🐹 Hamsterdam part one: the reboot (9/22 to 9/28, 2024).

A weekly marketing and AI content recap.

Welcome to the first week of the new Hamsterdam! 🐹

I’ve been away from the SEO game for a while, re-envisioning how I want to contribute.

👉 I’m back on the consulting horse, but with a new mission. 🙌

SEO is dead, in my view. 🪦

Long live holistic marketing. 👈

As a result, the format of Hamsterdam has changed slightly.

In the old version, I relied on social media, newsletters, and other sources of information for SEO news.

This time around, I’m focusing more broadly on marketing and AI.

I’m also giving up on socials to focus exclusively on my Google Discover feed, including choice content I found during the last week or so.

Below, you’ll find links to the articles along with detailed excerpts. All words are ascribed to the authors. All bolding is mine.

How about a quick summary of this week’s content: well, there’s a lot about knowledge graphs and RAG. If you’re still doing SEO the old way, pay attention. 😉

Without further ado, let’s dig in!

1. Fine-Tune Smaller Transformer Models: Text Classification – Ida Silfverskiöld, Towards Data Science

Excerpt: “If you’re new to transformer encoder models like BERT, this is a good learning experience. … I got inspiration for this piece from Fabian Ridder as he was using ChatGPT to identify clickbait and factual articles to train a model using FastText. … The model we’re building will use synthetic data rather than the real thing, though. … As we’re using binary classes, i.e., clickbait or factual, we will be able to achieve 99% accuracy. … While transformers have introduced amazing capabilities in text generation, they have also offered improvements within other NLP tasks, such as text classification and extraction. The distinction between model architectures is a bit blurry but it’s useful to understand that different transformer models were originally built for different tasks. A decoder model takes in a smaller input and outputs a larger text. GPT, which introduced impressive text generation back when, is a decoder model. A decoder primarily focuses on generating the next sentence rather than look at the text as a whole. While larger language models offer more nuanced capabilities today, decoders were not built for tasks that involve extraction and labeling. For these tasks, we can use encoder models, which take in more input and provide a condensed output. Encoders excel at extracting information by looking at the entire input at once to create a representation and thus are great at analyzing input data in its entirety.”

2. UX is dead. Long live UX. – Patrick Neeman, UX Collective

Excerpt: “I’m of the same opinion as Keith Ford about the future of UX: we are entering a golden age, but it will require change of the habits we have. … Every downturn has a different flavor, but they all have base ingredients. The original flavors of this one are COVID and eventually, the effect of Artificial Intelligence changes how we do our jobs. … Free money leads to overhiring. At the end of the COVID-19 pandemic, many companies did just that, a decision that later proved problematic. … The reality is that the talent pool isn’t meeting our needs. That’s not at all related to the designer oversupply; it’s a skills gap. There will be too many designers in the field for a while, but in the end, it will lead to a stronger field as designers close the talent gap in their effort to break in. … The age of the Figma or Sketch jockey are over. We have to embrace our contribution to the business more and learn more about the problems we’re solving. … We got paid too much for too long for skills that don’t contribute to the business other than being told what to do. … The more we talk about craft, the more we put ourselves at risk from being cut because it’s really hard to quantify craft.”

3. RAG Explained | Using Retrieval-Augmented Generation to Build Semantic Search – Yong Sheng Tan, Orkes

Excerpt: “With more and more LLMs released as open source and deployable as on-premise private models, it became possible for organizations to train, fine-tune, or supplement models with private data. RAG (retrieval-augmented generation) is one such technique for customizing an LLM, serving as a viable approach for businesses to use LLMs without the high costs and specialized skills involved in building a custom model from scratch. … RAG (retrieval-augmented generation) is a technique that improves the accuracy of an LLM (large language model) output with pre-fetched data from external sources. With RAG, the model references a separate database from its training data in real-time before generating a response. RAG extends the general capabilities of LLMs into a specific domain without the need to train a custom model from scratch. This approach enables general-purpose LLMs to provide more useful, relevant, and accurate answers in highly-specialized or private contexts, such as an organization’s internal knowledge base. For most use cases, RAG provides a similar result as training custom models but at a fraction of the required cost and resources. … First, the data is chunked and transformed into embeddings, which are vector representations of the data. These embeddings are then indexed into a vector database with the help of an AI algorithm known as embedding models. Once the data is populated in the index, natural language queries can be performed on the index using the same embedding model to yield relevant chunks of information. These chunks then get passed to the LLM as context, along with guardrails and prompts on how to respond given the context. … RAG can be used for various knowledge-intensive tasks: Question-answering systems, Knowledge base search engine, Document retrieval for research, Recommendation systems, Chatbots with real-time data.”

