Must See AI News This Week - 11 December 2024

Dec 11, 2024

Article voiceover

1×

0:00

-15:23

🤯 Most Interesting This Week

OpenAI released its o1 model out of preview, now available to ChatGPT Plus and Team users, with Enterprise and Education access rolling out next week.
The full o1 now handles image analysis and produces faster, more accurate responses than the preview, with 34% fewer errors on complex queries.
A new $200/month Pro plan includes unlimited access to o1, GPT-4o, Advanced Voice, and future compute-intensive features.
Pro subscribers get exclusive access to 'o1 pro mode,' featuring a 128k context window and stronger reasoning on difficult problems.
In testing, o1 Pro achieved:
- 80% reliability in math (AIME).
- 75th percentile in coding (Codeforces).
- 74% reliability in science (GPQA Diamond).
The full o1 appears to perform worse than the preview version on several benchmarks, though both surpassed the 4o model.
The livestream showcased o1 Pro tackling complicated thermodynamics and chemistry problems after minutes of thinking.

OpenAI launched Sora, its AI video generation model, now available to ChatGPT Plus and Pro subscribers.
Sora creates up to 20-second outputs in various aspect ratios, with a new ‘Turbo’ model for faster generation.
The web platform allows users to:
- Organize and view prompts.
- Get inspiration from other users’ prompts and featured content.
Creative tools include:
- Remix for scene editing.
- Storyboard for stitching outputs.
- Blend, Loop, and Style presets.
Available to ChatGPT subscribers, with additional features for Pro plan users:
- Unlimited generations.
- Higher resolution outputs.
- Watermark removal.
Content restrictions apply to:
- Real people and minors.
- Copyrighted material.
Rollout excludes the EU, UK, and other territories due to regulatory concerns.

Amazon introduced Nova, a new family of AI models competing with GPT and Claude, featuring fast text and full video generation capabilities.
Nova models support:
- Text, image, and video inputs.
- Enterprise workloads with a focus on cost and latency reduction.
Nova model lineup:
- Nova Micro: Text-only, 128K context.
- Nova Lite: Multimodal, 300K context.
- Nova Pro: Advanced multimodal, 300K context.
- Nova Premier: Coming in 2025.
- Nova Canvas: Image generation.
- Nova Reel: Video generation.
Nova Pro costs roughly 1/3 of GPT-4’s price while matching or exceeding its performance on benchmarks.
Technical capabilities include:
- Async calls for video generation.
- Fine-tuning on text, images, and video.
- Function calling for agents.
- RAG through Bedrock.
- Streaming responses and cross-region inference.

Google DeepMind’s new gemini-exp-1206 model has reclaimed the top spot on the Chatbot Arena leaderboard, surpassing OpenAI across multiple benchmarks.
Key highlights:
- Processes and understands video content, unlike ChatGPT and Claude (which only handle images).
- Features an impressive 2M token context window, allowing it to process over an hour of video content.
Released on Gemini’s one-year anniversary:
- Climbed from second to first place on the Chatbot Arena leaderboard.
- Freely available through Google AI Studio and the Gemini API.

Microsoft launched Copilot Vision, allowing its assistant to see and interact with web pages in real-time in the Edge browser.
Vision integrates into Edge, enabling Copilot to analyze text and images on approved websites when enabled.
The feature assists with tasks like:
- Shopping comparisons.
- Recipe interpretation.
- Game strategy on supported sites.
Previously revealed in October, Vision includes voice and reasoning capabilities.
Microsoft emphasizes privacy:
- Vision is opt-in only.
- Voice and context data are automatically deleted after each session.

X briefly rolled out Aurora, a new AI image generator integrated with Grok, producing more photorealistic images than the previous Flux model.
Aurora showed significant improvements, particularly with:
- Landscapes.
- Still-life images.
- Human photorealism.
The model had minimal content restrictions, allowing the creation of copyrighted characters and public figures.
Elon Musk called it a "beta version" that will improve quickly.
X Developer co-lead Chris Park revealed that Grok 3 is coming, taking aim at OpenAI and Sam Altman in the announcement.

Google revealed Willow, a new quantum computing chip achieving major performance breakthroughs in error correction and computation speed.
Willow reduced errors exponentially as more qubits were added, addressing a critical issue in quantum computing.
Key highlights:
- Willow completed a computation in under five minutes that would take today’s fastest supercomputers 10 septillion (10^25) years.
- Utilized 105 qubits, maintaining the quantum state for nearly twice as long as previous designs.
Manufactured at Google’s new quantum fabrication facility in Santa Barbara.