‹ BackHN Continuity

Thread

Jensen Huang says AI distillation is 'competition.'

73 points · 81 comments · cramer4next

  1. prodigycorp · · focus · HN ↗
    Can someone answer this: if anthropic cares so much, why don't they meganerf their thinking summaries (edit: in Claude Code) the way OpenAI does? OpenAI's thinking traces so much that the Codex app doesnt even show them.
    1. aqfamnzc · · focus · HN ↗
      Haven't they already? (at least in the regular chat UI)
      1. prodigycorp · · focus · HN ↗
        It's not there in the chatui but claude code shows more reasoning than they need to.

        im not suggesting that they should, but it's so much more than the garbage both antigravity and codex show to users. seems like a thing a company paranoid about distillation would reign in.

    2. Catloafdev · · focus · HN ↗
      Thinking traces aren't in the Claude app (at least, not visibly). I assumed it's only available (if at all) via API.
      1. prodigycorp · · focus · HN ↗
        they dont show thoughts in chat, but they do in code mode.
        1. Catloafdev · · focus · HN ↗
          Which version are you using? Using code mode in the desktop Claude app, I do not see thinking traces at all, and I don't see any way to toggle it.
          1. prodigycorp · · focus · HN ↗
            latest. click on the down caret next to the session title. hover over transcript view and click on 'thinking'
    3. arcanemachiner · · focus · HN ↗
      Pretty sure those are just summarized traces, which obscure the valuable reasoning traces used by the actual model.
      1. prodigycorp · · focus · HN ↗
        yeah i know they are summarized/condensed, but they are still an order of magnitude more useful that what they competitors present to users.
    4. ch_sm · · focus · HN ↗
      In Claude Code (at least inside VSCode) thinking traces are hidden by default. You can reenable them via JSON. They’re not the real thinking traces though, they are something like thinking summaries.
      1. prodigycorp · · focus · HN ↗
        I'm guessing the answer to this is that, despite being much more than what codex provides, claude current thinking summaries are not useful for training a frontier model.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.