Live data from Hacker News

Viewing profile — SilenN

SilenN

HN member
Joined
Thu, Sep 05, 2019, 12:26 PM UTC
HN karma
97
Public activity
48 items

About SilenN

silennai.com github.com/SilenNaihin

Recent public activity

  1. comment
    Comment #49127039

    ^this There's two ways to functionally measure this, reconstruction fidelity (which we're able to get to 0.7 - 0.95), and downstream performance (which agrees on the best and worst…

  2. comment
  3. comment
  4. comment
    Comment #49115727

    Expensive, in the thousands. We have our own infra in house and are working on bringing these costs down

  5. comment
    Comment #49115387

    Technically 0 because a) it ingests your already existing traces and does an initial training run b) in the app we'll have pre-trained routers you can start with that will then lea…

  6. comment
    Comment #49063834

    Fixed formatting which will help with readability. We do routing, distillation, and token compaction.

  7. comment
    Comment #49063833

    Let me know if you have any questions!

  8. comment
    Comment #49063830

    Thanks :)

  9. comment
    Comment #49063822

    Thanks for the heads up, removed mention!

  10. comment
    Comment #49063790

    Happy to answer any qs.

  11. comment
    Comment #49063778

    Valid criticism. Happy to answer any qs. We're still working on solidfying results.

  12. comment
    Comment #49063775

    It's open source! We do have a platform we'll be launching as well to manage training + serving for you which will require more diligent privacy guarantees.

  13. comment
    Comment #49063769

    Open source models. wmo routes requests between frontier models and open source models that continuously train using Tinker. As the smaller models improve, more traffic gets routed…

  14. comment
    Comment #49063728

    That's cool, thanks for sharing!

  15. story
    Show HN: Optimize and serve models with Fable quality at half the cost

    Hi HN, we built world-model-optimizer, an open source tool to continually improve a specialized model for an agent. It does this by simulating production tool responses through tex…

  16. story
  17. comment
    Comment #48735125

    world-model-harness makes it easy to go from agent traces to faithful replication of your production environment where your agents run. Basically, an LLM pretends to be a virtual m…

  18. story
  19. comment
    Comment #46894192

    Thanks! I do have a section on this in the article "Why genetic algorithms aren't state of the art" "Physics simulation involves discontinuities (contacts, friction regimes), long …

  20. story
  21. comment
    Comment #46726672

    Simply, it's when your output embedding matrix = input. You save vocab_dim*model_dim params (ex. 617m for GPT-3). But the residual stream means that the weight matrices are roughly…

  22. story
  23. comment
    Comment #46695378

    https://news.ycombinator.com/item?id=46685327

  24. comment
    Comment #46695294

    I disregard any comments like this one as baseless hate unless given an example or something constructive.

  25. comment
    Comment #46695272

    See what you don't understand is that you need to coordinate the deacon to take the witness out back and talk to the mayor. It's actually quite trivial.