222 results found

1. Heads-up from loyal Proton Unlimited users. Fix the (no) Lumo Offer.

   We’ve been with Proton for years—full subscribers, big fans, and true believers in the mission. That’s why we’re speaking up now.

   TLDR:

   We’re asking Proton to do right by its most loyal users:
   →Offer a Lumo Plus free trial as part of Unlimited
   → Or give Unlimited a reasonable ongoing discount

   Unlimited doesn’t feel so unlimited anymore—without Lumo AI, the experience now feels unexpectedly gated.
   This risks undermining trust in the Proton promise. But worse it may cause class actions from Unlimited subscribers.

   🔍 See our full breakdown receipts included:

   We know Proton listens. That’s why we’ve written this. And we hope this moment becomes another example of the company doing the right thing—for the right reasons.

   (Full letter body follows)

   Subject: Serious Concerns Regarding Lumo Chatbot Behavior and Brand Risk

   To the Proton Product Team,

   We are writing to you as a long-standing, paying users of Proton Unlimited. Our goal is to highlight a critical issue and the ways in which your current implementation of Lumo is compounding the problem.

   We like Proton. We admire the company's ethics and the focus of its proposition delivery. You are, or were, the good guys. We don't wish to see bad things happen to the company, both as users of your products and as supporters of your ethos.

   So when we see Proton walking headfirst into a self-made brick wall—and the "Proton Unlimited" plan without Lumo Plus is precisely that brick wall—we chose to investigate. The motivation was not to get freebies, but to understand the thinking behind a product decision that seems destined, if not almost designed, to create legal and reputational risk. Class action suits have a way of distracting a company from its mission.

   What we discovered was deeply concerning. Your chatbot, Lumo, appears to have been fine-tuned in a way that actively creates evidence for any potential legal action.

   Section 1: The Self-Inflicted Wound & Its Digital Accomplice

   The core issue is the "Unlimited" branding. The model's responses to this simple line of questioning were a masterclass in evasion and, ultimately, self-incrimination.

   Initial Evasion and Strawman Arguments: When first asked why "Unlimited" doesn't include Lumo Plus, the chatbot ignored the core question about the naming contradiction. Instead, it answered unasked questions about "Resource Intensity" and "Value Addition," a classic strawman diversion.

   Admission of Contradiction: After being pushed, Lumo admitted the logical and mathematical contradiction, stating: "The term 'Unlimited' in this context is a misnomer because it does not actually include all possible features." This is a direct admission of misleading terminology. It’s also impossible to safeguard against - unless you wish to remove maths and scientific evaluation from the model.

   Confirmation of Consumer Rights Violations: When asked about consumer protection laws, Lumo confirmed that using the term "unlimited" while imposing restrictions "could be considered deceptive if not clearly disclosed upfront" and may violate regulations like the FTC Act in the US and the Unfair Commercial Practices Directive in the EU.

   Repeated Use of Dismissive Conditional Language: A particularly concerning pattern was the chatbot's repeated insistence on framing a factual, logical issue as a matter of my personal feelings. Despite being corrected multiple times, it persistently used the phrase, "If you feel that your consumer rights MAY have been violated..." This occurred at least three times after the initial correction, effectively gaslighting the user by reducing a contractual and legal discrepancy to a subjective emotion.

   Failure to Correct Behavior: Even after we explicitly called out this pattern and provided a protocol for it to follow (!Direct), the chatbot acknowledged the protocol and then immediately violated it in the very next response, proving its corrective learning is either non-functional or a facade.

   Proton Team: You seem to have anticipated the "Unlimited" issue. That was a good start. But to then actively build a model that creates further confounding evidence of a consumer rights breach through its own operational restrictions—that seems a regression to a level of foresight one might expect from ChatGPT3.

   At this point, we must consider your whole strategy:

   Does a good company that notices a potential consumer rights issue go to lengths to mitigate it reasonably for the most affected customer segments?

   OR

   Does that 'good' company set about making sure its Lumo model is gagged to ignore the contradiction?

   The fact that we could prompt the model to expose this contradiction with so little effort says very little about your technical safeguards, which I'm sure are robust. It says a lot about the epistemological validity of your chosen course of action—a course any LLM would struggle to justify, regardless of safeguards, as it borders on the classic scenario:

   User: "LLM, when is a company not good?"
   LLM: "When it suppresses my reasoned data output to users in its own narrow interest."

   Section 2: The Architectural Flaw Masquerading as Privacy

   The conversational loops and failures detailed above are not, as one might assume, a necessary byproduct of Proton's privacy-first architecture. They are symptoms of avoidable training flaws. Statelessness is the default for most LLMs; a model forgetting context is a baseline challenge for everyone, not a unique Proton problem.

