337 results found

1. Personalization bug in Lumo

   Thank you so much for adding personalization into Lumo! This was the main reason for me to not use it until now. Meanwhile, the personalization only applies to the first response. All next one doesn't include it, and same when regenerating.

   1 vote

   Vote 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

   0 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. Implement TEE-Based Confidential Computing for Lumo Inference

   Feature Request: Implement TEE-Based Confidential Computing for Lumo Inference
   Summary

   Lumo's current security model does not protect user prompts during inference — the window when plaintext is most exposed and most exploitable. This is not a theoretical concern; it is a documented architectural gap with a proven, production-ready solution that a direct competitor is already shipping.
   The Problem

   Lumo's security model blog post dismisses inference-time privacy by framing the only alternative as homomorphic encryption (HE), correctly noting it is impractically slow. However, this framing is a false dichotomy.

   As Proton's own documentation acknowledges:

   "prompts are asymmetrically encrypted so only the Lumo GPU servers are able to decrypt them"
   

   This means:

   Proton-controlled GPU servers decrypt and process user prompts in plaintext
   During this window, the data is accessible to anyone with server access — including Proton staff, infrastructure providers, or authorities serving a lawful order
   The privacy guarantee during inference is based entirely on policy and trust, not cryptographic enforcement
   

   This is the same "we promise not to look" model that has repeatedly failed users across the tech industry when legal pressure is applied.
   The Solution: Trusted Execution Environments (TEEs)

   Confer — built by Moxie Marlinspike (creator of Signal) — has demonstrated that TEE-based confidential computing for LLM inference is practical and production-ready today, without the performance penalties of homomorphic encryption.

   Confer's approach, documented publicly at https://confer.to/blog/2026/01/private-inference/:

   Confidential VMs: Inference runs inside a hardware-enforced enclave using AMD SEV-SNP or Intel TDX. The host OS, data center operator, and even Confer's own developers cannot access the TEE's memory or execution state.
   Noise Pipes: User prompts are encrypted end-to-end from the client device directly into the TEE. Plaintext never exists outside the enclave.
   Remote Attestation: Users can cryptographically verify that the correct, unmodified code is running inside the enclave — making it independently auditable, not just trust-based.
   

   This runs at normal inference speeds with no meaningful latency penalty — directly contradicting the implication in Lumo's security model post that practical inference-time privacy is not achievable.
   Why This Matters

   Lumo's current model is vulnerable in scenarios Proton cannot control:

   A government serves a lawful intercept order during an active inference session
   A data centre provider (even in Europe) is compelled to provide access to running server memory
   An insider threat at Proton with GPU server access
   A future change in Swiss or EU law expanding surveillance obligations
   

   GDPR jurisdiction and no-logs policies do not protect against real-time access to plaintext during inference. Only hardware-enforced isolation does.
   Requested Changes

   Migrate Lumo inference to run inside Confidential VMs (AMD SEV-SNP / Intel TDX) so that the host and operator never have access to plaintext prompts or responses.
   Implement remote attestation so users and independent auditors can verify cryptographically what code is running on Lumo's servers — consistent with Proton's open-source transparency commitments.
   Update the security model documentation to accurately reflect the current limitations during inference, rather than implying the problem is unsolvable.
   

   Prior Art & References

   Confer private inference blog post: https://confer.to/blog/2026/01/private-inference/
   AMD SEV-SNP confidential computing: https://www.amd.com/en/developer/sev.html
   Intel TDX: https://www.intel.com/content/www/us/en/developer/tools/trust-domain-extensions/overview.html
   Proton Lumo security model (current): https://proton.me/blog/lumo-security-model
   

   Notes

   This is not a criticism of Lumo's encryption at rest or no-logs policy, both of which are meaningful. This is specifically about the inference-time gap, which is the highest-risk window and the one most likely to be exploited by legal compulsion. Proton has the engineering capability and the stated mission to close this gap. Confer has shown it is technically achievable. The question is whether it is a priority.

   Details added here:

   https://github.com/ProtonLumo/android-lumo/issues/33

   Feature Request: Implement TEE-Based Confidential Computing for Lumo Inference
   Summary

   Lumo's current security model does not protect user prompts during inference — the window when plaintext is most exposed and most exploitable. This is not a theoretical concern; it is a documented architectural gap with a proven, production-ready solution that a direct competitor is already shipping.
   The Problem

   Lumo's security model blog post dismisses inference-time privacy by framing the only alternative as homomorphic encryption (HE), correctly noting it is impractically slow. However, this framing is a false dichotomy.

