You are currently viewing LLM Slowdown, Creative AI in Music and Video, and Ethical Boundaries in Marketing

LLM Slowdown, Creative AI in Music and Video, and Ethical Boundaries in Marketing

Hello AI Friends,

This week there was a lot of ground to cover with video and music updates, a debrief of AI’s role in the recent U.S election but the overwhelming theme of the meetup was that LLMs are slowing down in their progress.

Indeed we might ask what even constitutes progress with LLM tech and can we even describe a model upgrade as an upgrade, as the CEO of Anthropic touched on in a recent interview with Lex Freidman. 

We even got an answer as to why the latest upgrade to Claude 3.5 was not labeled Claude 3.6.

We blew a lot of minds with the demos of Suno V4, with a lot of new AI enthusiasts joining this week that weren’t aware of text-to-music technology at all.

And it’s great to see regular faces getting more confident in their understanding of the different models and the AI landscape.

See you next week, 

Harry Verity

Bali AI Meetup host

Co-Founder, AI To The World

Attendees:

  • Harry Verity (host) – Tech journalist and AI consultant
  • Cameron – Operations consultant interested in AI
  • Chris – Freelance developer using GitHub Copilot
  • Victor – Founder of TabTrooper.com, an AI browser with Vision
  • Will M – Founder of Intelligent Marketer, AI marketing expert
  • Nave B – Building a SaaS product
  • Several other attendees

Key Topics Discussed:

1. Model Updates and Industry Trends

  • The group delved into recent developments in large language models (LLMs), focusing on OpenAI’s upcoming “Orion” model and Anthropic’s Claude series.

  •  A key point of discussion was the apparent slowdown in significant improvements between model iterations. Will provided insights on Claude’s capabilities, nothing that the Haiku model (3.5) is potentially more capable than the previous Opus (3) version.

  • The conversation touched on the challenges of model naming conventions and the nuances of incremental improvements.

  •  Anthropic’s CEO’s recent interview was referenced, highlighting the complexity of model development and the shift towards longer reinforcement learning periods with human feedback.

2. AI in Creative Industries

Music Generation:

  • A live demo of Suno AI’s V4 music generation capabilities impressed attendees.

  • The group discussed the rapid advancements in AI-generated music and the potential copyright implications. 

  • The conversation touched on the legal challenges faced by companies like Suno and Udio, highlighting the complex intersection of AI, creativity, and intellectual property law.

Video Generation:

  • Advancements in AI video generation were showcased, particularly the breakthrough in character consistency with Kling. 

  • The group examined examples from Kling AI, a Chinese video model, and discussed the potential for creating consistent characters across multiple video generations. 

  • The process of training these models using low-resolution adaptation (LoRA) techniques was briefly explained.

3. AI and the U.S. Election

  • The meetup coincided with the aftermath of the recent U.S. elections, prompting a discussion on how AI tools handled election-related queries. 

  • The group debated the implications of AI in information dissemination during critical political events and the measures taken by AI companies to prevent misinformation.

4. AI in Sales and Marketing

  • Harry presented several AI-powered sales tools, including:

  • Hey Reach: For LinkedIn automation and lead generation with multiple burner accounts

  • Clay: A versatile tool for gathering and analyzing company data that can be used with Ocean and Panda Match

  • The ethical implications of using AI for automated outreach were debated, with discussions on LinkedIn’s terms of service and email deliverability best practices.

5. Ethical Considerations and Future Implications

  • Throughout the meetup, conversations frequently returned to ethical considerations

Topics included:

  • The potential for AI to be used in creating fake or misleading content

  • How AI companies handle sensitive or potentially dangerous queries

  • The balance between AI advancement and responsible use

6. Bonus research papers we ran out of time to go over

  • Research from Byte Dance (Tik-Tok) and a Chinese University

Leave a Reply