• Lore
  • Posts
  • Lore Issue #45: GPT3.5 Turbo Fine-Tuning Is Here

Lore Issue #45: GPT3.5 Turbo Fine-Tuning Is Here

Good morning.

This week we’ve got updates from OpenAI with GPT3.5 fine-tuning, Midjourney inpainting, a significant Copyright ruling that impacts how people use AI, and much more.

Also, we’ll launch Lore Japan next week—a lot of other great news to share soon.

Thank you to our partner Alpha for sponsoring this issue.

Let’s get to it.

  • a16z-backed Genesis Tx raised $200M for drug discovery (Link)

  • Elemental Cognition raised $60M to develop AI that “thinks before it talks” (Link)

  • Prins AI raised $22M for its AI-driven digital persona training product (Link)

  • Air cargo industry startup Aircon raised $3.3M (Link)

  • Synthpop raised $2.6M to develop AI assistants for Healthcare workflow automation (Link)

  • Writerly raised $2M for AI-powered writing tools (Link)

Plus: Databricks, an enterprise big-data company, is looking to raise ‘hundreds of millions’ in fresh funding to fuel generative AI push (Link)

Unleash the Power of Generative AI with Alpha

Embarking on an AI-driven future? Partner with Alpha, the go-to agency for turning generative AI into business solutions. Here's what makes Alpha exceptional:

  • Design Excellence: User-friendly interfaces that fuse aesthetics with AI.

  • Optimized Models: Tailored AI selection for performance and efficiency.

  • Precision Engineering: Accurate prompts and fine-tuned AI models.

  • Seamless Integration: Agile API usage that adapts to AI changes.

  • Trustworthy Security: Privacy and data protection is our priority.

Alpha's concise three-step process - understand, build, unleash - guarantees not only cutting-edge AI products but ongoing optimization for enhanced outcomes.

Ready to redefine your AI experience with Alpha? Don't miss out - slots are limited!

Illustration: Ruby Chen on OpenAI.com.

OpenAI has introduced fine-tuning capabilities for GPT-3.5 Turbo, allowing developers to customize models with their own data for specific use cases.

This feature, set to expand to GPT-4 this fall, enhances steerability, output formatting, and custom tone, and ensures customer data privacy. Early tests have demonstrated that a fine-tuned GPT-3.5 Turbo can even outperform base GPT-4-level capabilities on certain tasks.

In addition to performance enhancements, fine-tuning with GPT-3.5 Turbo doubles the token handling to 4k and significantly reduces prompt size, speeding up API calls and cutting costs. Support for additional features like function calling will be coming later this fall, making this a versatile update for developers looking to tailor their models more closely to their business needs.

According to US District Court Judge Beryl A. Howell, artwork generated by artificial intelligence is not eligible for copyright protection because it lacks “human involvement.”

The judgment is expected to serve as a pivotal reference point for future legal battles, exploring the boundaries of intellectual property law in relation to AI.

YouTube has published a new set of AI music principles to highlight its commitment to responsible innovation in the music industry. These focus on responsible collaboration with AI, empowering music partners, and extending trust and safety.

At the same time, YouTube announced its Music AI Incubator, launched in collaboration with Universal Music Group.

Other news:

  • Meta announces SeamlessM4T, a Multimodal AI Model for Speech and Text Translations (Link)

  • Microsoft and Epic expand AI collaboration to accelerate generative AI’s impact in healthcare, addressing the industry’s most pressing needs (Link)

AI for Business Conference (SPONSORED)

The AI for Business Conference is set to unfold in San Francisco on October 23, 2023, at the Hilton San Francisco Financial District.

Focusing on the transformative impact of Generative AI, this gathering will explore innovative topics ranging from Multiplying Coding Productivity to Next-Generation AI Technologies. Whether you're an experienced professional or just beginning in the field, join expert speakers for invaluable insights across industries.

Midjourney inpainting has landed

The AI art community has been waiting for this for a long time, and social media is buzzing with the news. You can now change specific parts of your Midjourney images for endless customization or tweaking — and it also works on the Discord mobile app.

AI-generated cinematic videos that look like professional drone captures

Replit’s AI Ghostwriter

Automate the repetitive parts of coding with Replit’s AI Ghostwriter. Comes with an in-IDE AI chatbot (Link)

Midjourney 🤝 Photoshop 🤝 Blender 🤝 Unity

More Midjourney-to-game experiments

A multiplayer infinite scalable browser-based world with AI NPCs

  • Flair: AI-generated branded content (Link)

  • Voila: Personal AI assistant for supercharged productivity (Link)

  • Lovo: AI voice generator (Link)

  • Boomy: Create original music (Link)

  • AI Code Convert: Convert Code or Natural Language To Programming Language Code (Link)

Find more tools at FutureTools.io.

  • Generative AI Explained — a free, self-paced online course from Nvidia’s Deep Learning Institute (Link)

  • GPT-3.5 Turbo fine-tuning now available. (Link)

  • Humans are now slower and worse at solving CAPTCHAs than ML-powered bots (Link)

  • New survey paper on instruction tuning LLMs (Link)

  • Reinforced Self-Training (ReST) for Language Modeling (Link)

  • SceNeRFlow: Time-Consistent Reconstruction of General Dynamic Scenes (Link)

Future ToolsEverything you need to know in AI news and tools

That's it!

See you next week!

If you want to support Lore, please:

  1. Share this newsletter 

  2. Follow us on Twitter @NathanLands & LoreAiNews.

  3. Join our Generative AI community on LinkedIn.