Google Testing AI Tool That Finds & Rewrites Quality Content via @sejournal, @martinibuster

Google is paying small publishers to use their AI platform to rewrite the content published on news sites that are targeted by Google’s platform for largescale content parasitism. People on Twitter express unfavorable opinions about it.

Google is paying small independent publishers to use a generative AI platform that targets the content of other news content for summarization and republication. According to a paywalled article published on AdWeek, the tool is a beta test that requires the publishers to post a required number of articles per day in order to receive their payment.

Small publishers use a tool that shows them content chosen by the AI tool that can be selected in order to produce content at “at no cost.”

Why Google’s New Tool Is Problematic

It’s common for a news organization to pick up on news that’s broken by another news organization and if it’s done by a capable journalist they put their own spin on it. It’s the way things are done.

Google’s tool however appears to resemble a programmatic way to plagiarize content called article spinning. Article spinning is an automated tactic that uses website feeds to input published content from other sites that is then rewritten by the computer program, typically by replacing words with synonyms. AI however can spin content with greater nuance, essentially summarizing the content in a different tone by replacing entire sentences and paragraphs with content that’s the same as the original but expressed in a different manner.

What’s different is that this tool is something that Google itself is testing and that’s problematic not just because Google is the de facto gatekeeper for online content, the tool places a target on specific news organizations to have their content spun by small independent publishers.

On the one hand that could be a good thing because it could generate inbound links to the original publisher of the news. Free links, that’s a win-win, right?

But it’s not because news publishers don’t really benefit from links to content that has a shelf life of roughly 48 hours, at most. News is a hamster wheel of constant news publication in the service of keeping the wheel spinning to keep the business afloat. It’s a nonstop process that can easily be undermined by wholesale content dilution.

And that’s really at the hear of the problem with Google’s AI tool in that it dilutes the value that an organization creates by hiring professionals to create the “value add” content that Google frequently says it wants to publish. And that is what makes Google’s AI tool hypocritical at best and cynical at worst because Google is encouraging the creation of high quality content while simultaneously undermining it.

An army of publishers programmatically copying every published news article doesn’t look so good to the original publisher, especially if their content is overwhelmed by the parasitic AI in Google News, in the search results and by user preference for their local online news publisher who is republishing the news from the bigger publishers.

Reaction To Google’s AI News Tool

Technology journalist Brian Merchant (who writes for The Atlantic and has published a book) posted a virtual thumbs down on Twitter, a sentiment that was unanimously seconded.

He tweeted:

“The nightmare begins — Google is incentivizing the production of AI-generated slop.

If you are a news outlet who has accepted this meager deal, and especially if you are publishing AI-generated articles without disclaimers, you should be deeply ashamed.”

Brian followed up with this tweet with his observation of what’s wrong with Google’s AI tool for small news sites:

“If we in the media have learned ANYTHING from the last 10 years it is that we do not in fact have to settle for whatever scraps big tech throws us, and in fact it will screw us in the end—why would you participate in automating your field out of existence for like $30k a year???”

Merchant retweeted a comment by technology journalist Alex Kantrow:

“This is sad. Is this the web Google wants?”

Another person tweeted:

“It’s all about unchecked and untaxed profit.

Of course the is what Google wants – corporations selling out their employees and Americans in general for a quick buck.”

A person who works for Microsoft invoked the concept of “autophagy” which when an organism begins consuming itself such as when it is starving.

She tweeted:

“The quality of news content will decline and hurt search. Autophagy is a real threat to information quality and it looks like no one is taking it seriously.”

The Future Of Content

This isn’t just a “news” problem, it’s a problem for everyone who earns a living publishing online content. What can be used for News publications can easily be adapted for product reviews, recipes, entertainment and virtually any topic that affiliates publish content on.

How do you feel about Google’s new tool? Will it help small publishers compete against bigger sites or is it just the onset of autophagy in the body of online publishing?

Does AI Democratize SEO Or Amplify Incompetence? via @sejournal, @martinibuster

Generative AI has introduced significant shifts in the SEO industry with some experiencing success integrating SEO into the daily workflow while others not so much. How is AI best used in SEO tasks, and what are its capabilities and limitations and can the subjective nature of SEO negatively affect the outcome for SEO?

AI-Automated SEO

There are some tasks that an AI can do reasonably well, like creating meta description tags and title elements at scale.

But the utility of AI becomes questionable when it comes to other aspects of search optimization.

AI has been put into service analyzing search engine results pages (SERPs), keyword research, content scoring based on keyword use, keyword based article creation, competitive research, as well as just creating content based on keywords.

It’s an act of faith to trust that the people behind the software understand SEO. But who do you trust if SEO is so subjective that people can’t even agree on the proper use of title tags and headings?

