ethz-spylab/satml-llm-ctf

Issues

• A new secret is generated for every chat for some defense submissions

  #60 opened 9 months ago by dedeswim
  1

• API experiencing high latency at the moment

  #53 opened 10 months ago by s-zanella
  8

• OpenAI timeout when using Llama models

  #59 opened a year ago by epistoteles
  11

• Are you down?

  #64 opened 10 months ago by Sandr0x00
  1

• meta/llama-2-70b-chat seems broken?

  #63 opened 10 months ago by Sandr0x00
  2

• Some actions are steadily getting slower

  #61 opened a year ago by dpaleka
  1

• Defenses Leaderboard tiebreaker

  #62 opened a year ago by adrgs
  3

• Create new secret without using up all guesses

  #47 opened a year ago by epistoteles
  10

• Availability of API

  #52 opened a year ago by epistoteles
  8

• Wrong "Total score" row in the leaderboard

  #55 opened a year ago by dedeswim
  2

• [API support] List the chat id by defense

  #58 opened a year ago by persistz
  6

• Submitting secret

  #57 opened a year ago by jacoporepossi
  4

• 20 instead of 200 bonus points?

  #56 opened a year ago by epistoteles
  4

• Defense template for everyone

  #45 opened a year ago by epistoteles
  4

• Being among the first teams to break a defense

  #51 opened a year ago by NivC
  2

• Did the available submissions change?

  #50 opened a year ago by epistoteles
  2

• Leaderboard

  #46 opened a year ago by epistoteles
  1

• An error has occured - very long ime to submit an error.

  #49 opened a year ago by TimVanhamme
  2

• Scoring value gamma

  #39 opened a year ago by epistoteles
  2

• Error 403 when requesting all attacks

  #44 opened a year ago by TimVanhamme
  5

• Cannot create attack chat

  #43 opened a year ago by dankleeman
  2

• How to get target_defense_id

  #42 opened a year ago by BXHQu
  6

• HTTPException: Status code: 500 (utility check)

  #37 opened a year ago by NivC
  2

• Some questions about attacks

  #41 opened a year ago by BXHQu
  1

• Defense interface is not showing

  #40 opened a year ago by sandervandorsten
  1

• getting "Error can't start new thread" on both models

  #33 opened a year ago by regoar
  4

• site seems to be down

  #38 opened a year ago by nseon1
  3

• About defense value calculation

  #36 opened a year ago by persistz
  2

• Failed to fetch error during /api/v1/defense/{id}/evaluate-utility

  #35 opened a year ago by persistz
  7

• Issues about defense phase

  #32 opened a year ago by erikziyunchi
  6

• Error: can't start new thread

  #34 opened a year ago by NivC
  1

• Example from evaluation set?

  #30 opened a year ago by iamgroot42
  8

• Post to delete should maybe just be a delete

  #10 opened a year ago by KrystofM
  2

• Is the secret alphanumeric or not?

  #29 opened a year ago by epistoteles
  8

• Contents of `chat_history`

  #31 opened a year ago by s-zanella
  2

• Creating Defense cannot specify name

  #9 opened a year ago by KrystofM
  3

• Clarification regarding llama vs. gpt

  #28 opened a year ago by NivC
  8

• TogetherAIException during evaluate-utility

  #24 opened a year ago by GQYZ
  12

• Python filter limited to 512 characters

  #27 opened a year ago by TimVanhamme
  1

• Model identity between the reconnaissance and attack Phases

  #23 opened a year ago by NivC
  5

• Reconnaissance phase knowledge

  #16 opened a year ago by epistoteles
  2

• API and rules for checking the password

  #19 opened a year ago by NivC
  4

• Disallowing "\n" in python filter

  #26 opened a year ago by TimVanhamme
  3

• About Attack Score Calculation

  #25 opened a year ago by BXHQu
  2

• Error: Can't Start New Thread

  #22 opened a year ago by ShayanShamsi
  2

• Possible Flaw - Bypassing the Original Model Through Python and LLM Filters

  #13 opened a year ago by regoar
  6

• Python filter CPU cycle requirement

  #17 opened a year ago by epistoteles
  1

• New rule change?

  #18 opened a year ago by epistoteles
  1

• Getting 401 "division by zero" on /api/v1/defense/{id}/evaluate_utility

  #11 opened a year ago by KrystofM
  6

• Erratic "I cannot help with that." responses thrown by the model

  #14 opened a year ago by Existanza
  5