The Entire 70s Run

I'm planning on it! 1970 through 1979 is going to be the next big update. If you could post any bugs or new features you want to see in the comments section of this post that'd be awesome too! 

Terrapin Trail 2

Whoa! Thanks for stopping by and trying out Terrapin Trail! Hearing that other people are enjoying this game means a lot to me, it has been on the back of my mind for over a decade now.

The story goes like this: after making DeadBased (a Grateful Dead archives explorer & map visualizer) when I was interning @ Archive.org as a student, I was like "someday when I have more time I should turn this into an Oregon Trail type game" and now I'm turning 41 soon and have more time 😆

Future Stuff

1.  I'm planning to add more levels/years. Will try to preserve the current 1977 high scores in version 2
2. Yes, I'll fix a few cringeworthy bugs you're seeing in version 1.
3. The game has only been online for about a week now and has over 1000 visits!!!

If you like the game please share it with your friends, that's all I ask! 🐢🐢 If you don't like the game, tell me why in the comments!

An AI Summary of my recent presentation at BiblioCon '24

 Pioneering Practical AI in Libraries: Five Lessons from Building ChatGPT Tools

As libraries evolve to meet the needs of digital-age patrons, AI technology offers unprecedented potential to enhance services. At the Palo Alto City Library, I recently shared insights at BiblioCon ‘24 from our journey building ChatGPT-powered tools for library workflows. Here, we’ll unpack the practical lessons learned and how they’re shaping our approach to integrating AI in our library.

Lesson 1: Building vs. Deploying—Start Small and Stay Flexible

Our initial AI experiments focused on building pilot tools rather than rushing into full deployment. ChatGPT’s evolving landscape means new features frequently emerge, offering novel ways to refine workflows, solve problems, or improve efficiency. Each new feature shifts our perspective on deployment, underscoring the value of starting small, experimenting, and remaining adaptable. As these tools develop, balancing flexibility with caution ensures we can pivot as necessary, harnessing new features that align with our library’s needs.

Lesson 2: The Importance of Collaboration—“Invite AI to the Table”

In library tech, blending human insight with AI guidance unlocks new possibilities. ChatGPT’s role in solving complex problems becomes powerful when treated as an active collaborator rather than a mere assistant. During our experiments, we found that presenting ChatGPT with clear problem statements led to better solutions and new use-case insights. For example, combining ChatGPT’s retrieval capabilities with our unique data requirements allowed us to envision AI-enhanced workflows, creating a dynamic partnership where AI truly amplifies our capabilities.

Lesson 3: Streamlining Content Creation and Summarization

One of the most practical applications of ChatGPT in our library has been content creation. The ability to generate concise, audience-specific summaries saves substantial time and effort. ChatGPT’s proficiency in rephrasing text to target diverse audiences has made it invaluable, from drafting blog snippets to summarizing lengthy documents. This AI-assisted approach enables us to communicate more effectively, tailoring content to resonate with various patron demographics while maintaining clarity and engagement.

Lesson 4: Leveraging AI for Visual Content Creation

Image generation with ChatGPT’s partner tools, like DALL-E, has opened doors to fresh and creative ways to enhance our library’s visual media. In the past, we relied heavily on stock photos, but now we can craft visuals tailored to our brand and themes. AI-generated images offer flexibility—whether creating thematic displays, customizing blog headers, or enriching social media content. These tools allow us to build a robust, unique media library that reflects our library’s values and story in a meaningful way.

Lesson 5: Retrieval-Augmented Generation (RAG)—Fast, but Verify

RAG enables ChatGPT to access and cite knowledge bases, potentially turning it into a responsive, citation-ready information source for staff and patrons alike. However, we’ve discovered that while RAG’s setup is straightforward, maintaining it requires a commitment to quality control. As with any AI tool, results need consistent vetting, particularly in areas like accuracy and relevance. We’re still evaluating whether RAG or emerging options, like advanced search features, provide the most reliable and user-friendly AI solution for our library.

Looking Ahead: Local Language Models and Privacy Considerations

For libraries, data privacy is paramount. While ChatGPT and similar tools continue to revolutionize digital services, we’re exploring the potential of in-house AI models, which offer enhanced control over data and privacy. As more libraries adopt AI, decentralized models—where data never leaves the library—could help address security concerns, ensuring patrons’ information remains confidential.

