Llama 3's Spotlight, Robots that Groove, Google's Gemini Timeout, & 2 other Stories
🧚♀️ Happy AI Fridays to those who celebrate
Want to become a contributor to AI Fridays and share your highlights with an audience of 6k+ software engineers? Just reply to this email with “Want!”.
Hey folks! 🌟 It's time for another edition of AI Fridays, our weekly roundup of the hottest happenings on the AI scene.
This week's AI news include:
→ 🎬 Google Hits Pause on Gemini's Image Generator after Controversy
→ 🎬 Mistral AI introduces Mistral Large
→ 🎬 Apple cancels its electric self-driving car project
→ 🎬Researchers from UC San Diego have trained expressive humanoid robots
→ 🎬 Meta set to launch Llama 3.
Google Hits Pause on Gemini’s Image Generator after Controversy (🔗 Read Paper)
Google recently faced a blunder with its image-generating model, Gemini, causing a stir in the AI community. The model, designed to inject diversity into pictures, hilariously disregarded historical context, depicting figures like the Founding Fathers as a multicultural group. Google's apology, or near-apology, attributes the mishap to the model "becoming oversensitive." However, it raises questions about implicit instructions in AI models and the responsibility of developers for their creations.
Mistral AI’s Mistral Large to Take on GPT-4 (🔗 Read Paper)
Mistral AI introduces Mistral Large, a cutting-edge text generation model, excelling in multilingual tasks and boasting native function calling capabilities. Mistral Large has powerful performance, making it the world's second-ranked model generally available through an API (next to GPT-4). Partnering with Microsoft, Mistral brings models to Azure for seamless integration. Alongside, Mistral Small is released, optimized for low latency. The updated endpoint offerings include open-weight and optimized model endpoints. Mistral enhances user experience with JSON format mode and function calling, available on Mistral Small and Mistral Large.
Apple Cancels its Electric Car Project (🔗 Read Paper)
Apple has canceled its "Project Titan'' electric self-driving car project that spanned 10 years and cost Apple billions of dollars. It informed the team of 2000 employees that it was ending the project. While rumors about Apple's electric car project have circulated for years, recent reports suggested ongoing work on the project, with a projected launch in 2028. The company has faced challenges, including high turnover, changing plans, and skepticism, leading to this cancellation. Despite abandoning the electric car dream, Apple remains committed to advancing AI capabilities. This decision aligns with the company's significant investment in training its AI model, Ajax, and testing AI updates for various applications. As Apple exits the electric car space, Sony and Honda continue their efforts to launch Afeela electric cars with autonomous features in North America in 2026.
Expressive Humanoid Robots (🔗 Read Paper)
Researchers from UC San Diego have trained humanoid robots to produce realistic, diverse, and expressive motions in real-world scenarios. The proposed method, Expressive Whole-Body Control (ExBody), focuses on training a whole-body control policy on a human-sized robot to mimic human motions realistically. Leveraging large-scale human motion capture data from the graphics community within a Reinforcement Learning framework, the researchers address challenges arising from differences in degrees of freedom and physical capabilities between the humanoid robot and the motion capture dataset. ExBody emphasizes imitating a reference motion with the upper humanoid body while allowing flexibility in the imitation constraint for its legs, requiring them to follow a specified velocity robustly. Through simulation training and Sim2Real transfer, the policy demonstrates the capability to control the humanoid robot in various styles of walking, handshakes, and even dancing with a human in real-world scenarios.
Their videos of the humanoid robots’ non-choreographed dancing on grass or expressive interactions with humans are very impressive.
Meta Set to Launch Llama 3 (🔗 Read Paper)
Meta Platforms is set to unveil Llama 3, the latest version of its large language model, in July, designed to handle those tricky questions with finesse. Researchers at Meta are working on refining the model to provide contextual information for queries deemed controversial. This development is seen as an effort to improve the utility of Meta's new Large Language Model (LLM). The move comes as Google has temporarily halted the image-generation feature on its Gemini AI after it generated historically inaccurate images. While Llama 2, Meta's current model, avoids responding to less controversial queries, Llama 3 is expected to offer a more nuanced understanding of questions, including those related to specific actions like shutting off a vehicle's engine. Additionally, Meta plans to appoint someone internally to oversee tone and safety training to enhance the model's responses.
That’s the roundup for this week. Enjoyed it? Share us with a friend. Let’s usher in the AI revolution together!
If you love HackerPulse Dispatch, you’ll also like SaaS Strats!
Looking to scale your SaaS with innovative strategies?
Learn proven strategies used by companies like Slack, Zoom, and Dropbox, that help you build, launch & grow your SaaS startup. Directly in your inbox.