledge to test if AI models understand local cultural contexts.
Realtime from Trigger.dev shows your users live progress updates while their tasks are running, like showing a progress bar when uploading files or streaming AI responses as they're generated.
Rerank 3.5 from Cohere finds the most relevant information in your search results by reordering documents based on how well they actually match your query, now with better reasoning and support for 100+ languages.
Qwen Agent helps you build chatbot interfaces using Gradio 5, requiring Python 3.10 or higher for running the graphical user interface, and QwQ-32B-Preview reasons through complex problems with a 32K context window, though it's currently an experimental research model atm.

FROM OUR PARTNERS

Want to build AI projects faster and cheaper?

Want to build AI projects faster and cheaper? Join Speed Read AI's founders and Dell's experts on Dec 12th to discover how to run AI locally with Precision workstations and NVIDIA RTX™.

NEW SECTION: Best AI Image Round-up

We see a TON of awesome AI images every week—and we’re sure you do, too.

Below is a round-up of our favorite 6 images we saw this week, as well as a link to submit your own creations for a potential feature next time!

Low key this guy looks like someone we know…

Vote below for your favorite of the six images. Got a fave AI of your own? If you found or created your own AI contender for next week, fill out the submission form here.