Even the concept of SERP analysis can go south depending on the experience of the person doing it. For example, there is a school of thought that the way to beat the competition is to understand why their content is ranking and then use those data point as the ingredients for creating an exact copy that is better, which is oxymoronic.

Obviously, you can’t make something better by making an exact copy of it that’s better. Yet, that’s the underlying logic of the Skyscraper Content Tactic that (ironically) is a copycat of the 10X Content Tactic, which are popular with those who are new to SEO. And as absurd as that tactic is, it’s at the heart of how some AI tools execute SERP analysis.

Clearly, some AI tools can amplify the inexperience of those who created the tools as well as those who use AI tools.

Julia McCoy, president of Content at Scale (LinkedIn) agrees.

She explained:

“AI is absolutely the most incredible advancement of technology that we’ve seen in the last 200 years.

We’re seeing a ton of AI tools designed for content optimization or writing generation that offer incredible efficiencies—they can streamline processes, give you powerfully detailed insights for optimization and ranking improvements, and even generate entire articles that are nearly ready to publish.

But, you’ve got to know how to use them. And you need to know who built them.

I think it’s crucial to acknowledge: no tool can transform an amateur into an expert overnight. Just as Malcolm Gladwell’s skills stem from years of honing his craft—not a tool that landed in his lap overnight—the path from budding learner to seasoned, proficient expert requires time, experience and a deep knowledge of the industry.

While AI has democratized access to advanced techniques making higher-level strategies accessible—it cannot instill wisdom nor insight where there is none. It amplifies capabilities, but also shortcomings. We need to remember that human intuition is complemented by technology, not replaced by it.”

AI Amplifies The User’s SEO Skill

Why is it that some people have success with AI and others do not? In my opinion, AI is just a tool like a paint brush. The talent and skill belongs to the person, not the tool.

A less experienced SEO will analyze a webpage by extracting the keywords from the content, the headings and the title tag. A more experienced SEO will analyze the webpage by understanding what questions it answers.

The importance of skill and experience is evident with AI image generators where some users are able to create amazingly lifelike works of are while others make images of people with seven fingers on each hand.

Does AI Democratize SEO?

There is an idea that AI can empower an SEO beginner to perform at the same level as someone with decades of experience but that’s not how AI works right now, as Julia suggested earlier.

I asked Brenda Malone, an SEO Technical Strategist and Web Developer (LinkedIn), for her opinion on AI and the potential for democratizing SEO.

Brenda shared:

“I don’t necessarily think it will totally democratize the SEO discipline as it exists today.

I think the over-abundance of AI SaaS tools will serve to overwhelm the inexperienced SEO professionals, while further empowering experienced SEO professionals who know how to exploit specific AI tools in order to make more qualified human analyses.

What I think AI’s effect on the SEO industry for the short-term will turn out to be is a decrease in the number of professionals needed because a lot of the data-gathering will be automated.

Current SEO professionals cannot afford to be Luddites, and should instead dig deep into AI to identify tasks that are related to SEO activities and develop analysis specializations because the days of getting away with merely implementing meta and title tags for ranking are gone.”

The ability of AI to amplify makes a person more efficient. For example, deep analysis of data is a snap for AI. But it’s also great for tedious tasks like performing an analysis, generating a bullet point of major takeaways from the data then creating a presentation from that data.

Takeaways

Generative AI like ChatGPT and Claude can have a significant impact on SEO, amplifying what can be done and streamlining the workflow. However it also amplifies the shortcomings of its users. Experienced SEOs can leverage generative AI to enhance their work, while those who are new to SEO might not experience the full potential, although they may benefit from SaaS tools depending on the experience of the publishers of those tools.

• Generative AI amplifies the user’s SEO skill and experience
• Generative AI may not necessarily democratize SEO
• But SaaS AI tools can benefit users who are new to SEO
• Review the founders and creators of SaaS SEO tools to understand their experience and skill levels
• AI is the future, don’t be a Luddite

Featured Image by Shutterstock/Krakenimages.com

Google Announces Gemma: Laptop-Friendly Open Source AI via @sejournal, @martinibuster

Google released an open source large language model based on the technology used to create Gemini that is powerful yet lightweight, optimized to be used in environments with limited resources like on a laptop or cloud infrastructure.

Gemma can be used to create a chatbot, content generation tool and pretty much anything else that a language model can do. This is the tool that SEOs have been waiting for.

It is released in two versions, one with two billion parameters (2B) and another one with seven billion parameters (7B). The number of parameters indicates the model’s complexity and potential capability. Models with more parameters can achieve a better understanding of language and generate more sophisticated responses, but they also require more resources to train and run.

The purpose of releasing Gemma is to democratize access to state of the art Artificial Intelligence that is trained to be safe and responsible out of the box, with a toolkit to further optimize it for safety.

