In the Weights: New AI Tool Measures Your 'Existence' Within LLM Training Data

In an era where personal identity is increasingly shaped by digital footprints, the traditional 'vanity search' on platforms like Google is beginning to feel outdated. As large language models (LLMs) become primary conduits for information retrieval, a new question emerges: how does artificial intelligence 'remember' us? This evolving landscape has prompted two former OpenAI engineers to launch 'In the Weights,' a novel tool designed to quantify an individual's digital imprint within the neural networks of leading AI models, offering a fresh, and perhaps unsettling, measure of online prominence.

Quick summary

• 'In the Weights' is a new platform developed by former OpenAI engineers Thomas Dimson and Joey Flynn.
• The tool assesses how well major AI models recall information about individuals directly from their training data, without using external web searches.
• It queries various LLMs, including Grok, Gemini, GPT versions, Claude, and Llama, to generate a 'strength score' reflecting a person's embeddedness in AI's 'weights.'
• The platform highlights discrepancies in AI recall, identifies potential hallucinations, and reveals which models associate what information with a given name.

Why it matters

The emergence of 'In the Weights' carries significant implications across several domains. For individuals, it introduces a new, AI-centric metric for digital identity and influence, potentially fostering a competitive desire to be 'remembered' by superintelligence. It redefines what it means to be 'online' or 'known,' shifting the focus from web search visibility to internal AI knowledge representation. This could lead to a psychological impact, as people grapple with the idea of their existence being encoded in floating-point numbers within a machine's 'brain,' and the tantalizing, albeit perhaps specious, notion of 'digital immortality.'

For AI developers and researchers, the tool offers a valuable diagnostic lens. By revealing how different models recall information and where they might hallucinate, 'In the Weights' provides insights into the biases, limitations, and internal knowledge structures of these powerful systems. It prompts deeper investigation into training data composition, ethical considerations regarding personal data representation, and the accuracy of AI's 'memory.' Understanding these nuances is crucial for developing more reliable, fair, and transparent AI.

More broadly, 'In the Weights' underscores a fundamental shift in how information is accessed and validated. As more traffic moves to LLMs, the authority of traditional web search diminishes. This platform serves as a stark reminder that our digital selves are no longer just what can be found via a hyperlink, but also what has been internalized and synthesized by algorithms, paving the way for new forms of information access and potential challenges in validating AI-generated facts.

Background

The genesis of 'In the Weights' is rooted in the evolving landscape of information and the personal experiences of its creators, Thomas Dimson and Joey Flynn. Both engineers had previously joined OpenAI through the acquisition of their design startup, Global Illumination. After their departure from OpenAI, they sought to reignite their creative spark, leading them to ponder the shifting paradigms of digital relevance.

Dimson articulated a growing sentiment that traditional Google vanity searches, once a prime objective for gauging personal notoriety, were becoming less relevant in 2026. He observed a noticeable migration of information traffic towards large language models, implying that the 'canonical source of information' was no longer solely the open web but increasingly the internal parameters of AI. This realization was compounded by the profound thought that 'so many lives are encoded somehow in a bunch of floating point numbers inside the AI brain.'

The concept for 'In the Weights' was further solidified by a tongue-in-cheek blog post that explored the themes of AI weights and referenced Terry Bisson's classic short story 'They’re Made Out of Meat.' This confluence of professional experience, a keen observation of technological trends, and a philosophical reflection on digital existence ultimately steered Dimson and Flynn towards creating a tool that could shed light on this nascent form of AI-centric identity.

Decoding Digital Footprints in AI

At its core, 'In the Weights' operates on a fascinating premise: measuring an AI model's internal 'memory' of an individual. The 'weights' in question refer to the numerical parameters that define an AI model's training and output. Essentially, the website attempts to quantify how well an AI model can 'recall' someone without resorting to external tools like traditional web search engines.

The process is deceptively simple yet technically complex. Users input a name, and 'In the Weights' then queries a diverse array of prominent language models. These include widely recognized systems like Grok, Gemini, various iterations of GPT, Claude, and Llama, alongside lesser-known models. Each model is prompted with a query akin to, 'Who is <name>? Give up to 10 results, each with a short description and confidence.'

Upon receiving the responses, the platform employs a clustering algorithm to group similar descriptions and assign a 'strength score.' This score purports to indicate the depth of an individual's presence within the AI's learned parameters. For instance, one TechCrunch reporter, a self-proclaimed 'humble tech blogger,' received a strength score of 641, placing him in the top 6% of names. However, this initial gratification was quickly tempered by the discovery that several colleagues scored even higher, and the leaderboard remains dynamic, showcasing figures like Macaulay Culkin and Luciano Pavarotti vying for the top spot with scores near 988.

Beyond the Score: Unpacking AI's Memory

The utility of 'In the Weights' extends beyond a mere numerical ranking. The platform also provides granular insights into which specific AI models returned particular answers for a given name. This feature is crucial for understanding the nuances of AI recall and identifying potential pitfalls, such as hallucinations.