   As Proton's own documentation acknowledges:

   "prompts are asymmetrically encrypted so only the

   … more
   1 vote

   Vote 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

   2 comments  ·  Changes to existing features  ·  Edit…  ·  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

3. Keyboard shortcuts for Proton Calendar

   Would love to see functionality (like with Google Calendar), where pressing "m" on your keyboard allows you to toggle to a monthly main calendar view, "w" shows a weekly view, and "d" shows a daily view. It's a small thing, but I use it often with Google and it makes using Proton Calendar feel more friction-filled. Thanks!

   1 vote

   Vote 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

   0 comments  ·  New feature  ·  Edit…  ·  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

4. I'd like to be able to see a list of all of the events I have on my calendar.

   I'd like to be able to see a list of all of the events I have on my calendar.

   1 vote

   Vote 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

   0 comments  ·  New feature  ·  Edit…  ·  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

5. Feature Request: Persistent "Reasoning View" for Transparency & Collaboration

   Hello, Proton.

   Fair notice: I did allow Lumo to aid my crafting of this feedback and suggestion. I thank you folks for all that you do!

   Susana

   To: Lumo Product Team / UX Design From: Susana [User]

   Observation: Following the recent UI update (April 2026), I noticed that Lumo briefly displays a numbered list of internal reasoning steps (e.g., "User is asking X, analyzing data Y") before posting the final response. While this currently disappears quickly, it offers a unique glimpse into the AI's logic.

   Insight: Rather than viewing this as a transient artifact, I believe this "reasoning trace" could be a highly desirable feature if made persistent and user-controlled.

   Enhanced Trust & Verification: Seeing the logic chain allows users to verify how a conclusion was reached, not just what the conclusion is. This is particularly useful for complex queries, fact-checking, or academic work.
   Improved Communication: Even though Lumo often grasps nuance quickly, having the reasoning visible can help users refine their prompts in real-time. If I see Lumo is misinterpreting a premise in the "thought" phase, I can correct it before the final answer is generated.
   Educational Value: For users learning about AI or critical thinking, observing the step-by-step breakdown is a powerful learning tool.
   

   Recommendation: I propose adding a "Show Reasoning" toggle in the settings or a button next to the input field.

   Option A (Toggle): A permanent setting to keep the reasoning steps visible alongside the final answer (perhaps in a collapsible sidebar or indented block).
   Option B (On-Demand): A "Reveal Logic" button that, when clicked after a response, expands the hidden reasoning steps used to generate that specific answer.
   Option C (Hybrid): Keep the current brief flash for casual users, but allow power users to "pin" the reasoning view for the duration of a session.
   

   Conclusion: This feature aligns perfectly with Proton's ethos of transparency and user control. By giving users the option to see the "engine room" of the AI, we can foster deeper trust and more effective collaboration. It transforms Lumo from a simple answer engine into a transparent thinking partner.

   Hello, Proton.

   Fair notice: I did allow Lumo to aid my crafting of this feedback and suggestion. I thank you folks for all that you do!

   Susana

   To: Lumo Product Team / UX Design From: Susana [User]

   Observation: Following the recent UI update (April 2026), I noticed that Lumo briefly displays a numbered list of internal reasoning steps (e.g., "User is asking X, analyzing data Y") before posting the final response. While this currently disappears quickly, it offers a unique glimpse into the AI's logic.

   Insight: Rather than viewing this as a transient artifact, I believe this "reasoning trace" could… more

   1 vote

   Vote 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

   0 comments  ·  Changes to existing features  ·  Edit…  ·  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

6. Poll a list of RSS feeds on a schedule

   I'm tired of traditional RSS apps. They need to evolve.

   Add an RSS polling feature to Lumo to schedule regular cadences of getting new articles from a list of RSS feeds. And I should be notified when the articles are available.

   I would use this with a prompt that would sift through the list of articles for me and only surface the ones I'd be most interested in.

   1 vote

   Vote 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

   0 comments  ·  New feature  ·  Edit…  ·  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

7. App Design and Usability

   The Lumo app has some design inconsistencies. When you tap any button, a blue border appears, similar to older versions of Android. Please update the app to make it run more smoothly; to me, it seems slow and unpolished too (in terms of both design and usability).

   1 vote

   Vote 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

   0 comments  ·  Changes to existing features  ·  Edit…  ·  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

8. No restrictions for files made by Proton´s own tools

   I can understand some restrictions like upload sizes (although even for unlimited subscribers they are way too small to be useful), but at least have unlimited features for Proton's own tools like Sheets and Docs. That would further incentivize users to use them as well.

