LUMA Competes with SORA for AI Video Generation

Hello, AI Enthusiasts!

Welcome to FavTutor’s AI Recap! We’ve gathered all the latest and important AI developments from the past week in one place, just for you.

In Today’s Newsletter: 😀 

• LUMA’s Dream Machine Competes with SORA for AI Video Generation

• Stability AI launched Stable Diffusion 3 Medium text-to-image model

• Study Finds AI Fails with a Simple Problem Even Kids Can Solve

• Microsoft Copilot+ PC’s Recall AI feature gets delayed

• New NVIDIA AI Models Can Generate Training Data for LLMs

LUMA
💡 LUMA’s Dream Machine Competes with SORA for AI Video Generation

Luma AI is the San Francisco-based artificial intelligence startup that dropped the next big thing called Dream Machine, an AI video generator like OpenAI’s SORA. It’s also open for access to the public.

Insights for you:

• LUMA’s Dream Machine can create high-quality videos with a speed of 1 frame per second, so a 5-second video in 2 minutes only.

• Check out these amazing examples of AI videos created by Dream Machine, along with prompt ideas.

• Dream Machine can comprehend interactions between humans, animals, and things in the real world.

STABILITY AI
⚡️ Stability AI launched Stable Diffusion 3 Medium text-to-image model

Stability AI says that Stable Diffusion 3 Medium is their “most advanced text-to-image open model yet”. SD3 Medium can understand complex prompts involving spatial relationships, compositional elements, actions, and styles.

Insights for you:

• Stable Diffusion 3 Medium has a smaller size, which makes it perfect for consumer PCs and laptops.

• The model can produce photorealistic outputs with less error-prone typography.

• The weights are available under an open non-commercial license and a low-cost Creator License.

AI
👎️ Study Finds AI Fails with a Simple Problem Even Kids Can Solve

This new study showed that current powerful LLMs fail when drawing simple logical conclusions when giving incorrect responses to a simple elementary-level puzzle. This is quite an important experiment that sheds light on LLMs' vulnerability.

Insights for you:

• Research shows that current powerful LLMs fail to solve a simple ‘Alice In Wonderland’ problem that kids can solve.

• The models not only provided incorrect responses, they also blatantly tried to justify their incorrect reasonings.

• GPT-4o performed better than other models like Claude 3, Mistral Medium and many more.

MICROSOFT
🧐 Microsoft Copilot+ PC’s Recall AI feature gets delayed

The controversial Recall feature won’t be available to Copilot+ PCs, while these PCs will be launched this week, according to their update blog post. This happened because of the security concerns raised by the privacy advocates.

Insights for you:

• With Recall, you can retrieve almost anything you’ve seen or done on your computer like having a photographic memory.

• The decision to delay happened after receiving feedback on the Recall from the Windows Insider Community.

• Microsoft has said that Recall is designed to remain private on-device with encryption.

NVIDIA
🛠️ New NVIDIA AI Models Can Generate Training Data for LLMs

NVIDIA’s Nemotron-4-340B-Instruct is a Large Language Model that can be utilized in a synthetic data production pipeline to produce training data that will aid researchers and developers in developing LLMs of their own.

Insights for you:

• NVIDIA announces Nemotron-4 340B, an open-source family of LLMs that can generate synthetic data.

• It is built on the conventional decoder-only transformer model.

• Rivals GPT-4 and can be fine-tuned using the NeMo framework or LoRA.