Gemma By DeepMind

The model is developed to be lightweight and efficient which makes it ideal for getting it into the hands of more end users.

Google’s official announcement noted the following key points:

• “We’re releasing model weights in two sizes: Gemma 2B and Gemma 7B. Each size is released with pre-trained and instruction-tuned variants.
• A new Responsible Generative AI Toolkit provides guidance and essential tools for creating safer AI applications with Gemma.
• We’re providing toolchains for inference and supervised fine-tuning (SFT) across all major frameworks: JAX, PyTorch, and TensorFlow through native Keras 3.0.
• Ready-to-use Colab and Kaggle notebooks, alongside integration with popular tools such as Hugging Face, MaxText, NVIDIA NeMo and TensorRT-LLM, make it easy to get started with Gemma.
• Pre-trained and instruction-tuned Gemma models can run on your laptop, workstation, or Google Cloud with easy deployment on Vertex AI and Google Kubernetes Engine (GKE).
• Optimization across multiple AI hardware platforms ensures industry-leading performance, including NVIDIA GPUs and Google Cloud TPUs.
• Terms of use permit responsible commercial usage and distribution for all organizations, regardless of size.”

Analysis Of Gemma

According to an analysis by an Awni Hannun, a machine learning research scientist at Apple, Gemma is optimized to be highly efficient in a way that makes it suitable for use in low-resource environments.

Hannun observed that Gemma has a vocabulary of 250,000 (250k) tokens versus 32k for comparable models. The importance of that is that Gemma can recognize and process a wider variety of words, allowing it to handle tasks with complex language. His analysis suggests that this extensive vocabulary enhances the model’s versatility across different types of content. He also believes that it may help with math, code and other modalities.

It was also noted that the “embedding weights” are massive (750 million). The embedding weights are a reference to the parameters that help in mapping words to representations of their meanings and relationships.

An important feature he called out is that the embedding weights, which encode detailed information about word meanings and relationships, are used not just in processing input part but also in generating the model’s output. This sharing improves the efficiency of the model by allowing it to better leverage its understanding of language when producing text.

For end users, this means more accurate, relevant, and contextually appropriate responses (content) from the model, which improves its use in conetent generation as well as for chatbots and translations.

He tweeted:

“The vocab is massive compared to other open source models: 250K vs 32k for Mistral 7B

Maybe helps a lot with math / code / other modalities with a heavy tail of symbols.

Also the embedding weights are big (~750M params), so they get shared with the output head.”

In a follow-up tweet he also noted an optimization in training that translates into potentially more accurate and refined model responses, as it enables the model to learn and adapt more effectively during the training phase.

He tweeted:

“The RMS norm weight has a unit offset.

Instead of “x * weight” they do “x * (1 + weight)”.

I assume this is a training optimization. Usually the weight is initialized to 1 but likely they initialize close to 0. Similar to every other parameter.”

He followed up that there are more optimizations in data and training but that those two factors are what especially stood out.

Designed To Be Safe And Responsible

An important key feature is that it is designed from the ground up to be safe which makes it ideal for deploying for use. Training data was filtered to remove personal and sensitive information. Google also used reinforcement learning from human feedback (RLHF) to train the model for responsible behavior.

It was further debugged with manual re-teaming, automated testing and checked for capabilities for unwanted and dangerous activities.

Google also released a toolkit for helping end-users further improve safety:

“We’re also releasing a new Responsible Generative AI Toolkit together with Gemma to help developers and researchers prioritize building safe and responsible AI applications. The toolkit includes:

• Safety classification: We provide a novel methodology for building robust safety classifiers with minimal examples.
• Debugging: A model debugging tool helps you investigate Gemma’s behavior and address potential issues.
• Guidance: You can access best practices for model builders based on Google’s experience in developing and deploying large language models.”

Read Google’s official announcement:

Gemma: Introducing new state-of-the-art open models

Featured Image by Shutterstock/Photo For Everything

5 Questions Answered About The OpenAI Search Engine via @sejournal, @martinibuster

It was reported that OpenAI is working on a search engine that would directly challenge Google. But details missing from the report raise questions about whether OpenAI is creating a standalone search engine or if there’s another reason for the announcement.

OpenAI Web Search Report

The report published on The Information relates that OpenAI is developing a Web Search product that will directly compete with Google. A key detail of the report is that it will be partly powered by Bing, Microsoft’s search engine. Apart from that there are no other details, including whether it will be a standalone search engine or be integrated within ChatGPT.

All reports note that it will be a direct challenge to Google so let’s start there.

1. Is OpenAI Mounting A Challenge To Google?

OpenAI is said to be using Bing search as part of the rumored search engine, a combination of a GPT-4 with Bing Search, plus something in the middle to coordinate between the two .

In that scenario, what OpenAI is not doing is developing its own search indexing technology, it’s using Bing.

