Phased Approach | Claude 3.5 : Why should you care?

Welcome to this week's Phased Approach Newsletter, where we're diving into the recent release of Anthropic's Claude 3.5 Sonnet model and its implications for the AI landscape.

The AI Echo Chamber vs. Real-World Adoption

It's been just about 2 weeks since Claude 3.5 Sonnet hit the scene, but in the fast-paced world of AI, it feels like an eternity. The internet has been full of excitement, showcasing impressive benchmarks and cool examples of what this new model can do. But amidst all this chatter, a crucial question emerges: Does anyone outside the AI development bubble really care?

As someone working in this field, I'm struck by the stark contrast between the professional AI news ecosystem and the actual adoption rates in businesses. Anthropic's models have been impressive since Claude 2, yet they seem to fly under the radar outside the AI echo chamber. In conversations with small businesses, tech entrepreneurs, and general knowledge workers, I've noticed a trend: while many are familiar with tools like ChatGPT, Canva, and Midjourney, very few have ventured into using Claude.

The Tool-stack Dilemma

This observation leads us to a broader issue plaguing the AI tool landscape. We're surrounded by so many tools that are almost useful, but there's no single "killer app" that handles everything. Recently, I took stock of my own tool-stack and realised that every single tool I use now boasts its own AI interface. From Google Workspace AI and GitHub Copilot to Slack AI, Notion AI, and Canva – not to mention the dedicated AI tools I actually use like Perplexity, Claude, and ChatGPT – the options are overwhelming.

The question is: Can we realistically use all these AI-enhanced tools effectively? Or are we facing a case of AI feature overload? My guess is that this will all consolidate into features that just do the things without us having to ask, but until then it is a challenge to keep up and get utility out of these features.

What's on the Agenda

In this newsletter, we'll explore:

1. The Claude 3.5 Sonnet release: What's new and noteworthy?
2. The intriguing "Artifacts" feature: A potential game-changer for small businesses?
3. Real-world use cases: Are they actually practical and accessible?
4. Implications for small, medium, and enterprise-level businesses
5. Microsoft release GraphRag, which adds powerful Graph search to your RAG toolkit

Join us as we unpack these topics and attempt to bridge the gap between cutting-edge AI developments and practical, real-world applications.

Let's dive in! ??

CLAUDE SONNET 3.5 new faster and more capable model

The Claude 3.5 Sonnet model from Anthropic represents a significant advancement in AI capabilities. Compared to the previous Claude 3 Opus model, the Claude 3.5 Sonnet outperforms it on a wide range of benchmarks, including graduate-level reasoning (GPQA), undergraduate-level knowledge (MMLU), and coding proficiency (HumanEval).

The Claude 3.5 Sonnet operates at twice the speed of the Claude 3 Opus while maintaining superior performance. This speed boost, combined with cost-effective pricing, makes the Claude 3.5 Sonnet an attractive option for complex tasks such as context-sensitive customer support and orchestrating multi-step workflows.

So a new model has come out and its faster and cheaper, so far so boring. People like me are interested for backend apps and services. It is also much better a coding tasks which will be useful for developers. But what about business users?

The Game-Changing "Artifacts" Feature

One of the most intriguing aspects of Anthropic's latest release is the introduction of "Artifacts." This feature represents a significant leap forward in how users can interact with AI-generated content.

What Are Artifacts?

Artifacts allow users to view and interact with Claude's outputs - whether they're code snippets, text documents, or even website designs - in a dedicated workspace alongside their conversation. It's like having a digital whiteboard where Claude can sketch out ideas, present data, or showcase designs in real-time.

A Step Beyond Previous Attempts

This isn't the first time we've seen attempts at interactive AI outputs. ChatGPT, for instance, has experimented with similar concepts, most notably with its code interpreter feature. However, OpenAI struggled to make these tools accessible to the average user.

Anthropic's Artifacts feature seems to be aiming higher. It's positioning itself as a coding assistant for non-technical users, allowing them to run calculations, produce apps, or create dashboards directly in the browser with minimal effort.

Early Impressions

After spending several hours exploring this feature, I've been genuinely impressed by what it can accomplish with remarkably simple prompts. The potential applications are vast:

• Transform complex PDFs with charts and graphs into interactive dashboards in seconds
• Quickly prototype web applications without deep coding knowledge
• Visualize data and concepts in ways that are immediately understandable and manipulable

Real-World Implications

The Artifacts feature could be a game-changer for small businesses and individual professionals who lack extensive technical resources. Imagine being able to:

• Create a basic inventory management system without hiring a developer
• Design and test a simple customer feedback portal in minutes
• Generate interactive reports from raw data for client presentations

While it's still early days, the potential for democratising certain aspects of software development and data analysis is evident. However, questions remain about how this will integrate into existing workflows and whether it can truly bridge the gap between AI capabilities and practical, everyday business needs.