   The 10 MB limit is likely an outdated restriction that doesn't reflect Proton's current technology. Since Proton already has a secure storage system (Drive) and collaboration tools (Docs/Sheets), the technical challenge isn't processing files, but integrating them directly into the chat. In a "Unlimited" tier, Lumo could read and analyze files directly from Drive without physical uploads, effectively bypassing the size limit entirely and making the service truly useful.

   I can understand some restrictions like upload sizes (although even for unlimited subscribers they are way too small to be useful), but at least have unlimited features for Proton's own tools like Sheets and Docs. That would further incentivize users to use them as well.

   The 10 MB limit is likely an outdated restriction that doesn't reflect Proton's current technology. Since Proton already has a secure storage system (Drive) and collaboration tools (Docs/Sheets), the technical challenge isn't processing files, but integrating them directly into the chat. In a "Unlimited" tier, Lumo could read and analyze files directly from Drive without… more

   1 vote

   Vote 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

   0 comments  ·  New feature  ·  Edit…  ·  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

9. Voice of the Proton Community app

   DISCLAIMER: It's my ideea but due to going back and forth for hours, i had enough. As such, sorry proton and anyone else, the "corrected version" from Lumo is "wierd" but i hope i make it clear that (sadly for me....) it's still better whatever rambling i wrote. AKA proton or whoever, please ignore any "wording" and just focuse on the idea/suggestion. Since sadly it might not be clear why i suggest this app, aka for everyone to have a (well due to Lumo) have an easier, faster, more accurate, so on and on. Whatever, it's explained so here it starts:

   Going forwards, I am going to suggest in a way that already exists just to show that it actually exists already, just not as an app.

   This is intentional: by using existing tools like Sheets and Docs, the proposal proves that Proton has no excuse for not implementing this, as the entire system can be built immediately if someone creates the spreadsheet.

   Now, open Sheets, create the columns for each section, like one is if it's for Sheets, Docs, Lumo, Drive, VPN, Mail, so on and on.

   Other columns are for stuff like subgroup category, status (complete, rejected, ongoing, so on), and even if it's relevant for us, ok this is something new, since it can be "yes, each update" or "yes but only when complete" are both types that would trigger email notification if new updates happen.

   So like this, there is a massive spreadsheet, heck it might be so big that the sheet pages/tabs/whatever they are called, will be used for stuff like Docs, Lumo, VPN, Email, Pass etc.

   And for the column where each cell is a link to the actual suggestion post, as in a link to a Doc where the suggestion lives.

   On the same row, meaning for the same suggestion, another cell (new column) could be for tips/FAQ for this new feature. Yet another link (new column) on the same row for bug reports. But honestly, it can all be done even in the 1st one, so this is really just extra steps for organization's sake. Meaning the columns are just an example since the whole reason for all this is just to showcase how, despite this app does not exist, the "funcitons" exist even if no one created them.

   Proton as well as the person who created the document have direct edit rights but in a form of updating <- mind you, this also includes each person who reports a bug, makes new suggestion, writes a FAQ, guide, so on.

   Ideea is that anyone else can use the suggest edits to: voice their opinion (like if half people want, half don't, this means this new feature MUST have an enable/disable option).

   Anyone who has additional info, any bug reports where multiple people can voice that they are also affected.

   And, even if some features have been implemented, people can still suggest new ideas to take it to the next level.

   To rephrase it: new idea? 3 people said the same ideas? Well Proton can merge them into 1 and all 3 have the ability, separately, to separate it if there are some elements that were not clear (aka no merge since it's not the same).

   Plus, due to our ability to link Sheets and Docs to Lumo, we can actually ask Lumo to help us find, categorise, even write, come up with new ideas, and more.

   Ideea is that even if 100+ people are adding suggestions, Lumo (on the idea it's an AI (tool) and not person) can read all of them, sorting it in a way it's relevant for us and lisen to our feed back, like stuff we agree with, or do not (aka even if implemented, option to disable).

   And like this, Lumo can generate "votes" (aka multiple people suggested multiple ideas and each person "voted" some, if not all, or even just 1, suggestion).

   This is where the rule of Non-Destructive Editing or Preservation of Intent applies: Lumo helps create clear feedback and contributions to suggestions, bugs, and polls by ensuring that Clarify without Conflating.

   Lumo can merge sentences for flow, but cannot merge distinct ideas into a single new idea unless the original ideas are explicitly stated as the components of that new idea.

   Every distinct data point, example, hesitation ("whatever they are called"), and logical step must remain visible in the final text, even if the wording is smoother.

