The week in AI: Major new resources and updates from Anthropic and OpenAI

In partnership with

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each Thursday, we aggregate the major developments in artificial intelligence - we pass along the news, useful resources, tools and services; we highlight the top research in the field as well as exciting developments in open source. Even if you aren’t a machine learning engineer, we’ll keep you in touch with the most important developments in AI.

NEWS & OPINION

Enhance your AI skills with Anthropic’s new prompt engineering resources

-------------------------

Before diving in here, we’d like to note that if you weren’t aware already, Anthropic’s Claude 3.5 Sonnet is a state-of-the-art large language model with some incredible capabilities - and it’s free to use (potentially limited by server capacity). If you haven’t tested it out yet, we highly recommend you do so!

Anthropic just released two valuable resources for those interested in mastering prompt engineering, particularly with their suite of Claude models. The first is an “Interactive Intro to Prompting Tutorial”. This intro spans nine comprehensive chapters that cover all the foundational elements, including basic prompt structure and best practices for clarity, all the way down to more complex industry use cases. If you’re unfamiliar with terms like Chain-of-Thought reasoning and few-shot prompting, this is the place for you to become an LLM power user.

Anthropic has been at the forefront of effective prompt engineering and created one of the first prompt libraries for public use about 5 months ago. The intro tutorial provides hands-on exercises within an "Example Playground" to experiment with prompts and see their impact on Claude's responses in real-time. The interactive portion requires an API key - if you don't have one, you can sign up via the Anthropic Console (if you don’t want to use an API key, you can view the static tutorial answer key instead).

The second resource is a Real-World Prompting Course targeted at developers already familiar with the basics and looking to deepen their skills. This course delves into more advanced techniques, demonstrating how to incorporate key prompting strategies into complex, real-world applications. The five-lesson course begins with a recap of essential concepts and progresses through specific use cases like medical prompts, call summarization, and customer support bots. It also includes practical exercises to help developers refine their prompting abilities in real-world scenarios, ensuring they can maximize Claude's performance in production environments.

OpenAI also has a less robust (but more accessible) prompt engineering guide. You can view Anthropic’s entire course catalogue here.

OpenAI announces fine-tuning for GPT-4o: custom versions increase performance and accuracy

-------------------------

OpenAI has introduced fine-tuning capabilities for GPT-4o, allowing developers to tailor the model’s behavior to better fit their specific application needs. Fine-tuning GPT-4o can greatly enhance performance and accuracy across a variety of domains, from coding to creative writing. To help kickstart this process, OpenAI is offering 1 million free training tokens per day to every organization until September 23, 2024 (side note: Google is giving developers 1.5 billion tokens for free every day on its Gemini 1.5 Flash model and 1.6 million tokens on its Gemini 1.5 Pro model). Key points:

• Customization: Developers can now fine-tune GPT-4o to adjust the tone, structure, and response accuracy - even with small datasets - ideal for creating domain-specific models.

• Cost structure: Fine-tuning costs $25 per million tokens, with inference costs set at $3.75 per million input tokens and $15 per million output tokens. GPT-4o mini offers 2 million free training tokens daily until September 23.

• Success stories: Early testing with partners like Cosine and Distyl has shown significant improvements in AI performance, achieving state-of-the-art results in technical benchmarks.

• Safety and privacy: OpenAI has tried to emphasize data security, ensuring that fine-tuned models remain entirely under the user's control, with comprehensive safety measures in place to prevent misuse. We recently attended Section’s course on navigating AI data privacy, and if you have questions or concerns about the privacy of your data with OpenAI, we recommend watching the recording.

This release marks a big step forward in the way developers interact with and utilize OpenAI’s flywheel. It’s worth noting that at this point many fine-tuned open-source models can potentially yield increased performance for a fraction of the cost. By leveraging adapters, you don't have to pay up to 25x more or so just because it’s ChatGPT. However, for some organizations, the ecosystem and support provided by OpenAI might justify the additional cost, especially if they require the reliability and infrastructure that OpenAI offers.

