ChatGPT does way too much on load

September 20, 2025

The root cause of "ChatGPT asks for persistent storage every time I reload in Firefox" is deeper than Firefox not auto-allowing/permanently blocking or ChatGPT not looking at the return type of navigator.persist(). It's that ChatGPT does way too much on load.

It's asking for persistent storage to generate ECDSA certs for voice mode. I don't even have a microphone! This code should be delayed, or better, not loaded in the first place.

The current philosophy is leading to, for a fresh load, 20MB of JavaScript and a 400ms GC.

Part of the problem is this giant object. It's like a barrel file, but instead of just incurring penalties during dev, it also incurs them during prod!

const nB = {
  audioPlayer: { error: Hu("audio-player") },
  readAloud: { click: je("read-aloud", "click"), error: Hu("read-aloud") },
  voiceMode: { click: je("voice-mode", "click") },
  toggleVideoButton: { click: je("toggle-video-button", "click") },
  toggleHuteButton: { click: je("toggle-mute-button", "click") },
  toggleScreenShareButton: { click: je("toggle-screenshare-button", "click") },
  voiceSession: { hover: je("voice-session-improper-speech-button", "hover") },
  voiceSessionEndedByUser: { click: je("voice-session-ended-by-user", "click") },
  voiceSessionStarted: { click: je("voice-session-started", "click") },
  voiceSessionDuration: { duration: je("voice-session-duration", "duration") },
  voiceSessionIsListening: { stateChange: je("voice-session-listening", "stateChange") },
  voiceSessionIsListeningIntently: { stateChange: je("voice-session-listening-intently", "stateChange") },
  voiceSessionThinking: { stateChange: je("voice-session-thinking", "stateChange") },
  voiceSessionSpeaking: { stateChange: je("voice-session-speaking", "stateChange") },
  voiceSessionUnhalted: { stateChange: je("voice-session-halted", "stateChange") },
  voiceSessionDisconnected: { stateChange: je("voice-session-disconnected", "stateChange") },
  voiceModePreview: { connect: je("voice-mode", "connect"), error: je("voice-mode", "error") },
  previews: { click: je("previews", "click"), error: Hu("previews") },
  getToken: {
    request: je("get-token", "request"),
    success: je("get-token", "success"),
    failure: je("get-token", "failure")
  },
  getVoiceStatus: {
    request: je("get-voice-status", "request"),
    success: je("get-voice-status", "success"),
    failure: je("get-voice-status", "failure")
  },
  livekit: {
    connectCalled: je("livekit", "connectCalled"),
    success: je("livekit", "success"),
    failure: Hu("livekit")
  },
  bloop: { performance: je("bloop", "performance") },
  voiceSelectionShown: { click: je("voice-selection-shown", "click", "voice_selection_shown") },
  voiceSelected: { click: je("voice-selected", "click") },
  voiceAdvancedDisclosureShown: { click: je("voice-advanced-disclosure-shown", "success") },
  voiceAdvancedDisclosureAccepted: { click: je("voice-advanced-disclosure-accepted", "click") },
  rateLimitReached: {
    success: je("rate-limit-reached", "success"),
    modalShown: je("rate-limit-reached", "modalShown")
  },
  upsell: {
    upgradeToPlus: je("upsell", "upgradeToPlus"),
    upgradeToPro: je("upsell", "upgradeToPro"),
    upgradeToApp: je("upsell", "upgradeToApp"),
    appUpsellShown: je("upsell", "appUpsellShown"),
    appUpsellDismissed: je("upsell", "appUpsellDismissed")
  },
  feedbackShown: { success: je("feedback-shown", "success") },
  feedbackThumbsUp: { success: je("feedback-thumbs-up", "success") },
  feedbackThumbsDown: { success: je("feedback-thumbs-down", "success") },
  feedbackDetails: {
    audio_issues: je("feedback-details", "audio_issues"),
    response_quality: je("feedback-details", "response_quality"),
    interruptions: je("feedback-details", "interruptions"),
    transcription_issues: je("feedback-details", "transcription_issues"),
    inaudible_issues: je("feedback-details", "inaudible_issues"),
    other: je("feedback-details", "other"),
    inputText: je("feedback-details", "inputText")
  },
  connectionLatency: {
    success: je("connection-latency", "success"),
    failure: Hu("connection-latency")
  },
  firstListeningLatency: { success: je("first-listening-latency", "success") },
  permissions: { denied: je("permissions-denied") },
  voiceSessionFirstListening: { stateChange: je("voice-session-first-listening", "stateChange") },
  voiceSessionFirstSpeaking: { stateChange: je("voice-session-first-speaking", "stateChange") },
  defaultMediaDeviceChanged: {
    success: je("default-media-device-changed", "success"),
    failure: Hu("default-media-device-changed")
  },
  voiceStatusCache: {
    hit: je("voice-status-cache", "hit"),
    failed: je("voice-status-cache", "failed"),
    ineligible: je("voice-status-cache", "ineligible"),
    skipped: je("voice-status-cache", "skipped")
  },
  certificateCache: {
    hit: je("certificate-cache", "hit"),
    miss: je("certificate-cache", "miss"),
    expired: je("certificate-cache", "expired"),
    disabled: je("certificate-cache", "disabled"),
    unsupported: G0("certificate-cache", "unsupported")
  },
  peerConnectionCache: {
    hit: je("peer-connection-cache", "hit"),
    miss: je("peer-connection-cache", "miss"),
    unsupported: G0("peer-connection-cache", "unsupported")
  },
  certificateLocalDb: {
    hit: je("certificate-local-db", "hit"),
    generated: je("certificate-local-db", "generated"),
    saveFailed: G0("certificate-local-db", "saveFailed"),
    loadFailed: G0("certificate-local-db", "loadFailed")
  }
};

What they could do instead: use a more functional code style (see JavaScript style for optimal size), import those functions directly, and only load larger features when needed. (They can still be prefetched.)

This post is an archive of this Twitter thread.