🌟Alibaba’s Competitor to OpenAI’s o1

Alibaba unveils competitor to OpenAI’s o1 reasoning model

Key Points 

• Alibaba’s QwQ-32B-Preview outperforms OpenAI’s o1 in logic puzzles and math problem-solving.

• The focus on reasoning models comes as AI labs question the effectiveness of scaling laws.

👨🏻‍💻News - A new “reasoning” AI model, QwQ-32B-Preview, has made its debut, challenging OpenAI’s o1. Developed by Alibaba’s Qwen team, QwQ-32B-Preview is the first reasoning model to be made available for download under a permissive license. 

With 32.5 billion parameters, it can handle prompts up to 32,000 words in length, and it outperforms OpenAI’s o1-preview and o1-mini on certain benchmarks.

🕵🏻‍♂️How does it measure up? According to Alibaba’s testing, QwQ-32B-Preview beats OpenAI’s models on two key benchmarks: AIME and MATH. AIME evaluates a model’s performance using other AI models, while MATH tests a model’s ability to solve word problems. 

The new model has shown strong abilities in solving logic puzzles and answering math questions. However, it's not flawless, occasionally switching languages unexpectedly, getting stuck in loops, or underperforming on tasks requiring common sense reasoning.

🧐Are reasoning models the next big thing? The growing attention on reasoning models comes amid increasing doubts about the effectiveness of scaling laws—the theory that adding more data and computing power would continuously enhance a model’s capabilities. 

As major AI labs like OpenAI, Google, and Anthropic see slower improvements, companies are shifting focus to new approaches like test-time compute, which provides extra processing time during task completion and underpins models like o1 and QwQ-32B-Preview. Google has also expanded its reasoning model team significantly, signaling growing confidence in this approach.

ElevenLabs launches new AI podcast feature

Key Points 

• ElevenLabs has launched GenFM, enabling users to create multispeaker AI podcasts in 32 languages.

• GenFM adds "ums" and pauses to AI podcasts, creating a more natural, conversational flow.

🎙️News - Voice AI startup ElevenLabs has launched a new feature called GenFM, allowing users to upload different types of content to create a multispeaker podcast. 

Available on the ElevenLabs Reader iOS app, GenFM currently supports 32 languages, including English, Hindi, Portuguese, Chinese, Spanish, French, German, Japanese, and more. Users can upload YouTube videos, text, or documents, and the app will automatically select two voices from over a dozen options to create the podcast.

✨What makes it unique? What sets GenFM apart is its unique touch of adding human-like elements to the AI-generated podcast. Instead of removing filler words like “um” and “ah,” the app includes them, making the conversation sound more natural and relatable. 

Jack McDermott, ElevenLabs' head of mobile growth, explained, the goal is to strike the right balance between a smooth conversation flow and human-like pauses, similar to long-form podcasts with insightful dialogues. 

👾What's more? Looking ahead, ElevenLabs plans to introduce more customization options, giving users the ability to add multiple sources for a richer, generative AI podcast experience.