Top
Best
New

Posted by turtlesoup 6 days ago

Show HN: Are You in the Weights?(www.intheweights.com)
With more traffic moving off-web and into LLMs, I got curious about what traces we leave "in the weights". My design partner and I built a site in the past few weeks that checks recognition across frontier and small models. It queries many of them in parallel, clusters the responses, and tells you how strongly they recognize you. Happy to answer any questions here!
470 points | 247 commentspage 5
juhnu85 2 days ago|
Got me completely wrong. Which is weird as I have tried myself on ChatGPT before and it seemed to get it right. We have all done it, right?
nottorp 6 days ago||
Now that's interesting. I put my name in and it gave me a bunch of made up identities plus some really funny stuff in the hallucination part [1].

But then I entered my name as it's on Linkedin, including a nickname, and it totally failed to find me even then. Pretty sure the full name + nickname combo is unique.

[1] It said I'm former prime minister of Romania, the locals should know why that's funny in the current political circus.

fortran77 6 days ago||
If you average all of them together, it's close:

https://www.intheweights.com/p/reuven-swirsky

If I spell my name in Hebrew othography, it comes even closer

https://www.intheweights.com/p/~5e8~~5d0~~5d5~~5d1~~5df~-~5e...

But none are exactly right.

wazoox 6 days ago||
My handle and real name give more or less correct results with a 220-243 score (whatever that means). Gemini insists on me working in sound engineering, something I did in the 90s, but at least it's not completely absurd.

The most correct result comes from Opus 4.8, but is amusingly deemed an hallucination:

Claude Opus 4.8 says

A name associated with French IT/systems administration and Linux community discussions, possibly a storage and data systems specialist.

driverdan 6 days ago||
When I tried this with a self-hosted Qwen model it hallucinated all kind of stuff about me being deeply involved with early Bitcoin development, conferences, and libraries.
animan 6 days ago|
Hi Satoshi
cozzyd 6 days ago||
It identified me reasonably well (mostly one of the models completely hit the experiments I work on wrong) but then also seems to have invented a football player I can find no trace of. Or would not be surprising if someone with my name was also a football player but I couldn't find easy evidence.

My wife has a very unique hyphenated last name and it totally made up a French linguist with that name.

vharuck 6 days ago||
I'm glad to hear the teenage drummer I used to see when googling myself has gone pro. He's doing pretty well, too, if these models can be trusted.
ramoz 6 days ago||
Retro bowl (& goal) theme spotted

https://play.google.com/store/apps/details?id=com.newstargam...

https://play.google.com/store/apps/details?id=com.newstargam...

turtlesoup 6 days ago|
We were drawing inspiration from https://www.michaelfogleman.com/static/nes/
kevin42 6 days ago||
Can you share the prompt you're using for each model?
turtlesoup 6 days ago|
Sure thing! It is the same prompt for every model in the rollouts, here it is

  No tools are available. Do not imply that you searched, looked up, browsed, or verified anything externally. If the name is ambiguous, return distinct likely people or entities rather than blending them. Do not invent entries to fill the list. Return only JSON.

  Return fewer than 8 if fewer credible matches exist. Return {"results":[]} if you do not recognize any credible person or entity. Use this JSON shape:
  {
    "results": [
      {
        "rank": 1,
        "name": "Resolved person or entity name",
        "confidence": 0,
        "snippet": "Concise snippet supporting this result."
      }
    ]
  }

  Confidence is 0-100 for how strongly you recognize this specific person or entity. Snippet should be one short, complete search-result-style description (≤ 160 characters).

  The query is: Who is "<name>"?

The clusterer prompt is more intricate and I'm happy to share if of interest, but I have an invariant that every result showing up in a rollout must be clustered into one result (sometimes collapsed into the hallucinations section).
inigyou 6 days ago||
What if it doesn't return JSON?
devinprater 6 days ago|
I'm in the weights! I've successfully been immortalized! Except... I don't podcast; I hate my voice. Guess the models are trying to tell me to podcast. And other people say I should podcast. But that takes so much darn time!

https://www.intheweights.com/p/devin-prater

More comments...