The next gen models from Google & OpenAI

Also: OpenAI's safety team is falling apart

Author

Philip Mohr
May 18, 2024

Hey, it’s Philip.

what a crazy week of AI announcements 🤯 this was.

Today we cover some relevant aspects of Google’s and OpenAI’s product reveals, also concerning developments at OpenAI with Illya leaving, Fujifilm joining the CAI and Google focusing on watermarking technology.

For new readers, sign up here

THIS WEEKS RUNDOWN

  • OpenAI released GPT-4o (omni) with multi-modality across video and audio as well many model efficiency improvements - Link

  • ElevenLabs debuts its model for generative music - Link

  • Fujifilm joins the Content Authenticity Initiative and plans implement C2PA in top of the line cameras - LINK

  • IIlya Sutskever and other key figures in OpenAIs safety team resigned - Link

  • Google continues to lean on its watermarking technique by bringing SynthID to text and video - Link

  • Instagram Co-Founder Mike Krieger is joining Anthropic AI - Link

TOP STORIES

Google introduces its text-to-video model “Veo” to compete with OpenAI’s Sora as part of their Google I/0 announcements

Google has introduced a new generative AI video platform called Veo, aiming to compete directly with OpenAI's Sora.

Veo is capable of creating high-quality, 1080p video clips that can exceed 60 seconds in length, leveraging advancements from Google's DeepMind.

Like Sora, Veo is also not yet publicly available and only accessible for select users like artists and production companies. Source

OpenAI brings useful multi-modality to ChatGPT 4 and makes it free

Credit: OpenAi

OpenAI has introduced GPT-4o, a new flagship model offering GPT-4-level intelligence with improved speed and capabilities in text, voice, and vision.

Additionally, advanced tools and features, including image discussion, file uploads, and voice interactions, are being made available to ChatGPT Free users, with certain usage limits. Link

VIDEO OF THE WEEK

Impressive examples of Google’s new Veo model:

YOU GOTTA KNOW

SLOP and SLOM are the hot contenders as a new term for unwanted AI-generated content (aka AI SPAM). More here

TO CHECK OUT

Interesting stuff:

  • On the end of stock photography - Link

  • Podcast with OpenAI co-founder John Schulman on progress in AI development and his timeframe for AGI - Link

  • Sony Music sent letters to 700 AI companies to disclose if they trained on their content - Link

  • Study finds that 70% of consumers are afraid of becoming a victim of a deepfake fraud - Link

  • Researchers build a microphone to verify speech to counter voice cloning attacks - Link

  • How OpenAI stole Google’s Thunder - Link