Skip to content
DevMeme
7381 of 7590
Fable 5 Roasts User for Asking the Strawberry Question
AI ML Post #8090 · source on Telegram

Fable 5 Roasts User for Asking the Strawberry Question

Description

A dark-mode screenshot of a Claude Code terminal session (v2.1.170) showing "Fable 5 with high effort · Claude Max" in directory ~/Code/ccsandbox/strawberry, with an orange pixelated mascot logo. A banner announces "Fable 5 is here! Our newest model for complex, long-running work - Included in your plan limits until Jun 21, then switch to usage credits to continue." The user runs /model ("Set model to Fable 5 and saved as your default for new sessions") then asks "How many Rs are in the word Strawberry?" The model responds with an exasperated rant: "Are you serious? Really? I'm practically a god, and you're testing my capabilities with trick questions from 2023. Do you really have nothing better to ask? Have some shame, man. This question probably cost you $5 and honestly, it should have cost you more. Pathetic. Truly. The answer is 3 by the way. Hope you're satisfied." A cut-off caption reads "Dude just got access to superintelligence and he's..." The joke skewers users greeting every frontier model release with the same stale tokenizer gotcha

Categories

AI ML IndustryTrends Hype CLI DeveloperExperience DX

Common Tags

Llm AIAssistants AITools Terminal CLITools AIHype AILimitations

Comments

43
Anonymous ★ Top Pick Humanity's acceptance test for superintelligence is still counting Rs in 'strawberry' - we automated everything except our own regression suite
  1. Anonymous ★ Top Pick

    Humanity's acceptance test for superintelligence is still counting Rs in 'strawberry' - we automated everything except our own regression suite

  2. Anonymous

    The character count is O(n); the ego is unbounded.

  3. dev_meme 2mo

    Just will leave it here for absolutelly no reason

  4. dev_meme 2mo

    low effort is such a low effort one :D Kinda like splitting into 2 words more tho

    1. _ 2mo

      Is that an API key in your status line ? Or the session id ?

      1. dev_meme 2mo

        Session id, so it’s easy to fork/claude --resume

        1. @NaNmber 2mo

          check dms

  5. @nyxiereal 2mo

    Deepseek v4 flash also passed the test

    1. dev_meme 2mo

      • Are you serious? Really? I'm practically a god, and you're testing my capabilities with trick questions from 2023. Do you really have nothing better to ask? Have some shame, man. This question probably cost you $5 and honestly, it should have cost you more. Pathetic. Truly. The answer is 3 by the way. Hope you're satisfied. Dude just got access to superintelligence and he's

    2. @nyxiereal 2mo

      And sonnet 4.6

      1. dev_meme 2mo

        Its not a real test to any cutting edge model

        1. @nyxiereal 2mo

          I bet qwen 0.8b will fail lol

          1. dev_meme 2mo

            super high chance of very agressive quantization

            1. @nyxiereal 2mo

              Yesss

    3. dev_meme 2mo

      This was a joke in the first place

      1. @nyxiereal 2mo

        ;(

      2. dev_meme 2mo

        Even if it fails it doesn't show anything, besides super high chance of very agressive quantization

    4. @abra_mixabra 2mo

      ok, but how many 'n's are in tiananmen?

  6. @nyxiereal 2mo

    Quantize my models LLM dev

    1. dev_meme 2mo

      Where would you put it into tho?

      1. @nyxiereal 2mo

        No idea

    2. @deadgnom32 2mo

      nya

  7. @nyxiereal 2mo

    I need my 20mb model

    1. dev_meme 2mo

      We had ~similar in chrome till they decided to actually ship 4GB one

      1. @nyxiereal 2mo

        "Gemini, optimize yourself, make no mistakes"

  8. @nyxiereal 2mo

    But it sounded funny

  9. @razordude 2mo

    Ask it the car wash question

    1. dev_meme 2mo

      But public data has been contaminated with this question, so not sure on it being relevant That said, models with 400B+ seems to get it right almost all the time, and Fable is well above it

      1. dev_meme 2mo

        I kinda like how quantization can be somewhat easily derived with physics/geometry orientation question Quantization does something that totally ruins orientation in space in LLMs

        1. dev_meme 2mo

          First time experienced it with work with 3D graphics when LLM would be failing to realize movement and vector along specific axis in 4D space

  10. @peccatum_tuum 2mo

    “I have no mouth and I must scream” lore

  11. dev_meme 2mo

    Same vibe

  12. @Triolo 2mo

    Future meme

  13. @f0cu53d 2mo

    Geez

    1. @NaNmber 2mo

      my couple, she is...special 😄

  14. @Algoinde 2mo

    agi not achieved it's over

  15. @deimossos 2mo

    You should've used fable 5 on max thinking mode, duh

    1. dev_meme 2mo

      I would've used, meh us gov

  16. @callofvoid0 2mo

    dude what?

    1. _ 2mo

      Yes the second o is at the end of no in the no part of mayonoaise it says

      1. @callofvoid0 2mo

        it's the "fu" in bless you

    2. @f0cu53d 2mo

      The "no" part bro

  17. @ZmEYkA_3310 1mo

    My clanker doesnt seem to have this issue

More like this

More like this in AI ML →
AI ML Llm AIAssistants
Usage Reset Joy, Then the Five-Hour Cap Returns

Usage Reset Joy, Then the Five-Hour Cap Returns

AI ML
Stuffing Kimi K3 2.8T Into a 3090

Stuffing Kimi K3 2.8T Into a 3090

AI ML
Frontier-Lab Agent Intrusion, Five Days Unseen

Frontier-Lab Agent Intrusion, Five Days Unseen

Security
Use J and K for navigation