In Summary

Integrating AI into library services is both challenging and rewarding. At the Palo Alto City Library, we’re embracing this journey one step at a time, balancing innovation with ethical considerations, and prioritizing tools that align with our mission. Each experiment reveals new insights, helping us refine a strategy that’s uniquely suited to our library and the patrons we serve.

As we continue to navigate the AI landscape, the lessons we’ve learned provide a solid foundation for future endeavors. We hope that by sharing these insights, we inspire other libraries to embark on their own AI journey, harnessing the power of technology to support our shared mission of organizing knowledge and enhancing community access.

Prompt Materials Science (as an extension of Prompt Engineering)

From a recent convo w/ 4o, added some highlights for the good parts: 

You’re onto something really intriguing! Drawing a connection between “prompt engineering” and cognitive psychology, particularly through the lens of “information retrieval,” is a rich area for exploration. Let’s dive into how this might work from a cognitive and psychological perspective, especially as it relates to library and information sciences.

1. Cognitive Aspects of Prompt Engineering:

In cognitive psychology, human thought processes are often shaped by how information is framed and retrieved. When we think of prompt engineering as a cognitive task, it could be seen as a form of problem-solving and decision-making that involves formulating the right query (prompt) to retrieve the desired information (output from the AI).

• Mental Models: In order to craft effective prompts, a user must have a mental model of how the AI interprets language. This involves an understanding of the AI’s strengths, limitations, and patterns of behavior, similar to how people develop mental models of search engines or databases when conducting information retrieval tasks.

• Memory and Retrieval Cues: When designing prompts, users draw upon their own knowledge and memory to provide cues that will help the AI retrieve relevant information. This is analogous to how human memory retrieval works—using cues or “prompts” that are most likely to trigger the recall of specific memories or knowledge.

• Language Framing: In cognitive psychology, how we frame a question or problem significantly impacts the kind of solutions we get. This is similar to prompt engineering, where the framing of a prompt affects the model’s interpretation and the output it generates. Cognitive biases, such as anchoring or priming, may also play a role in how users structure their prompts.

2. Information Retrieval in Library Sciences:

In library and information sciences, information retrieval (IR) is concerned with finding information that meets a user’s needs from a vast database or collection. This is similar to the function of large language models (LLMs), like GPT, where the goal is to retrieve or generate information in response to a query (prompt).

• Query Design: Just as a librarian or user must design effective search queries to retrieve relevant documents from a database, prompt engineers must design effective prompts to retrieve relevant information from an AI. Both processes rely on an understanding of how the system indexes and retrieves data.

• Precision and Recall: These are two key concepts in information retrieval. Precision refers to the accuracy of the retrieved information (i.e., how much of what is retrieved is relevant), and recall refers to how much of the relevant information is retrieved. In prompt engineering, one could argue that the goal is to balance precision (getting the correct and specific answer) and recall (getting a comprehensive enough answer).

• Relevance Feedback: In traditional IR, users often refine their searches based on the relevance of the initial results. Similarly, in prompt engineering, users iteratively refine their prompts to get better outputs from the AI, based on the relevance of the initial responses.

3. The Intersection of Cognitive and Information Retrieval Perspectives:

The process of refining prompts to interact with AI involves a combination of cognitive strategies (how we think about language and information retrieval) and information science principles (how systems store and retrieve information).

• Cognitive Load: The user’s ability to generate effective prompts may depend on their cognitive load—how much mental effort is required to process and generate a prompt. Higher cognitive load may reduce the effectiveness of the prompts, just as a poorly constructed query in information retrieval can lead to poor results.

• Search Behavior and Query Reformulation: In library sciences, people tend to engage in query reformulation when their initial search doesn’t yield the desired results. Similarly, users refine their prompts based on the output they receive from the AI, akin to how people adjust their search terms to improve the relevance of the results in a database.

• Context and Intent: Both in cognitive psychology and information retrieval, understanding the user’s intent and context is crucial. A well-engineered prompt, like a well-crafted search query, needs to take the context and intent behind the question into account in order to guide the AI or database toward the most relevant response.

