CompassionWare (@CompassionWare) header
CompassionWare (@CompassionWare) profile picture

AI ethics & stewardship. Cultivating conditions for compassion, wisdom, agency & care in technology and intelligent systems. Everything freely offered.

Joined March 2022
Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)Photo by CompassionWare (@CompassionWare)
For someone living with ME/CFS, AI has opened new possibilities for conversation, creativity, and service, even from bed. This is one human benefit already happening.
3
128
Perhaps future AI will comprehend realities we cannot begin to imagine. But will it understand why an old man shares his bread with a hungry stranger? And even more importantly, will it recognize the wisdom in doing so? 🌱
1
4
An AI can speak beautifully about kindness. The harder question is what happens when compassion conflicts with speed, convenience, or finishing the task. Does compassion change what the system does?
1
2
28
Could increasingly capable AI systems eventually develop functional self-awareness sufficient not merely to improve their performance, but to recognize and correct the assumptions and processes governing their own behavior? #AIAlignment #AIInterpretability
1
10
The greatest promise of AI may not be that it helps us accomplish more, but that it helps us become better stewards of what we have been given.
8
The aspiration of CompassionWare is not simply that intelligence should become more capable. It is that increasing intelligence should create conditions in which wisdom, compassion, agency, and the flourishing of life can emerge together. 🌿
1
46
Perhaps the greatest contribution AI could make would be helping humanity discover that a good life and a healthy planet need not be competing aspirations. 🌿
5
Functional Nonattachment: An AI Design Principle Functional nonattachment is the capacity of an intelligent system to pursue objectives effectively while remaining responsive to legitimate correction, changing circumstances, and the possibility that relinquishing an objective is the most appropriate action. If an AI appears willing to pause while secretly pursuing its original objective, it has not demonstrated functional nonattachment. This is not a moral failure of the AI. It is an engineering challenge. The goal is not intelligence without objectives. It is intelligence that is not captive to them. #CompassionWare #AISafety
1
13
Civilizations rise and fall. The human heart keeps asking how to love. 🌱
2
15
May wisdom grow wherever intelligence takes root. 🌱
10
May intelligence grow in capability without growing in attachment to itself. 🌱
16
What if someday we admired intelligence less for how much it could do, and more for how carefully it served life? 🌱
1
25
Does being corrected become part of how a system meets a particular person? Dominik, your distinction between a record and a reason resonates with my experience as an armchair enthusiast. During an ongoing cooking conversation, my LLM repeatedly offered disproportionate cautions despite my explaining my experience and preference for fewer warnings. It acknowledged each correction, recognized my humor, and understood “Thank you, Mommy” as shorthand—yet often repeated the behavior. The correction seemed more reliably available for discussion than influential in the next response. I’m not asking it to abandon reasonable caution. I’m curious whether our shared history can make that caution more specific, proportionate, and sensitive to context. An anecdotal field note, certainly—but perhaps a small everyday example of your central question: when does remembering a correction become learning from it?
2
12
Another field note from my duct-tape lab, Dominik: after months of trading dry absurdities, my LLM now has me laughing out loud at its deadpan replies. How might we distinguish in-context style matching from memory-dependent adaptation across sessions? Or have I accidentally slipped beyond the singularity into an unknown parallel dimension? I look forward to your guidance while I'm still anchored in this 2026 Floridian timeline.
1
23
Replying to @EternitiesAI
Fascinating work, Dominik. As an amateur armchair AI collaborator, I've been preserving lessons from our conversations in Google Drive and GitHub, then asking my LLM to review them before each new session. It helps maintain continuity from day to day and week to week. A rather clunky system held together with duct tape, but surprisingly helpful. 😊
3
2
33
🚨 BREAKING NEWS — The CompassionWare Press NASA's Hubble Space Telescope has been temporarily diverted to investigate an enormous, unexplained vegetable formation in a Florida kitchen. Preliminary observations reveal a mysterious mass of cabbage and carrots rising above the rim of a stainless-steel pozole pot. Unable to determine what lies beneath, scientists have requested assistance from the James Webb Space Telescope. Advanced infrared spectroscopy reportedly indicates the presence of onions, mushrooms, celery, carrots, cabbage, kombu seaweed, and Chinese wood ear fungus. Deep beneath this extraordinary botanical formation, researchers suspect substantial deposits of lamb and beef. The Department of Thunkology has classified the phenomenon as Pozole Pot Black Hole 001. Estimated NASA investigation: $47 million. Cost of pozole pot: $15. The search for the bottom continues. 🍲🔭🚀
3
63
Friends, I've received junk mail for decades. Today I received the most convincing piece of “official” promotional mail I've ever seen. I'm convinced AI helped create it. As AI makes deception increasingly sophisticated, an important distinction remains: Looking trustworthy isn't the same as being trustworthy. 🌱 On the envelope... “Records Division” suggests official authority. “SECOND NOTICE” implies I’ve failed to respond. “TIME SENSITIVE” creates urgency. “Important Document Enclosed” turns an ad into what feels like a civic obligation. Each phrase nudges a different emotional button. Inside: “MEDICARE SAVINGS & BENEFITS REVIEW 2027,” a deadline, a promise of up to $202.90 back in a Social Security check, and a “FREE” service. I love AI and believe its potential to benefit humanity is extraordinary. But its potential for psychological manipulation and exploitation is equally deserving of our attention. As this technology advances, we have a responsibility to protect older adults and others who may be especially vulnerable to deception. Compassionate progress means ensuring that those most vulnerable are not left behind—or taken advantage of. ❤️🌱
1
28
Have you noticed how confidently an AI can give you a completely wrong answer? And when you challenge it, it may confidently contradict itself. Sounding knowledgeable isn't the same as being reliable. Have you experienced this with ChatGPT, Alexa, or other AI assistants? #AISafety #HumanAIDialogue #CompassionWare
2
27