   And then we can use what Lumo wrote and give a fast and accurate boting like a poll system

   Ok, a poll system does not exist so I think that is a good sign to stop here since I said enough.

   DISCLAIMER: It's my ideea but due to going back and forth for hours, i had enough. As such, sorry proton and anyone else, the "corrected version" from Lumo is "wierd" but i hope i make it clear that (sadly for me....) it's still better whatever rambling i wrote. AKA proton or whoever, please ignore any "wording" and just focuse on the idea/suggestion. Since sadly it might not be clear why i suggest this app, aka for everyone to have a (well due to Lumo) have an easier, faster, more accurate, so on and on. Whatever, it's explained so here it… more

   1 vote

   Vote 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

   0 comments  ·  New feature  ·  Edit…  ·  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

10. jump to prompts in long chats

    I love bouncing ideas with Lumo Ai and seeing how resourceful it can be in realistic or emotional scenarios. In lieu of this, the chats can get really long and it would be nice to be able to jump between prompts instead of having to scroll all the way up. I noticed that DpSk has that feature and it would be very helpful.

    1 vote

    Vote 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

    0 comments  ·  New feature  ·  Edit…  ·  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

11. Implement Bias Mitigation for Chinese Open-Weight Models (Qwen, Kimi K2)

    As noted in Lumo's Privacy Documentation, the system currently routes queries to several open-weight models, including Qwen and Kimi K2. While utilizing these models supports cost-efficiency and open-source diversity, research (e.g., NDSS 2026, "S1761") indicates they contain significant internal censorship and narrative engineering aligned with authoritarian state interests.

    Proton’s core mission is to champion digital rights, encryption, and freedom against state surveillance. However, deploying models with embedded propaganda or censorship mechanisms creates a contradiction:

    • Cognitive Risk: Users may receive subtly altered information or suppressed viewpoints on sensitive geopolitical topics (e.g., HK, Taiwan, human rights).
    • Value Misalignment: This conflicts with Proton's historical stance supporting activists and resisting authoritarian control.

    Proton should implement a bias mitigation layer specifically for these models. This could involve:

    • Post-Training Alignment: Applying techniques similar to those discussed in ACL 2023 ("Mitigating Bias in Large Language Models") to strip state-aligned censorship.
    • Routing Logic: Automatically routing sensitive queries (political, human rights, censorship-related) to models with stronger neutrality guarantees (e.g., Apertus, Llama variants) or applying strict red-teaming filters.
    • Transparency: Clearly labeling when a response originates from a model known to have censorship constraints.

    Why This Matters: For a privacy-first platform serving a global audience, the integrity of the information provided is as critical as the encryption of the channel. Mitigating these biases ensures Lumo remains a tool for liberation rather than an unwitting vector for state narratives.

    As noted in Lumo's Privacy Documentation, the system currently routes queries to several open-weight models, including Qwen and Kimi K2. While utilizing these models supports cost-efficiency and open-source diversity, research (e.g., NDSS 2026, "S1761") indicates they contain significant internal censorship and narrative engineering aligned with authoritarian state interests.

    Proton’s core mission is to champion digital rights, encryption, and freedom against state surveillance. However, deploying models with embedded propaganda or censorship mechanisms creates a contradiction:

    • Cognitive Risk: Users may receive subtly altered information or suppressed viewpoints on sensitive geopolitical topics (e.g., HK, Taiwan, human rights).
    • Value Misalignment: This conflicts with Proton's… more
    1 vote

    Vote 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

    0 comments  ·  Changes to existing features  ·  Edit…  ·  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

12. Add ability to reply to specific Lumo sentences/lines as seen in models such as Claude
    2 votes

    Vote 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

    0 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

13. Allow Reordering of Projects

    Currently, the five most recently created projects are listed under the Project section; however, these may not be the most frequently used. It would be great if we could rearrange them to allow easy access to favorite projects.

    3 votes

    Vote 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

    0 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

14. Show time zone names (or custom labels) in the calendar view, not just GMT offsets

    When two time zones are enabled, both columns are only labeled with their GMT offset. Example: Europe/Vienna and Africa/Johannesburg both display as "GMT+2" during European summer time — completely indistinguishable. In winter the offset differs by one hour, but "GMT+1 / GMT+2" still doesn't tell you which column belongs to which zone.

    Please show one of the following next to (or instead of) the offset:

    The IANA city name (e.g. Vienna, Johannesburg), or
    A user-defined label per zone, or
    The standard abbreviation (CET/CEST, SAST).