California weakens bill to prevent AI disasters before final vote

-------------------------

California’s SB 1047, a bill designed to prevent AI disasters, has undergone significant revisions following input from key industry players, including Anthropic. Initially intended to give California’s government strong oversight over AI developers, the bill has been softened to reduce the state's power. Notably, the attorney general can no longer sue AI companies for negligent safety practices before a catastrophe occurs, a change that aligns with recommendations from Anthropic. Other amendments include the removal of the Frontier Model Division and relaxing legal obligations for AI labs regarding safety certifications.

SB 1047 still aims to hold developers accountable for AI-related disasters, but the modifications make the bill more palatable to the tech industry as has been the norm in the US for developing regulations around AI. The bill has advanced through California's legislature with relative ease, and its softened stance might help secure Governor Newsom’s approval without alienating the AI sector. However, the core issue - developer liability - remains contentious, with critics arguing that the amendments fail to address the bill's fundamental flaws. SB 1047 now heads to the Assembly floor for a final vote, with its future still uncertain.

BCLP Law has developed an interactive state-by-state legislation map, updated quarterly.

MORE IN AI THIS WEEK

• Wyoming voters face mayoral candidate who vows to let AI bot run government

• Nvidia unveils advances in digital humans and avatar tech at Gamescom

• Walmart earnings: stock rises to a record, company using AI to improve its business and save money

• AMD aims to catch up to Nvidia in AI battle - chipmaker acquires ZT Systems for $4.9 billion to help design better AI data center products

• SoftBank's Intel AI processor plans in doubt as insiders say it is now considering a TSMC partnership

• The humanoid robotics race between the US and China is intensifying (related: US tops AI ranking index with triple the investment, job postings as China)

• More AI lawsuits: authors sue Anthropic for copyright infringement

• In the search for less energy-hungry artificial intelligence, some scientists are exploring living computers

• International Data Corporation forecasts that worldwide AI spending is expected to reach $632 billion by 2028 - generative AI to account for 32% of that

🦾 Master AI & ChatGPT for FREE in just 3 hours 🤯

1 Million+ people have attended, and are RAVING about this AI Workshop.
Don’t believe us? Attend it for free and see it for yourself.

Save your spot here. (100 free spots only)

Highly Recommended: 🚀

Join this 3-hour Power-Packed Masterclass worth $399 for absolutely free and learn 20+ AI tools to become 10x better & faster at what you do

👉Save your seat  now (FREE for First 100)

🗓️ Tomorrow | ⏱️ 10 AM EST

In this Masterclass, you’ll learn how to:

🚀 Do quick excel analysis & make AI-powered PPTs 
🚀 Build your own personal AI assistant to save 10+ hours
🚀 Become an expert at prompting & learn 20+ AI tools
🚀 Research faster & make your life a lot simpler & more…

👉 Register here (Offer valid for First 100 people only)🎁

TECHNICAL, DEVELOPMENT, RESEARCH & OPEN SOURCE

• Microsoft releases Phi-3.5-MoE: an advanced, efficient and compact model that rivals the reasoning capabilities of much larger models

• xGen-MM (formerly BLIP-3): the latest foundational Large Multimodal Models (LMMs) developed by Salesforce AI Research

• Newly forming research area: Automated Design of Agentic Systems (ADAS)

• To code, or not to code? Exploring the impact of code in pre-training

• Google DeepMind publishes a retrospective of their work over the past 1.5 years across AI frontier safety, oversight, interpretability, and more

• CodeRabbit raises $16M to bring AI to code reviews

• Adobe’s TurboEdit: instant text-based image editing

• MistralAI publishes a multi-agent role-playing and knowledge graph construction example

• Unveiling Hermes 3: the first full-parameter fine-tuned Llama 3.1 405B model is on Lambda’s cloud

• Groq’s Distil-Whisper: a version of OpenAI’s Whisper text-to-speech model that is 240 times faster and significantly cheaper

That’s all for this week! We’ll see you next Thursday.