Confessions to a Data Lake

37 points by kkl 146 days ago | 13 comments

gus_massa 144 days ago |

> You could say that LLMs are the first technology where the medium actively invites confession.

What about ELIZA? https://en.wikipedia.org/wiki/ELIZA

How does this work technically?

I am unable to go to their website because:

> "This application requires passkey with PRF extension support for secure encryption key storage. Your browser or device doesn't support these advanced features"

Is this really necessary for a product's webpage? I would understand for the application itself.

3s 139 days ago | |

It uses confidential computing primitives like Intel TDX and NVIDIA CC, available on the latest generations of GPUs. Secure hardware like this is a building block to enable verifiably private computation without having to trust the operator. While Confer hasn’t released the technical details yet, you can see in the web inspector that they use TDX in the backend by examining the attestation logs. This is a similar architecture to what we’ve been developing at Tinfoil (https://tinfoil.sh) if you’re curious to learn more!

TimC123456 139 days ago | |

There's a more-recent post on the same blog that gets into the details of how they're using the WebAuthn PRF extension to store key material, but for platforms and browsers that don't yet support the extension, you'll need a password manager that does. There's a table midway down the post with details: https://confer.to/blog/2025/12/passkey-encryption/

abeyer 138 days ago | | |

This kind of insistence that their way is "better" and thus justifying removing agency from the user is the exact same thing that's kept me away from signal, too. Even their own blog post acknowledges a perfectly good current method for supporting what they want to do without any of this, yet they reject even allowing it as an option because they don't like the ux.

This arguably is more interesting than yet another closed messaging platform, but still not gonna use it with this requirement in place.

viccis 144 days ago |

>You could say that LLMs are the first technology where the medium actively invites confession.

I don't think LLMs really constitute a medium by McLuhan's definition. A medium is an extension of man, and LLMs don't extend so much as they replace. An LLM is closer to a secretary or similar force multiplier than an extension of oneself.

Waterluvian 144 days ago | |

I think it absolutely is a medium based on his big picture idea of how the way we consume information influences us. A TV series or a web search or a university lecture or an evening in the stacks all deeply influence how we might study a subject and in what ways the facts take hold within us. An LLM query or “conversation” is a distinct sibling to those. And a secretary is a medium.

viccis 144 days ago | | |

His idea of media wasn't about "the way we consume information", it's about each medium's ability to extend parts of ourselves. For example, a light bulb is a medium he gives as an example, as it extends our vision. He considers money a medium for extending work and skill.

>And a secretary is a medium.

Not in McLuhan's use of the word. His examples are all pieces of technology that extend our senses; another person helping you do something is not a medium.

I'm not saying you are wrong about the word "medium" in general, but simply that McLuhan had a very specific usage of the word that is not the same as the lay meaning and includes things like light bulbs, numbers, and money. The more modern usage of the word to generically refer to sources of information is very different. In that sense, AI is media insofar as it's feeding us content and information, but it does not meet the specific (and more interesting IMO) definition McLuhan put forward in Understanding Media. Obviously, others can and will disagree with me there, but I simply don't view LLMs are an extension of our thought. Things like the internet, encyclopedias, the education system, etc., are McLuhan-esque media of thought extension, whereas LLMs actually serve the opposite purpose. They abridge rather than extend it by doing it for you.

recursivetree 139 days ago |

Neither the article nor the linked application explain how confer maintains privacy. As much as I'd like to see something like this, I'm still a bit skeptical as long as they can't explain how the technology at the heart of their product works.

exo762 144 days ago |

Post is lacking in technical details. What it seems to be doing echoes the way ChatGPT is integrated into iOS - your requests are anonymized so your profile can't be (easily) built.

How can I make confer.to work on my Linux machine? Modern CPU.

sosodev 144 days ago |

I’ve spent some time thinking about privacy and LLMs. I developed the impression that encryption isn’t meaningful in this space. It seems like end to end encryption only truly works when both ends are outside of the system and can manage their keys independently of it. In this case one end is the system. My message has to be decrypted for processing by the LLM. So is “end to end encryption” in this case any different than HTTPS? It doesn’t seem like it

lisbbb 144 days ago |

I always said our data lake was frozen over. You could get tons of data into it, but barely anything back out.

jmulho 144 days ago | |

That is literally the definition of a data lake. The one you can get anything back out of is called a database.