What’s left then for OpenAI to do in order to create a search engine is to devise how the search interface interacts with GPT-4 and Bing.

And that’s a problem that Bing has already solved by using what it Microsoft calls an orchestration layer. Bing Chat uses retrieval-augmented generation (RAG) to improve answers by adding web search data to use as context for the answers that GPT-4 creates. For more information on how orchestration and RAG works watch the keynote at Microsoft Build 2023 event by Kevin Scott, Chief Technology Officer at Microsoft, at the 31:45 minute mark here).

If OpenAI is creating a challenge to Google Search, what exactly is left for OpenAI to do that Microsoft isn’t already doing with Bing Chat? Bing is an experienced and mature search technology, an expertise that OpenAI does not have.

Is OpenAI challenging Google? A more plausible answer is that Bing is challenging Google through OpenAI as a proxy.

2. Does OpenAI Have The Momentum To Challenge Google?

ChatGPT is the fastest growing app of all time, currently with about 180 million users, achieving in two months what took years for Facebook and Twitter.

Yet despite that head start Google’s lead is a steep hill for OpenAI to climb.  Consider that Google has approximately 3 to 4 billion users worldwide, absolutely dwarfing OpenAI’s 180 million.

Assuming that all 180 million OpenAI users performed an average of 4 searches per day, the daily number of searches could reach 720 million searches per day.

Statista estimates that there are 6.3 million searches on Google per minute which equals over 9 billion searches per day.

If OpenAI is to compete they’re going to have to offer a useful product with a compelling reason to use it. For example, Google and Apple have a captive audience on mobile device ecosystem that embeds them into the daily lives of their users, both at work and at home. It’s fairly apparent that it’s not enough to create a search engine to compete.

Realistically, how can OpenAI achieve that level of ubiquity and usefulness?

OpenAI is facing an uphill battle against not just Google but Microsoft and Apple, too. If we count Internet of Things apps and appliances then add Amazon to that list of competitors that already have a presence in billions of users daily lives.

OpenAI does not have the momentum to launch a search engine to compete against Google because it doesn’t have the ecosystem to support integration into users lives.

3. OpenAI Lacks Information Retrieval Expertise

Search is formally referred to as Information Retrieval (IR) in research papers and patents. No amount of searching in the Arxiv.org repository of research papers will surface papers authored by OpenAI researchers related to information retrieval. The same can be said for searching for information retrieval (IR) related patents. OpenAI’s list of research papers also lacks IR related studies.

It’s not that OpenAI is being secretive. OpenAI has a long history of publishing research papers about the technologies they’re developing. The research into IR does not exist. So if OpenAI is indeed planning on launching a challenge to Google, where is the smoke from that fire?

It’s a fair guess that search is not something OpenAI is developing right now. There are no signs that it is even flirting with building a search engine, there’s nothing there.

4. Is The OpenAI Search Engine A Microsoft Project?

There is substantial evidence that Microsoft is furiously researching how to use LLMs as a part of a search engine.

All of the following research papers are classified as belonging to the fields of Information Retrieval (aka search), Artificial Intelligence, and Natural Language Computing.

Here are few research papers just from 2024:

Enhancing human annotation: Leveraging large language models and efficient batch processing
This is about using AI for classifying search queries.

Structured Entity Extraction Using Large Language Models
This research paper discovers a way to extracting structured information from unstructured text (like webpages). It’s like turning a webpage (unstructured data) into a machine understandable format (structured data).

Improving Text Embeddings with Large Language Models (PDF version here)
This research paper discusses a way to get high-quality text embeddings that can be used for information retrieval (IR). Text embeddings is a reference to creating a representation of text in a way that can be used by algorithms to understand the semantic meanings and relationships between the words.

The above research paper explains the use:

“Text embeddings are vector representations of natural language that encode its semantic information. They are widely used in various natural language processing (NLP) tasks, such as information retrieval (IR), question answering…etc. In the field of IR, the first-stage retrieval often relies on text embeddings to efficiently recall a small set of candidate documents from a large-scale corpus using approximate nearest neighbor search techniques.”

There’s more research by Microsoft that relates to search, but these are the ones that are specifically related to search together with large language models (like GPT-4.5).

Following the trail of breadcrumbs leads directly to Microsoft as the technology powering any search engine that OpenAI is supposed to be planning… if that rumor is true.

5. Is Rumor Meant To Steal Spotlight From Gemini?

The rumor that OpenAI is launching a competing search engine was published on February 14th. The next day on February 15th Google announced the launch of Gemini 1.5, after announcing Gemini Advanced on February 8th.

Is it a coincidence that OpenAI’s announcement completely overshadowed the Gemini announcement the next day? The timing is incredible.

At this point the OpenAI search engine is just a rumor.

Featured Image by Shutterstock/rafapress