📆 September 17th 2024📆 

Touching on another topic related to shaping up your data organisation, the practical considerations that you must think about before you start writing a data & AI strategy.

In the world of AI news, we are not seeing a slow down:

• OpenAI releases a new model which greatly improves on reasoning tasks, claiming PhD level thinking.

• Two big API announcements from Luma & Runway bringing video generation to developer workflows

• xAI’s training cluster is now considered one of the largest in the world, leave it to Musk to one up the competition….. again

• Slack are opening up to the world of agents - allowing agents from other systems to help consolidate communications across multiple tools

Dive in for all the details!

🤖 Unlock AI 🤖 

AI Strategy for Your Business

I’m a firm believer in executable strategies. Developing a strategy that checks all the right boxes can be both challenging and time-consuming. Often, the focus shifts toward crafting something that sounds good on paper but lacks actionable follow-through. To ensure reasonable actions can emerge from your strategy it must be informed by the realities of your business. Below are the key areas to understand so that that can be true.

Do You Have the Rights to Data?

A critical first step is cataloguing your data, as I’ve written about before. But just as important is understanding what data you’re allowed to use and how you can use it. Many AI labs have faced legal challenges over data collection and training practices, which is understandable. After all, content creation isn’t cheap, and using data without proper rights can lead to costly consequences.

Even if you have a proprietary system, there are risks. Big AI models are typically trained on publicly available datasets, making your proprietary data highly valuable. But that also comes with legal risks. Before diving into your AI initiatives, ensure your terms of service allow you to use customer-generated data.

Also, consider data protection laws like GDPR in Europe or CCPA in the U.S., particularly the right to be forgotten. Using retrieval-augmented generation (RAG) solutions instead of retraining models can help you comply with these regulations, as it’s easier to remove customer information. Additionally, learn to leverage data obfuscation techniques to anonymize sensitive information and stay within legal boundaries.

Focus on One Business Objective

Don’t aim to solve all your company’s problems with one AI strategy—you don’t have the resources, team, or experience for that. Neither do the fancy consultants. Instead, pick one impactful and achievable problem.

A major issue with strategies is that they’re often overly ambitious and, as a result, unattainable. To avoid this, clarify who you’re trying to help. Do you want to enhance customer experiences to add value to your product, or are you aiming to boost internal efficiency? Whatever the focus, stay aligned and realistic.

What Tools Do You Have?

The tools you use can make or break the execution of your strategy. Start by assessing what tools you already have, how they function, and whether they meet your needs. Recognize the limitations in your current tech stack, and be practical about which constraints matter most for achieving your goals.

Sometimes, investing in adjacent areas is unavoidable, which could extend your timeline. But in the long run, it will position you for success. For instance, investing in solid data management tools can build trust in your data. Off-the-shelf solutions can also accelerate your time to market, but with so many options out there, choose a partner, not just a vendor. A good partnership will focus on mutual success, fostering a more engaged relationship than a simple subscription model.

Do You Have the Skills?

Your team is critical. People who are well-trained and invested in the process can make or break your AI initiative. While retraining existing staff is possible, it’s often time-consuming and expensive for certain roles. Carefully evaluate the strengths of your current team, identify skill gaps, and hire accordingly. Be selective when hiring—look for strong engineering mindsets, as these will be invaluable throughout the journey. Look for demonstratable skills and abilities - with the right team your project will reach the stars.

Is Your Data of Good Quality?

Data governance is often intimidating, but it doesn’t have to be. Instead of focusing on governance itself, think about data quality. Poor data quality is the number one obstacle to getting your AI project into production. After all, what’s the point of weeks of work if your AI tells staff to sell office equipment to save time or directs customers to a competitor?

By emphasizing data quality, you can ensure your data is:

• Accurate: Reflects true values and is free from errors.

• Complete: No missing fields that could impact usability.

• Consistent: Aligned across all datasets and systems.

• Timely: Current and available when needed.

• Valid: Meets the required formats and standards.

• Unique: Free from duplicate records.

Achieving this isn’t easy, but investing in the right monitoring tools to flag issues can make a big difference. Strong data foundations reduce unexpected problems and pave the way for smoother AI deployments.

Next Week 

Next week I’m going to explore Cursor & the new capabilities of coding models as practical means to improve the efficiency of your coding workflows.

