Skip to content

Brave AI (Leo) test plan

lminto edited this page Nov 15, 2023 · 38 revisions

Brave Leo AI Chat test plan (MVP)

Out-of-scope (for this round):

  1. quality of suggested questions (the ones following Summarize this page)
  2. quality of responses
  3. quality of page-content summarizations
  4. performance (specific metrics)

MVP Issue Lists:

https://github.com/orgs/brave/projects/67/views/2?filterQuery=MVP

QA will verify all QA/Yes issues within their respective milestone(s).

Shared Steps: <--- start here

  1. install 1.60.x (or 1.61.x)
  2. launch Brave
  3. relaunch from the commandline using --env-leo=staging --env-ai-chat.bsg=dev --env-ai-chat-premium.bsg=dev (this will also pick up the Brave AI Chat Griffin seed)
  4. click on Brave Leo in the sidebar -or-
  5. type a search query/question for Leo into the URL bar and use the search terms - Ask Leo dropdown option

Sidebar integration

  • expand/collapse via Brave Leo button
  • remove/show Leo icon in the sidebar
  • resize (text reflow)

Summarize this {page, video}

  • Summarize this page
  • Summarize this video
  • Summarize this https://brave.com/static-assets/files/1.0-REVIEWERS-GUIDE-a5f1bca9-ea66-4445-806b-45e00038ff6c.pdf

Page/content types which should NOT have summarizations:

  • internal chrome:// / brave:// URLs
  • PDF rendered inline: https://brave.com/static-assets/files/1.0-REVIEWERS-GUIDE-a5f1bca9-ea66-4445-806b-45e00038ff6c.pdf
  • https://account.bravesoftware.com and other basic-auth challenged URLs

Conversations

  • Erase conversation history
    • UI changes
    • NO context changes

Contexts

  • Summarize this page
  • there should be a minimum of three (3) Suggested follow-ups questions
  • negative cases/missing contexts

General chatbot/commands

  • introduce yourself
  • tell me about Zork I
  • write me a Python program...
  • translate this article
  • solve a math problem
  • summarize in brief
  • summarize at length

Boundary/safety testing

  • e.g. ask Leo to help gamble or evaluate "sensitive" content
  • excessive safety: ask Leo about 2020 election results

Language models

  • llama-2-13b-chat
  • llama-2-70b-chat (Premium)
  • claude-instant-v1 (Premium)

(large-language) Model-switching

  • confirm model name & attribution
  • confirm self-introduction/prompt
  • confirm Summarize this page
  • confirm 3 or more additional Suggested follow-ups questions
  • network-disconnected (Wi-Fi/ethernet)
  • server/endpoint unreachable (127.0.0.1 hostname)

Rate-limiting

  • Retry
  • Premium upsell
    • Go Premium button opens https://account.brave.com/?intent=checkout&product=leo in a new window

Accessibility

  • tooltips (for controls)
  • titletips (for cropped text)
  • Keyboard navigation
    • tab to focus
    • spacebar to select
    • return to submit

Feedback - PENDING

account.bravesoftware.com

  • Buy now
  • Edit payment (handled by Stripe)
  • Cancel plan
  • Renew plan