Projects
Randos.club
This is a simple but robust random number generator where users can test whether their intention exerts any force on random number generators. Bear with me... Why would intention have anything to do with digital dice rolls? This experiment has a long history, and it starts with games, where the roll of a dice, sheep knuckles, or a few sticks could be leveraged by ancient peoples for fun and divination:
- Bagatelle is a precursor to pinball, where players launch a ball onto a board filled with pins, aiming for it to land in high-scoring holes. The ball’s unpredictable bounces make the game a physical random number generator (RNG), as small variations in force and angle produce different outcomes.
- Pachinko is a Japanese gambling game where players fire small steel balls into a vertical pin-filled board, hoping they land in winning pockets. The chaotic motion of the balls, influenced by gravity, bumpers, and launch variations, creates an RNG effect as the balls fall down.
- Payazzo is a Finnish coin-pusher arcade game where players slide a coin down a slanted surface with pegs, aiming for it to land in designated slots for prizes. The coin’s bounces and deflections off the pegs introduce randomness, functioning as a real-world RNG where each drop is unpredictable.
- Astragali: Ancient people used sheep knucklebones and cast them in both games and rituals as tools for divination
- Senet: In ancient Egypt around 3100 BCE, this board game had bits of chance sprinkled in to determine at what rate players moved through the board.
Interest beyond just 'chance' and 'fate' but into the paranormal, psychic phenomena, telekenesis, and new directions in quantum physics drew statisticians and experimenters to a new test: can someone get what they want from a random number generator and beat the odds? In a meta analysis of 380 studies in "Examining Psychokinesis: The Interaction of Human Intention with Random Number Generators" by three researchers, the answer can be summed up as:
Maybe.
“A significant but very small overall effect size was found. The study effect sizes were strongly and inversely related to sample size and were extremely heterogeneous.”
“Unknown to most academics, a large amount of experimental data has accrued testing the hypothesis of a direct connection between the human mind and the physical world.”
“If human intention can influence physical systems, it challenges materialistic conceptions of reality, suggesting a more interconnected relationship between consciousness and the physical world.”
“Over time, experimental and statistical methods improved, and, in 1991, Radin & Ferrari undertook a meta-analysis of the dice experiments, finding small but significant deviations from chance.”
“The transition from dice experiments to random number generators marked a shift toward greater experimental control but also introduced questions about whether psi effects operate on a quantum or macroscopic level.”
“Despite methodological refinements, the evidence for psychokinesis remains controversial, with skeptics attributing observed effects to experimenter bias, selective reporting, and statistical artifacts.”
My hope is that with a global userbase, a large number of sessions, and incredible random randomness, we can figure this out together.
How random is randos.club? I've implemented drand, a public service spearheaded by Cloudflare (also known as the The League of Entropy) which bills itself as "Verifiable, unpredictable and unbiased random numbers as a service." The drand random number is mixed in with the timestamp, a salt of the username, and stirred up each second (stored as milliseconds after session start) for a mega number.)
How random numbers are generated on Randos.club:
drand_randomnessvalue is obtained from the Drand network every 30 seconds.saltis generated from the user IDtimestampis recorded for each second past the initial session start, as millisecondcombinedInput=drand_randomness+salt+timestamphash= SHA-256(combinedInput)randomNumber=parseInt(hash.charAt(0), 16) % 2(Take the first character of the hash, convert to a number 0-15, then mod 2 to get 0 or 1)
Caregiver Companion GPT
In an estimated 36.5 million households across the country an adult is providing unpaid care to a family member according to the “Caregiving in the U.S.” study by the National Alliance for Caregiving.
There's an enormous body of literature and resources available on the web for navigating the caregiving experience, and I wanted to create a tool that would help serve as a quick reference for navigating the process. Questions like:
- What should I know about caring for someone with alzheimers, specifically on the topic of financial planning and personal care agreements? example answer
- In New York, what resources are available to caregivers? example answer
- How do I deal with caregiver burnout? example answer
- How do I begin talking with my parents about their end of life care? example answer
Googling this information is tough when a great deal of this knowledge is locked up in huge publications, behind advertising, and hidden in PDFs on government websites. This tool brings those resources closer to those in need with vetted techniques and strategies to help caregivers access this information.
When OpenAI announced custom GPT models, I kept encountering GPTs built for Open AI's demographic: 'custom websites', 'help creating Excel formulas'... etc... but for one of the most challenging aspects of life on Earth -- caregiving -- especially in the U.S. -- I wanted to test if a tool like this would be actually useful. I've done a great deal of testing, prompt crafting and work on this and I hope it's useful to others.
Visit the Caregiver Companion
This custom GPT has been primed with a custom prompt that is designed to help caregivers with their day-to-day activities. It includes tools for planning, tracking, and managing tasks, as well as resources for finding support and information about mental health issues. Beyond the custom prompt, the GPT also has extensive knowledge about common issues caretakers encounter as they assist ailing adults. Answers improve as questions it receives improve, so the more detail provided in your questions the better.
The GPT has been trained on free resources published by leading elder care resources:
- US Department of Health & Human Services National Institute on Aging - Caregiver's Handbook
- Medicare.gov - Medicare and You Handbook
- Women's Institute for a Secure Retirement - Financial Steps for Caregivers: What You Need to Know About Protecting Your Money and Retirement
- Alzheimer's Family Center
- And more. If you'd like to suggest something, please reach out. At the moment it is geared towards U.S. caregivers.
I will say that I'm aware of all the issues that this type of service poses:
- Large language models can hallucinate and they are also seeking to soak up more information to improve their models. Their training data inevitably consists of material under copyright. Questions asked of LLMs may be used to train future models at Open AI. It's not super easy to opt out but that is one way to ensure that your questions and data do not get used, even anonymously, for financial gain. What I recommend is keeping history on, but turning off model training by following these steps.
- I don't intend for this model to be used for healthcare (medical) advice. Caregiving as a practice is a very personal thing, but it also requires a great deal of logistics, record keeping, mental health check-ins, and emotional intelligence. This model may answer empathically, but it's still just spicy autocorrect.
- Finding a balance between data protection and finding help for caregiving questions is a huge discussion. If I wasn't impressed with the answers I've received from this tool I wouldn't make it public.
Moonphase Generator
Building a service to show "what's today's moonphase" was a fun project. Today the moon looks like: 
Learn more here.
