I am retiring from fulltime writing (& pseudonymity) to launch Guardian Angel
An influential long‑time pseudonymous writer is leaving full‑time writing to launch “Guardian Angel,” a startup that builds highly personalized LLM “digital twins” meant to act as trusted, user‑aligned agents. Commenters are sharply divided: supporters see a path to safer, more empowering AI that preserves individual autonomy, while critics warn of elitism, deep privacy risks, reinforcement of users’ worst traits, and a broader AI “mania” that could concentrate power and worsen social and economic divides. The exchange surfaces deeper tensions over whether more personalized AI can genuinely protect human agency, or whether structural regulation and limits on such technologies are the only realistic safeguards.
Project concept and goals
- Discussion centers on a new startup proposing “Guardian Angels” (GAs): highly personalized LLM-based “digital twins” that emulate a single user’s personality, values, and preferences.
- Claimed aims: enhancement over replacement, preservation of “mental sovereignty,” and support for self‑actualization; oppose generic, ad‑driven, lab‑controlled chatbots “for everyone.”
- Technical sketch: continual/online learning, preference learning from the user, local/CLI‑style UX, and strong focus on privacy and security.
Positive reactions
- Some see the idea as ambitious and timely given fears that generic agents will replace humans and optimize them out of workflows.
- A collaborator strongly vouches for the founder’s character and concern for downstream impacts.
- Supporters like the focus on user alignment, data security, and the possibility of a “smarter version” of oneself as a tool for work, politics, or long‑term legacy.
- Others appreciate the attempt to give individuals an AI advocate in a world of powerful centralized systems.
Ethical and psychological concerns
- Many find the notion of renting a digital copy of oneself “dehumanizing,” narcissistic, or nightmare‑inducing.
- Worries about creating simulacra of real people, especially persisting after death, are frequent.
- Several see the essay as evidence of “AI psychosis” or quasi‑religious belief in LLMs as future “quasi‑gods.”
- Some argue that the rationalist/AI milieu behind the project has a track record of harmful outcomes and elitist attitudes.
Alignment, safety & misuse
- Skeptics dispute the claim that a GA is “trustworthy by definition” because it shares the user’s values:
- What if the user is malicious or unstable?
- A self‑echoing AI could entrench delusions and biases instead of challenging them.
- Others argue that valuable AI partners should be “other,” not copies of oneself, to provide genuine critique and growth.
- Concerns that GA‑driven political participation could degenerate into automated spam and further degrade public discourse.
Privacy, ownership, and power
- Strong anxiety about giving a company enough data to fully emulate one’s mind.
- Critics say a model “of you” that is owned and rented back by a firm intensifies dependence and exploitation; real sovereignty would require local, user‑owned hardware and open models.
- Some insist they would only consider such a system if self‑hosted and fully controlled by the user.
Economics & access
- Proposed early cost target (> $1,000/month in 2026) is seen as cementing GA as an elite tool that widens inequality.
- Some expect rapid cost declines; others note that vendor market power, hardware bottlenecks, and VC losses may keep prices high.
Broader AI trajectory
- Deep disagreement over whether current progress implies an imminent “singularity” or another overhyped bubble.
- Some view LLM advances (e.g., in math) as early signs of general superhuman cognition; others stress current brittleness and argue that solving human problems like war or hunger is fundamentally different.
- Debate over whether tech “for good” should be more tech (like GA) or primarily regulation and personal/collective restraint.