4. GraphRAG: Elevating RAG with Next-Gen Knowledge Graphs – Converge Technology Solutions

Excerpt: “Despite significant progress, challenges remain, such as the ‘hallucination’ phenomenon where inaccurate information is provided. This often results from misinterpreting user intent, leading to irrelevant data retrieval. Consequently, this issue causes hesitation in the enterprise world, preventing full adoption of the technology. Efforts to improve involve four main approaches: Building models from scratch: Ensures clear data context but is costly. Fine Tuning: Cost-effective and accurate but difficult to balance general and domain-specific contexts and as constantly updated data. Adding additional context to queries: Cost-effective but risks subjectivity and bias. Providing extra context during responses: Allows for up-to-date, cost-effective responses but is complex to implement. … Baseline RAG was created to address the hallucination issue in a cost effective manner, but there are situations where it underperforms: Connecting the dots: Baseline RAG struggles to traverse disparate pieces of information to provide synthesized insights. Holistic understanding: It performs poorly when asked to comprehend summarized semantic concepts over large data collections or even single extensive documents. To address these issues, the tech community is developing methods to extend and enhance RAG. Microsoft Research’s new approach, GraphRAG, uses large language models to create a knowledge graph based on private datasets. … Knowledge graphs have been around for a while, with Google’s knowledge graph being the most notable example. … Historically, knowledge graphs were used to unveil hidden relationships across different data silos, requiring a time-consuming manual process that only big companies invested in due to the effort and domain expertise required. … However, with the rise of LLMs, knowledge graphs have taken a new turn. Now, LLM RAG systems use knowledge graphs differently. They create explicit connections between terms, reducing hallucinations, adding context, and providing memory and personalization for LLMs. This makes RAG systems enterprise-ready and automates knowledge graph creation, increasing their relevance and accessibility. … Knowledge graphs in RAG serve as: Data stores for information retrieval. Semantic structures for retrieving vector chunks. … By combining entity descriptions with their properties and relationships, GraphRAG facilitates deeper insights and better comprehension of specific domains.”

5. Figma’s new branding is designed for growth mode – Lilly Smith, Fast Company

Excerpt: “Figma’s been making moves to expand beyond its founding idea of being being a single product company for designers, to a multi-product company for multi-role creative teams. Now, the company’s refreshed brand is catching up and speaking to an expanded audience that includes developers and supporting team members like project managers, who help bring a design deliverable to life. … It’s using copy and visual references that connect the product to broader audiences, and in turn, the company with major growth opportunity. … While their original concepts all touch different points of creative processes, Adobe, Zoom, Slack, Notion, Canva, WordPress, Wix, Squarespace and, well, pretty much any creative platform you use for one thing wants to be the one platform your team uses for everything, as a way to keep users in-product. … This is clearer than ever with Figma’s brand refresh. The brand introduced its own bespoke typeface with Grilli Type, Figma Sans. It’s a pared-down version of the typeface the company used before, which had super exaggerated inkwells. The new wordmark uses the same typeface, which slims its once super round, humanist sans counters that were popular among aughts-era tech startups. They designed both to be a somewhat neutral, flexible counterpart to its otherwise colorful brand.“

6. 10 Reasons Why People Tune You Out – Nuala Walsh, Inc.

Excerpt: “So why do people tune you out when you need them to tune in? Is it you, or is it them? It could be either — and you need to know which, whether pitching to investors, giving a keynote, or interviewing for a job. … Your Message Is Irrelevant. … You Interrupt. … You Contradict or Criticize. … You Ramble. … You Brag. … They Lack Incentive. … They See You as an Outsider. … They Don’t Like You. … They Dislike Your Message. … They Doubt Your Credibility.”