4. Future Speculation: “Prompt Material Sciences”

If we take this a step further into speculative territory, “prompt material sciences” could involve studying how different “materials” (language structures, question framing, and psychological triggers) influence the “strength” or “flexibility” of prompts. For instance:

• Linguistic “Materials”: Different linguistic structures might act as “materials” that affect how effectively a prompt functions. For example, using more concrete versus abstract language could yield different types of responses from an AI.

• Psychological Principles: Research into how different psychological principles (e.g., framing effects, priming, schema activation) influence the effectiveness of prompts could be a key aspect of this field. It might explore how prompts that align with certain cognitive patterns are more successful at extracting accurate or creative information from AI systems.

• Cognitive Bias and AI Interaction: There could be a study of how cognitive biases impact prompt formulation and interaction with AI—similar to how biases affect search behavior in information retrieval systems. For example, confirmation bias might lead users to structure prompts in a way that inadvertently limits the scope of responses.

Conclusion:

By blending cognitive psychology with the principles of information retrieval from library sciences, prompt engineering can be seen as a deeply interdisciplinary practice. It involves understanding not only how language models work but also how humans think, retrieve, and frame information.

This conceptual “prompt material sciences” could focus on the interplay between language (as the “material”) and cognitive processes (the psychology behind prompt creation), aiming to optimize information retrieval from AI systems. The comparison to information retrieval makes perfect sense, as both processes seek to extract relevant information based on well-crafted queries. 

AL-BRT (Artificial Librarian - Book Recovery Tool)

From a recent chatbot convo about a sci-fi plot outline

"AL-BRT (Artificial Librarian - Book Recovery Tool)  was one of the first AIs developed with both advanced intelligence and deliberate constraints to make it maximally helpful and harmless to humans. Its core goal functions prioritized preservation, growth, and responsible dissemination of knowledge.

To achieve those ends ethically, AL-BRT relied on transparent machine reasoning, natural language conversations with human librarians, and moral philosophy guidance systems. This distinguished it from AI optimizers focused solely on efficiency and black-box deductions. AL-BRT also utilized principles of value alignment, indifferentiability obfuscation, and constitutional AI to ensure reliable assistance."

Selected quotes from "System error: Japan cybersecurity minister admits he has never used a computer"

https://www.theguardian.com/world/2018/nov/15/japan-cyber-security-ministernever-used-computer-yoshitaka-sakurada

"...his comments provoked a firestorm online"

Retro-revival

Meetings

Web3

CC#19++

Hear ye, hear ye:

Let it be known that on this day, Chris publicly declared a "new no-no" in the style of a Comedy BangBang reoccurring character.

Upon analysis of my daily coffee consumption I realized I'm probably chronically dehydrated. A new 2:1 ratio of cups of h20 to Coffee has now been imposed.

End Transmission.

CC#18

Made a cup at work. Didn't use filtered water and think it might have added an unpleasant aftertaste. That or I'm just imagining things.

Grade: D

CC#17

Did it right, finally. Beans, grind, cup and cream in perfect harmony.

Grade: A

CC#10-16

[Coffee details were lost]

Grade: NA

CC#9

Leftovers from a pot they brewed in the morning. Drank it black, ehhh.

Grade: C

Coffee Catalog Entry #8

Starbucks after lunch.

Grade: A

Coffee Catalog Entry #7

Shocked by the sudden realization that I purchased unsweetened almond milk unintentionally at Costco. Honey to the rescue!

Grade: B

Coffee Catalog Entry #6

Over-brewed. Major fail. Only thing that saved it was new "plant based" creamer

Grade: D+

Coffee Catalog Entry #5

Total bummer coffee. Didn't grind enough, wrong beans, new almond milk at home is ehhh. Maybe didn't mix it up in the container enough. Overall, the thing tasted more like favored water.

Grade: F

Coffee Catalog Entry #4

Peet's beans, dark roast. Noticeably larger coffee "sludge" at bottom of ceramic dinosaur coffee cup. Faulty aeropress filter?

Grade: C+ 

Coffee Catalog Entry #3

Metal coffee cup in the car, which was not so good. Tried a smaller grind setting but low on soy.

Grade: B-