Last Week in AI #213 - Midjourney video, Gemini 2.5 Flash-Lite, LiveCodeBench Pro
326 snips
Jun 26, 2025 Midjourney has taken a leap into video with a new AI model capable of generating 21-second clips, making video creation more accessible than ever. Google’s Gemini AI family introduces efficient models and voice interaction with search, transforming user experience. Meanwhile, groundbreaking benchmarks like Live Code Bench Pro and the Abstention Bench are set to challenge the reasoning capabilities of language models, revealing intriguing limitations. OpenAI's $200 million defense contract showcases a growing alliance between tech giants and government for innovative AI applications.
AI Snips
Chapters
Transcript
Episode notes
Midjourney's Video Leap
- Midjourney launched its first AI video generation model, offering up to 21-second clips at an affordable price.
- This marks a shift from text-to-image to video generation, enhancing creative possibilities for users.
Quirky AI Video Phase
- Daniel shares a funny example of early AI-generated videos, including surreal content like Donald Trump and Elon Musk making an "anti American sandwich."
- This phase of quirky AI content is amusing but likely to fade as quality increases.
Voice AI vs Text Usage
- Voice-based AI interactions are becoming more common with Google enabling back-and-forth voice conversations in search.
- However, most users still prefer text interaction, leaving voice mode adoption an open question.