7. Why Content Marketing Doesn’t Have to Be Boring: How to Create Engaging Content – Jason Hennessey, Rolling Stone

Excerpt: “… it is probably much easier to get customers excited to buy new gear for rock climbing than to come in for their biannual dental exam, but both types of businesses can benefit from content marketing. Even if your business is seen by some as boring, that doesn’t mean your content marketing has to be. Here’s the key: If you think of content marketing itself as boring, or if your marketing team views your business and product or service as boring, you will have boring, ineffective content. Many businesses end up following in the footsteps of their competitors when it comes to marketing and branding and don’t think outside the type of content others within their industry are publishing. But you don’t necessarily have to publish exciting content to stand out … Set a Vision for Content Marketing Outcomes … Get Your Marketing Team on Board With Your Vision … Keep Audience First … Create Quality Content … To avoid writing boring content, make sure that all of your marketing content checks the following boxes: Timely … Useful … Engaging. … Ultimately, it’s the human element that takes content from boring to engaging. In a world where things seem increasingly automated, people crave human connection; if you can connect with them using your business’s purpose to solve their problem and do it with a personal touch, you can create engaging content while building your business.”

8. KnowFormer: A Transformer-Based Breakthrough Model for Efficient Knowledge Graph Reasoning, Tackling Incompleteness and Enhancing Predictive Accuracy Across Large-Scale Datasets – Asif Razzaq MarkTech Post

Excerpt: “Knowledge graphs (KGs) are structured representations of facts consisting of entities and relationships between them. These graphs have become fundamental in artificial intelligence, natural language processing, and recommendation systems. By organizing data in this structured way, knowledge graphs enable machines to understand and reason about the world more efficiently. This reasoning ability is crucial for predicting missing facts or inferences based on existing knowledge. KGs are employed in applications ranging from search engines to virtual assistants, where the ability to draw logical conclusions from interconnected data is vital. One of the key challenges with knowledge graphs is that they are often incomplete. … Path-based methods, which attempt to infer missing facts by examining the shortest paths between entities, are especially prone to incomplete or oversimplified paths. Moreover, these methods often face the problem of “information over-squashing,” where too much information is compressed into too few connections, leading to inaccurate results. … Current approaches to addressing these issues include embedding-based methods that convert the entities and relations of a knowledge graph into a low-dimensional space. These techniques, like TransE, DistMult, and RotatE, have successfully preserved the structure of knowledge graphs and enabled reasoning. However, embedding-based models have limitations. They often fail in inductive scenarios where new, unseen entities or relationships must be reasoned about, as they cannot effectively leverage the local structures within the graph. … Researchers from Zhongguancun Laboratory, Beihang University, and Nanyang Technological University introduced a new KnowFormer model, which utilizes transformer architecture to improve knowledge graph reasoning. This model shifts the focus from traditional path-based and embedding-based methods to a structure-aware approach. KnowFormer leverages the transformer’s self-attention mechanism, which enables it to analyze relationships between any pair of entities within a knowledge graph. … By utilizing a query-based attention system, KnowFormer calculates attention scores between pairs of entities based on their connection plausibility, offering a more flexible and efficient way to infer missing facts. … The KnowFormer model incorporates both a query function and a value function to generate informative representations of entities. The query function helps the model identify relevant entity pairs by analyzing the knowledge graph’s structure, while the value function encodes the structural information needed for accurate reasoning.”

9. Your Marketing Strategy Needs an Overhaul — This Approach Is What Separates Successful Campaigns From the Rest – Kelly Fletcher, Entrepreneur

Excerpt: “… it’s time to rethink your approach because the new generation of data generation is here, and it’s transforming the way brands connect with their audiences. … Picture this: 87% of marketers say data is their company’s most under-utilized asset, yet only 20% of marketing spend is data-driven. … We thought we had the content nailed down — funny posts, clever memes, etc. But the data told a different story. Our initial content strategy didn’t resonate as strongly as we anticipated. The audience wasn’t engaging with the funny posts and clever memes we had planned. Instead, they craved authenticity — real pets using real products. It was a shift we hadn’t anticipated, but one that made all the difference. By partnering with pet influencers and leveraging user-generated content, we tapped into a goldmine of engagement. … This example underscores the importance of being agile and responsive to data. … The harsh truth is that many don’t have the data infrastructure to support it. According to a Harvard Business Review report, only 31% of businesses have a single, 360-degree view of customer data. That’s a major roadblock. … The good news is that those who invest in the right tools and strategies can unlock incredible potential. … The future of marketing and PR lies in this new generation of data generation. It’s about more than just collecting data — it’s about transforming it into actionable insights that drive real results. Whether it’s through social media analytics, media relations metrics or advanced AI tools, the companies that get it right will be the ones leading the charge. … The landscape is shifting, and the days of gut-feeling marketing are numbered.”