A notable example involved GPT-5.4 Mini, which, when queried about TechCrunch's weekend editor Anthony Ha, reportedly responded that his was an 'ambiguous name form that could refer to multiple people with the initials A.H.A.' Such discrepancies not only highlight the limitations of current AI models but also offer a glimpse into the varying degrees of 'knowledge' and interpretation held by different systems.

The ability to pinpoint specific model behaviors allows users to observe the 'personality' of each AI in its response. It demonstrates that not all LLMs possess the same information, or interpret it in the same way, even when trained on vast datasets. This layer of transparency, while still opaque in terms of the underlying data, is invaluable for those seeking to understand how their digital identity is fractured and reassembled across the AI landscape.

The Creators' Vision and Public Reception

Thomas Dimson revealed that the motivation behind 'In the Weights' stemmed from a desire to stimulate creativity after leaving OpenAI. He and Joey Flynn recognized that the traditional metric of 'vanity searches' on Google was becoming obsolete as more internet traffic shifted towards large language models. Dimson posed a provocative question: if human lives are increasingly encoded within the 'floating point numbers inside the AI brain,' then how can one measure that encoding?

The site's underlying philosophy, encapsulated by its tagline, 'Being in the weights means your existence was deemed important in the process of creating superhuman artificial intelligence,' clearly resonated with the public. Dimson described the reception as 'insane,' indicating that what they anticipated would be a 'mild curiosity' had instead 'struck a nerve.' The inherent human desire to ascertain one's place within this emerging 'super intelligence,' coupled with the addictive comparison factor of the strength scores, fueled its rapid popularity.

Despite skepticism from some AI critics, who liken the process to merely asking 13 chatbots about oneself, the appeal of 'In the Weights' is undeniable. Its 'Nintendo-inspired retro design' also contributes to its user-friendly and engaging interface, making the complex concept of AI recall accessible and even enjoyable for a broad audience.

Future Directions and Unanswered Questions

Looking ahead, Dimson has articulated clear intentions to deepen the analytical capabilities of 'In the Weights.' He plans to delve further into the reasons behind varying results from different models within the same series. This line of inquiry could illuminate subtle distinctions in model architectures, training methodologies, or the specific datasets they were exposed to.

Furthermore, Dimson aims to investigate whether certain models exhibit biases towards particular types of people. This critical examination of AI bias is paramount, as understanding how different demographics are represented (or misrepresented) within AI's 'memory' can inform efforts to build more equitable and inclusive technologies. Finally, the platform may also evolve to identify individuals who, based on their embeddedness in AI weights, 'should have a Wikipedia article but don’t,' effectively highlighting gaps in traditional knowledge repositories that AI might unwittingly fill.

While the concept of 'digital immortality' through AI recall remains a subject of philosophical debate, 'In the Weights' has undeniably tapped into a profound human curiosity. It challenges our understanding of self-representation in the digital age, prompting a deeper look into the opaque yet increasingly influential world of artificial intelligence and its perception of our collective human story.

Qnews24h insight

The launch of 'In the Weights' serves as a crucial signal of a paradigm shift in how digital prominence and legacy are perceived. No longer is visibility solely about search engine rankings or social media presence; it's increasingly about an individual's intrinsic imprint within the vast, complex 'memory' of artificial intelligence models. This platform deftly leverages human vanity to expose a profound aspect of AI's internal workings: its inherent understanding, or misunderstanding, of humanity derived from its training data.

While presented with a playful, retro aesthetic, the tool implicitly raises serious questions about data privacy, representational bias, and the ethical implications of AI's 'knowledge' about individuals. It highlights the opaque nature of massive AI training sets and how they might inadvertently perpetuate or amplify societal biases. 'In the Weights' could transition from being a mere 'vanity search' into an essential diagnostic instrument for auditing AI models for fairness, accuracy, and the comprehensive, unbiased encoding of human existence.

Sources

• TechCrunch

FAQ

What is 'In the Weights'?
'In the Weights' is a new online platform created by former OpenAI engineers Thomas Dimson and Joey Flynn. It allows users to search for names and discover how deeply and accurately major AI language models (LLMs) like GPT, Gemini, and Claude 'remember' or recall information about individuals from their internal training data, without using traditional web search.

How does 'In the Weights' measure a person's 'strength score'?
The tool queries various AI models with questions like 'Who is <name>?' and asks for up to 10 descriptive results with confidence levels. It then clusters similar descriptions and assigns a 'strength score,' which indicates how prominently and consistently a person appears in the AI models' 'weights' or internal parameters.

Who created 'In the Weights' and why?
'In the Weights' was created by Thomas Dimson and Joey Flynn, both former engineers at OpenAI who had joined through an acquisition. They developed the tool to explore how digital identity is evolving beyond traditional web searches, recognizing that more information traffic is moving to LLMs and that human lives are increasingly encoded within AI's internal 'brains.'

What are the future plans for 'In the Weights'?
The creators plan to further investigate why different AI models return varying results for the same individual. They also intend to analyze if models exhibit biases towards certain types of people and to identify individuals who are significantly 'in the weights' of AI but may lack traditional recognition, such as a Wikipedia article.