💫Help Support my Work💫

I really enjoy researching and writing about AI & appreciate your support. If you enjoy this content please use any of the options below to help

• Buy Me a Coffee ☕️- AI is really fuelled by coffee - keep the tank filled

• Spread the Word 🗞️- Share on social or directly with your friends

• Follow me on X - Always happy to chat AI and software dev

• Data & AI Consulting - From struggling data teams to developing executable strategies - lets talk about how my experience can supercharge your next data project

📰 News 📰 

OpenAI Launches New Reasoning Models for Developers

OpenAI has made significant advancements with the launch of its new AI model family, o1, designed to improve reasoning capabilities through a technique called "chain of thought." This method enables the models to break down complex tasks into smaller steps, enhancing their problem-solving abilities, especially in areas like coding, mathematics, and scientific reasoning. The o1 family includes two models: o1-preview and o1-mini, both offering varying levels of performance.

The o1-preview model stands out for its PhD-level reasoning, excelling in high-level tasks like solving math Olympiad problems and performing advanced programming challenges. It significantly outperforms its predecessor, GPT-4, in these areas, although it comes with a higher inference time and cost. The o1-mini offers a more budget-friendly option, targeting developers and researchers who need reasoning power but can sacrifice some depth​.

These models are seen as a leap forward in AI, moving beyond simple language generation towards more sophisticated problem-solving. However, OpenAI acknowledges that not all tasks require this deep reasoning, and for certain applications, quicker models like GPT-4o may still be preferable​.

For developers, the new models open up exciting opportunities, particularly in STEM fields, where tasks like coding optimization, workflow automation, and document analysis can benefit from o1's advanced reasoning abilities​.

Luma AI & Runway Launch APIs

This week saw the release of two major AI video generation tools from Luma AI and Runway, intensifying competition in the space. Luma AI's Dream Machine API is designed for developers, offering flexible video generation features such as text-to-video and keyframe control. Its pricing structure, at $0.32 per million pixels, aims to make high-quality video creation more accessible, focusing on democratization and responsible usage.

Meanwhile, Runway's Gen-3 Alpha Turbo API focuses on enterprise use, with faster video generation and custom workflows, catering to industries like media and entertainment. Runway's phased rollout and pricing options ($0.01 per credit) make it appealing for businesses seeking advanced video solutions. Both releases demonstrate the growing sophistication of AI-driven video, with Luma targeting broader creative use and Runway focusing on industry-specific applications

Colossus: The World's Most Powerful AI System

The XAI Cluster, dubbed "Colossus," is now considered the most powerful AI training system in the world, with a staggering 100,000 Nvidia H100 GPUs powering its capabilities. Developed by Elon Musk's xAI, the system can process massive AI models with unprecedented speed and efficiency. Colossus is designed to train xAI's advanced language model, Grok, and promises to revolutionize the AI landscape. However, concerns have been raised regarding the system's energy consumption and storage capacity. Reports suggest it is currently powered by temporary solutions like diesel generators while awaiting additional grid support, highlighting potential sustainability challenges.

The name "Colossus" also attracted attention, referencing a 1970s sci-fi story about a supercomputer that turns on humanity, perhaps hinting at Musk’s penchant for dramatic symbolism. Despite these challenges, xAI plans to double the system's capacity by adding 100,000 more GPUs, further solidifying its dominance in AI model training​

Slack Integrates AI Agents for Enhanced Workflow

Slack's latest update introduces powerful AI agents and workflow automation features, designed to enhance productivity and streamline operations for users. With the new Agentforce framework, Slack integrates third-party AI agents from companies like Adobe, Asana, and Workday, allowing users to interact with these agents to manage tasks, generate content, and optimize workflows directly within the Slack interface. These agents can help teams with tasks like drafting emails, summarizing discussions, or managing project updates.

Additionally, Slack has expanded its Workflow Builder to include AI capabilities. Users can now build automated workflows through natural language prompts, making it easier to set up reminders or generate reports. The AI-driven search function has also been enhanced to sift through various file types and provide more relevant, contextual answers based on both conversations and documents shared across channels.

This update aims to further integrate AI into the everyday workflow, providing tools that can save time and help teams focus on more strategic, high-value activities

🔨 AI Powered Tools 🔨 

• SlideSpeak is an AI tool that helps users quickly create professional presentations by leveraging ChatGPT. It generates slides from simple prompts, automating content generation to save time. Ideal for professionals needing polished presentations with minimal effort.

• HumanizerAI transforms robotic-sounding AI-generated text into more natural, human-like language. It enhances any AI output, such as emails or chat responses, by making them sound warmer and more relatable.

• AutoGen by Microsoft is an open-source framework for building AI-driven systems with multi-agent capabilities. It facilitates advanced workflows by allowing large language models (LLMs) to interact, improving performance and flexibility across applications.