10. RIP RAG. RIG is Here – Sagar Sharma, Analytics India Magazine

Excerpt: “Google released its new Gemma model – DataGemma. While the world is experimenting with RAG to reduce hallucinations and increase accuracy, Google decided to use RIG (retrieval interleaved generation), a technique that integrates LLMs with Data Commons, an open-source database of public data. … Typically, when a user asks a question, the AI model begins answering based on its existing knowledge. However, with RIG, if the AI model encounters a need for more current or specific data, it pauses to search for this information from reliable external sources like databases or websites. The model then seamlessly incorporates this newly acquired data into its response, alternating between generating content and retrieving information as needed. This approach enables the AI to provide more up-to-date and accurate answers, especially for topics involving rapidly changing information such as current events or recent statistics. Unlike RAG, which performs retrieval once before generating the answer, RIG can adapt in real time while generating responses. This allows the model to refine its output iteratively as it retrieves new information. … One of the most important concepts when talking about RIG is Data Commons, which is described as “a publicly available knowledge graph containing over 240 billion rich data points across hundreds of thousands of statistical variables”. This vast repository of data serves as the foundation for grounding AI responses in factual information. The data in Data Commons comes from reputable organisations such as the United Nations (UN), the World Health Organisation (WHO), the Centers for Disease Control and Prevention (CDC), and the Census Bureau. This ensures that the information used is reliable and trustworthy.”

11. Meta Releases Llama 3.2—and Gives Its AI a Voice – Will Knight, Wired

Excerpt: “The more important upgrade for Meta’s long-term ambitions, though, is the new ability of its models to see users’ photos and other visual information. Meta today also announced Llama 3.2, the first version of its free AI models to have visual abilities, broadening their usefulness and relevance for robotics, virtual reality, and so-called AI agents. Some versions of Llama 3.2 are also the first to be optimized to run on mobile devices. This could help developers create AI-powered apps that run on a smartphone and tap into its camera or watch the screen in order to use apps on your behalf. “This is our first open source, multimodal model, and it’s going to enable a lot of interesting applications that require visual understanding,” Zuckerberg said on stage at Connect, a Meta event held in California today. … Meta has lately given its AI a more prominent billing in its apps—for example, making it part of the search bar in Instagram and Messenger. … The new version of Meta AI will also be able to provide feedback on and information about users’ photos; for example, if you’re unsure what bird you’ve snapped a picture of, it can tell you the species. And it will be able to help edit images by, for instance, adding new backgrounds or details on demand. … Powering Meta AI’s new capabilities is an upgraded version of Llama, Meta’s premier large language model. The free model announced today may also have a broad impact, given how widely the Llama family has been adopted by developers and startups already. In contrast to OpenAI’s models, Llama can be downloaded and run locally without charge—although there are some restrictions on large-scale commercial use. Llama can also more easily be fine-tuned, or modified with additional training, for specific tasks. … Large language models are increasingly becoming “multimodal,” meaning they are trained to handle audio and images as input as well as text. This extends a model’s abilities and allows developers to build new kinds of AI applications on top of it, including so-called AI agents capable of carrying out useful tasks on computers on their behalf.”

12. Graph RAG: Bring the Power of Graphs to Generative AI – Melliyal Annamalai, Oracle

Excerpt: “But how about data the LLM has not been trained on, such as your own business data? How can we use the power of generative AI on such data? We have the answer – Graph RAG. … Retrieval Augmented Generation (RAG or Vector RAG) can enhance the user query to an LLM (the prompt) by searching business data and adding relevant information to the user query. A user query in human language can be converted to a vector and used to search business data in a database. The user query enhanced with results from the vector search in the database (the enhanced prompt) can be input to the LLM which can then output useful results in human language even though it was not trained on that data. Graph RAG adds a new dimension to enhancing the prompt by adding information on how entities in a user query are connected to other data entities (a graph). Graphs capture information on how data is related, which is difficult to capture with other data models. … Graphs are powerful as they help capture and navigate relationships in data, which in this example is the relationship between customers and movies and between customers when they watch movies together. Instead of a generic recommendation using demographic information, MovieStream can make more informed recommendations based on the movie-watching preferences of a customer’s social network. … In addition to combining Operational Property Graphs and RAG, as we’ve just described, there are several other impactful uses of graphs when developing applications with generative AI: Multi-modal Connections in Data … Knowledge Graph for Improved Information Retrieval.”

Thanks for checking out the new Hamsterdam! 🐹

Until next time, enjoy the vibes:

Thanks for reading. Happy marketing! 🤗