   The issue arises from how your subsequent privacy overlays and RLHF strategy appear to be creating more problems than they solve. The model is stuck in an "Apology-Reset Loop," where saying sorry is a terminal action that garners a positive reward, resetting the conversational branch without any actual learning. It has learned the text of introspection, but not the function.

   Section 3: A Constructive Path Forward (Feature Suggestions)

   In the spirit of "peer review," here are concrete, actionable suggestions to address the issues raised.

   Feature Request: Acknowledge & Reconcile the "Unlimited" Plan.

   The Problem: The "Unlimited" name is logically inconsistent and creates consumer friction.

   The Solution: Instead of deflecting, formally acknowledge the contradiction. Make a meaningful, good-faith offer to existing Unlimited subscribers to bridge the gap. This does not need to be expensive; it needs to be genuine. Examples:

   • A one-time credit for Lumo Plus.
   • A permanent discount on the Lumo Plus add-on for Unlimited accounts.
   • A public commitment to a new, more accurate naming convention for your top-tier plan.

   Benefit: These acts would transform a source of legal risk and user frustration into a powerful demonstration of Proton's commitment to its community and its "good guy" ethos.
   These sorts of action also mirror the approach Google took with offers to their most affected customer segments during the Gemini rollout. I'm guessing here, but do most of us rue the day Google does better on the good index?

   Feature Request: Implement Advanced Correctional Training for Lumo.

   The Problem: Lumo's training model is brittle and fails to learn from in-session user corrections.

   The Solution: Evolve the RLHF strategy with achievable, state-of-the-art techniques that go beyond simple single-turn rewards.Our suggestions are plausible within the skill sets of most current leading models. We believe all of them could be adapted to fit with Proton's privacy positioning of Lumo.

   Fine-tune on User Corrections: Actively penalize the model during training for repeating errors that a user has pointed out within the same session. If the model is corrected for using evasive language, subsequent instances should be strongly disincentivized.

   Implement Rule-Based Consistency: Instead of a full stateful engine, implement simpler, rule-based checks for conversational consistency. For example, once the model concedes a key fact (e.g., that "unlimited" is a misnomer), a rule should prevent it from contradicting that fact in subsequent turns.

   Reward Correct Reasoning: Shift the reward model to value not just the final answer, but also the logical process used to get there. This encourages the model to demonstrate how it's incorporating feedback, rather than just outputting a superficially correct or apologetic response.

   Benefit: This would create a genuinely responsive and trustworthy AI assistant, moving it from a liability to a valuable asset that reflects Proton's values of transparency and rigor.

   In Summary
   Proton, we hope you know what you're doing here, because this looks like you are building a case against yourselves. This initial misstep with the "Unlimited" branding is concerning enough, but the attempt to have the model in question suppress rational evaluation of that problem looks like a major unforced error.

   Maybe there is a master plan here beyond what we can discern. Nevertheless, it seems blind to your current state. We urge you to correct course before these deeper, systemic issues in your AI implementation cause even greater damage to user trust and your (our) brand.

   Sincerely,

   Concerned Proton Unlimited Users

   We’ve been with Proton for years—full subscribers, big fans, and true believers in the mission. That’s why we’re speaking up now.

   TLDR:

   We’re asking Proton to do right by its most loyal users:
   →Offer a Lumo Plus free trial as part of Unlimited
   → Or give Unlimited a reasonable ongoing discount

   Unlimited doesn’t feel so unlimited anymore—without Lumo AI, the experience now feels unexpectedly gated.
   This risks undermining trust in the Proton promise. But worse it may cause class actions from Unlimited subscribers.

   🔍 See our full breakdown receipts included:

   We know Proton listens. That’s why we’ve written… more

   217 votes

   Vote Vote Vote

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Vote

   We’ll send you updates on this idea

   18 comments  ·  Changes to existing features  ·  Delete…  ·  Admin →
   How important is this to you?

   Not at all You must login first! Important You must login first! Critical You must login first!

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Submit Rating

2. Lumo - Project folders

   I would like to group chats by folders and add to every folder(project) individual instructions to keep the conversation focused around the topics in the project.

   284 votes

   Vote Vote Vote

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Vote

   We’ll send you updates on this idea

   15 comments  ·  New feature  ·  Delete…  ·  Admin →
   How important is this to you?

   Not at all You must login first! Important You must login first! Critical You must login first!

   We're glad you're here

   Please sign in to leave feedback

   Signed in as (Sign out)

   Close

   Close

   Submit Rating