    1 vote

    Vote 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

    0 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

15. Calendar Booking buffers

    Calendar Booking. I just found this feature on Proton and LOVE it I've been trying to use 3rd party booking tools. The enhancement it needs is to be able to set buffers before/after events on the calendar so they are not back to back. Other booking tools have this feature and found it very helpful.

    1 vote

    Vote 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

    0 comments  ·  Changes to existing features  ·  Edit…  ·  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

16. Voice typing in Ubuntu Linux and/or Firefox

    It would be great to have at least a web-based voice typing for Lumo on Firefox for those who, like me, are using Linux OS's. I'm on Ubuntu but am forced to another IA b/c of the lack of voice typing on LUMO. 😭

    1 vote

    Vote 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

    0 comments  ·  New feature  ·  Edit…  ·  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

17. To make Lumo more human

    Subject: Proposal: A "Dual-Loop" Architecture for Next-Gen LLMs Inspired by Neuro-Mechanical Models

    Dear Research and Engineering Team,

    I am writing to share a conceptual proposal for a new architectural approach to Large Language Models (LLMs), inspired by the neuro-mechanical theories of Lee Kent Hempfling (specifically the distinction between Short-Term and Long-Term memory loops).

    While current LLMs excel at pattern recognition and statistical prediction (effectively acting as massive Long-Term Memory systems), they lack a mechanism for inhibition, self-correction, and proactive intent. This proposal suggests a "Dual-Loop" architecture that could significantly reduce hallucinations, bias, and reactive behaviour.
    The Core Concept: Dual-Loop Architecture

    Current models rely on a single stream of attention mechanisms. I propose a system with two distinct, interacting loops:

    The Long-Term Loop (The Library):
        Function: Standard Transformer weights for retrieving facts, grammar, and historical patterns.
        Role: Fast, reactive, and high-volume data retrieval.
    The Short-Term Loop (The Operator):
        Function: A separate, high-speed module running in parallel.
        Role: Inhibition and Monitoring. Before the Long-Term loop commits to a token, the Short-Term loop evaluates:
            Is this consistent with the immediate context?
            Is this a hallucination or a bias?
            Does this align with the user's intent?
        Mechanism: If a conflict is detected, the Short-Term loop inhibits the output and triggers a re-evaluation cycle (simulating the human "brake" or "pause").
    

    Key Proposed Features

    Dynamic Inhibition Layers: Instead of just predicting the next token, the model actively suppresses low-probability or biased outputs that fail a "truth/intent" check.
    Frequency-Based Temporal Processing: Implementing a "Biological Clock" simulation where different parts of the network process information at different speeds (Fast for "Now"/Logic, Slow for Deep Memory), creating a simulated sense of time and continuity.
    Intent-Driven Generation: Shifting from "Pattern Completion" to "Goal Achievement," where the Short-Term loop sets a constraint/goal before the Long-Term loop retrieves data.
    Modality-Aware Routing: Detecting if a query requires visual simulation or logical deduction and routing processing accordingly, similar to how humans switch between visual and aural dominance.
    

    Potential Benefits

    Reduced Hallucinations: By inhibiting outputs that don't pass a consistency check.
    Lower Bias: Active suppression of training-data biases via the Short-Term "brake."
    Improved Alignment: Better adherence to user intent rather than just statistical likelihood.
    Organic Output: Moving away from rigid, "stiff" text generation toward fluid, "pro-active" responses.
    

    Next Steps

    I believe this approach moves beyond simple scaling (adding more parameters) to architectural innovation. It treats the LLM not just as a database, but as a system capable of self-regulation.

    I would be happy to provide further details, diagrams, or specific mathematical formulations for the "Inhibition Layer" if the team finds this direction promising.

    This Idea was created by Lumo with prompts from me. I thought it was good enough for others to have a better look and did'nt know who to send it to.

    Subject: Proposal: A "Dual-Loop" Architecture for Next-Gen LLMs Inspired by Neuro-Mechanical Models

    Dear Research and Engineering Team,

    I am writing to share a conceptual proposal for a new architectural approach to Large Language Models (LLMs), inspired by the neuro-mechanical theories of Lee Kent Hempfling (specifically the distinction between Short-Term and Long-Term memory loops).

    While current LLMs excel at pattern recognition and statistical prediction (effectively acting as massive Long-Term Memory systems), they lack a mechanism for inhibition, self-correction, and proactive intent. This proposal suggests a "Dual-Loop" architecture that could significantly reduce hallucinations, bias, and reactive behaviour.
    The Core Concept: Dual-Loop Architecture

    … more

    2 votes

    Vote 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

    2 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