TL;DR
In the Weights is a new website that queries multiple AI models to determine how well they can recall a person’s information, assigning a strength score. Created by ex-OpenAI employees, it aims to explore how AI ‘remembers’ individuals. The tool has gained rapid attention, but its significance and accuracy remain uncertain.
In the Weights, a new AI-driven vanity search tool, has launched, allowing users to see how well various AI models can recall their identities based on a name query. Developed by former OpenAI employees Thomas Dimson and Joey Flynn, the site assigns a ‘strength score’ indicating the model’s confidence in recalling a person’s information. This development matters because it shifts the focus from traditional web search to AI memory, raising questions about digital immortality and how AI models encode human identities.
The website queries multiple AI models, including GPT versions, Gemini, Claude, Grok, and Llama, asking them to list up to ten results for a given name, each with a brief description and a confidence level. It then clusters similar descriptions and assigns a strength score based on the models’ responses. For example, a user received a score of 641, placing them in the top 6% of names, while high-profile figures like Macaulay Culkin scored near 988. The site also highlights potential hallucinations, such as GPT-5.4 Mini suggesting that ‘Anthony Ha’ is ambiguous.
Dimson explained to TechCrunch that the project was motivated by a desire to explore how AI models ‘remember’ individuals and to challenge the idea that vanity Google searches are the future, especially as more traffic shifts to large language models (LLMs). The site’s playful design and provocative premise have garnered significant attention, with some seeing it as a reflection on digital legacy and AI’s role in personal identity.
Implications of AI Memory for Digital Identity
This development underscores a growing interest in how AI models encode information about people, potentially influencing notions of digital immortality. By quantifying how well models recall individuals, In the Weights prompts debates about AI’s role in preserving personal histories and the permanence of digital footprints. The tool also raises questions about bias, model differences, and the accuracy of AI-generated information, which are critical as AI becomes more integrated into personal and societal domains.
Ai fsem Toolbox Memory Matching Game Kids Educational Toys Felt Board Pieces Learning Activities Fine Motor Skills Toys Classroom Must Haves
- Complete 10-Piece Set: Includes felt pieces and storage bag
- Felt Board Compatible: Soft, durable flannel sticks easily
- Educational Memory Game: Builds concentration and visual memory
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Rise of AI as a Memory Repository
The concept of AI models ‘remembering’ individuals is rooted in the understanding that these models encode vast amounts of information during training, which can be retrieved or inferred during interactions. Previously, AI’s role was primarily as a generative or analytical tool, but recent developments focus on its capacity to serve as a form of digital memory. This shift is driven by the increasing sophistication of models like GPT-4 and beyond, and the growing public fascination with digital legacy and AI’s potential to ‘preserve’ identities.
The launch of In the Weights follows a wave of interest in measuring AI’s knowledge base, with some experts questioning whether models truly ‘know’ or merely generate plausible responses. The site also echoes broader conversations about bias, accuracy, and the ethical implications of AI recalling personal information.
“The idea of quantifying how well AI models remember individuals is fascinating but also raises questions about privacy and bias.”
— an anonymous researcher
Western Digital WUH721414ALE6L4 14TB Ultrastar DC HC530 SATA HDD – 7200 RPM Class, SATA 6 Gb/s, 512MB Cache, 3.5"
- Workload Rating: Up to 550TB per year
- MTBF: Up to 2.5 million hours
- Technology Generations: Five generations of HelioSeal
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Accuracy and Bias in AI Recall Scores
It remains unclear how accurately the models’ responses reflect real-world information or personal identities. The potential for hallucinations and bias, especially with newer or less-known models, is significant. The scoring system’s reliability and how it accounts for model differences are still under question, and the long-term implications of AI ‘memory’ are not yet understood.
ASUS Ascent GX10 AI Supercomputer, DGX Spark, NVIDIA GB10 Superchip, 128GB LPDDR5x, 1TB PCIe Gen4 NVMe SSD, Wi-Fi 7 & BT5.4, Agentic AI Ready, Supports OpenClaw, NemoClaw, Stackable Chassis
- AI Performance: Powered by NVIDIA GB10 Superchip with 1 petaFLOP
- Memory Capacity: 128GB LPDDR5x RAM for large models
- Storage: 1TB PCIe Gen4 NVMe SSD
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Further Research and Model Analysis
Developers plan to analyze why different models produce varying results for the same query, investigate biases toward certain types of individuals, and identify which figures lack sufficient AI representation. Additionally, the creators aim to refine the scoring system and explore broader applications of AI memory measurement, possibly influencing how models are trained and evaluated in the future.
ZNP Digital Badge AI Companion, Wearable Translation Translator with HD Touchscreen, Real-Time Interactive Reactions, Bluetooth 6.0 Portable Pin for Travel Business & Life
- AI-Powered Daily Companion: Assists with schedules, reminders, and facts
- Supports 101 Languages: AI conversations in multiple languages
- 60-Language Real-Time Translator: Instant translation for travel and business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
How does In the Weights determine a person’s strength score?
The site queries multiple AI models with a name, gathers their responses, clusters similar descriptions, and assigns a confidence score based on the models’ agreement and response strength.
Is this tool accurate or reliable for assessing AI memory?
Its accuracy is uncertain, as responses can include hallucinations or biases. The scores are more illustrative than definitive, serving as a playful exploration of AI recall capabilities.
What are the privacy implications of this project?
The tool does not collect personal data directly but relies on publicly available model responses. However, it raises questions about how AI models encode and potentially preserve personal information.
Could this influence how AI models are developed or trained?
Potentially, as understanding which models recall certain types of information could inform future training to reduce bias and improve accuracy in AI memory encoding.
Source: TechCrunch