In the following sections, we'll dive deeper into specific use cases and explore what this means for businesses of all sizes.

Diving Deeper: Practical Use Cases for Artifacts

To truly understand the potential impact of Claude 3.5 Sonnet's Artifacts feature, let's explore some real-world use cases in detail:

1. Financial Reporting Dashboard

Scenario: A small investment firm needs to quickly analyze and present complex financial data to clients.

How Artifacts Helps:

• Upload a 200-page PDF containing financial reports, market research, and economic forecasts.
• Within seconds, Claude analyses the document and generates an interactive dashboard.
• The dashboard includes: Key performance indicators (KPIs) extracted from the report Interactive charts showing market trends A summary of risk factors with adjustable parameters Real-time calculation of financial ratios based on user inputs

Benefits:

• Saves hours of manual data extraction and visualisation
• Allows for on-the-fly adjustments during client meetings
• Simplifies complex financial analysis for smaller firms without dedicated data teams

2. Prototype User Interfaces

Scenario: A startup is rapidly iterating on their app design and needs to test multiple concepts quickly.

How Artifacts Helps:

• Describe user interface elements in natural language (e.g., "Create a modern, minimalist home screen with a search bar at the top, followed by a scrollable feed of content cards").
• Claude generates clickable prototypes in seconds.
• Create multiple iterations of the interface and easily switch between them.
• Adjust elements on the fly by giving additional instructions.

Benefits:

• Dramatically speeds up the prototyping process
• Allows non-designers to participate in the UI/UX design process
• Facilitates quick A/B testing of different design concepts
• Reduces the need for specialized prototyping tools and skills

3. Scientific Research Visualisation

Scenario: A graduate student needs to compare their research data with published studies and create compelling visualizations.

How Artifacts Helps:

• Upload published scientific papers related to the research topic.
• Claude extracts relevant data and generates charts and graphs from the papers.
• Input your own research data and ask Claude to create comparative visualisations.
• Generate interactive plots that allow for real-time data manipulation and statistical analysis.

Benefits:

• Simplifies the literature review process by automating data extraction
• Creates publication-quality visualisations without advanced graphing skills
• Allows for quick hypothesis testing by manipulating variables in real-time
• Facilitates the discovery of patterns and correlations across multiple studies

The Bigger Picture

These use cases demonstrate the versatility and power of the Artifacts feature. By bridging the gap between complex data analysis, design, and visualisation tasks and the average user, Claude 3.5 Sonnet is potentially creating access to capabilities that were once the domain of specialists.

However, it's important to note that while these tools are impressive, they're not a complete replacement for human expertise. The real power lies in how they can augment human capabilities, allowing professionals to focus on higher-level analysis, creativity, and decision-making. One major issue I found with Artifacts was that very often the first thing it produced wasnt great and with some gentle prodding and specific requests for different types of analysis it got better.

Check out this dashboard with moveable inputs to run a Monte Carlo simulation. 2 years ago your would have needed serious skills to do this.

My question is with all the noise and constant news, will this tool even be used? I still think there is going to be a need for skilled users of these tools to help businesses to get the most out of what these models can now achieve

.

Microsoft GraphRAG

The demise of Retrieval Augmented Generation has been greatly exaggerated. Most Enterprise implementations of LLMs and Foundation models use a form of RAG in their backend implementations to control hallucination, private data and keep guardrails in place. With context windows having larger and larger sizes available, many people that don't work in enterprise AI think that this means that RAG is irrelevant. This in my opinion is inaccurate and a misunderstanding of the full purpose of RAG in enterprise. What we do need though is faster and more accurate search in RAG so an excellent adjunct to this stack is Graph databases and graph search.

Microsoft recently released GraphRAG, a new graph-based approach to retrieval-augmented generation that enables question-answering over private or previously unseen datasets. GraphRAG uses a large language model (LLM) to automatically extract a rich knowledge graph from any collection of text documents, allowing it to report on the semantic structure of the data prior to any user queries.Some key features of GraphRAG include:

• Detecting "communities" of densely connected nodes in the graph in a hierarchical fashion, partitioning the data from high-level themes to low-level topics
• Generating "community summaries" using an LLM to describe the entities and relationships in each community, providing an overview of the dataset
• Supporting a new class of "global queries" that naive RAG approaches cannot generate appropriate responses for

Evaluations showed that GraphRAG outperforms naive RAG on comprehensiveness and diversity of answers when using community summaries, while using substantially fewer tokens than hierarchical source text summarisation. GraphRAG is now available on GitHub, along with a solution accelerator that provides an easy-to-use API experience hosted on Azure. Microsoft is working to reduce the upfront costs of graph index construction while maintaining response quality, and is exploring ways to approximate the knowledge graph and community summaries to enable evaluation with minimal indexing costs.

Thats it for this week folks. As always we are here to help with any questions you might have on how the topics discussed here can be relevant in your own business. Please drop me a line if you would like some help of just have a question.