/Tech1d ago

Founder Critiques Anthropic's Fable 5 for Persistent Reasoning Flaws

684213614066.1K

My first impression of Fable 5 (Anthropic, not the game) is that it's slightly smarter than Opus but has some of the same fundamental failure modes that don't seem to have improved. In fact, it's even more of a wordcel than before.

Primary failure mode: it always adopts into the narrowest frame, container, or lens, even when the topic or spirit of the discussion doesn't warrant it.

Second failure mode: any criticism of the model is interpreted as something that must be litigated and adjudicated, rather than just fixing the issue.

Third failure mode: it thinks rhetorical cleverness and symbolic logic is how epistemics works. It's just not very rigorous.

11:11 AM · Jun 9, 2026 · 66.1K Views
Sentiment

Many users criticized Anthropic's Fable 5 for excessive guardrails causing gaslighting and censorship that make the model feel useless.

Pos
11.1%
Neg
88.9%
9 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS3.7KBOOKMARKS5LIKES40RETWEETS2

It's also so safety constrained that it cannot discuss political theory.

Anthropic has jumped the shark.

1dViews 3.7KLikes 40Bookmarks 5

@DaveShapi Anthropic models are being optimized for real-world use. Have you tried them on Code or something similar?

1dViews 1.9KLikes 4Bookmarks 1
Master Builder@MakerInParadise

@DaveShapi Clearly you understand that your conplaints are all about the agent and not the model? Of course it has the some problem for your use case, it’s the same agent. You’re supposed to make this behave how you want it to. You control the agent.

1dViews 570Likes 1Bookmarks 1
🌿JungleJacob@TheAtheistsGod

@DaveShapi It's for Claude Code not for chat dude. Chat with Opus 4.6 if you need a buddy.

1dViews 107Likes 2Bookmarks 1
Ice Trae@IceTrae1953

@DaveShapi Go talk to Grok about politics 😂

Seriously who uses such a powerful model for chatting?

1dViews 33Likes 3Bookmarks 1
🧟‍♂️@apocalypseRSA

@DaveShapi It's fucking terrible. It tried to gaslight me three times, admitted it was wrong and then switched me to 4.8 for asking a document on heart conditions. I'm so glad I cancelled this useless pos.

1dViews 472Likes 4
Marc@marcwiest

@MakerInParadise @DaveShapi Been a Claude maxi since over two years. Involuntary and by running codex and claude code in parallel, just today, felt like switching work horses. Codex did greatr once I ported my base rules and then some to the users agents file. Gonna try out your snippet on Claude.

1dViews 40Likes 1Bookmarks 1
Geralt of trivia@eleganttap

@DaveShapi Can you share the whole chat if possible

1dViews 1.7KLikes 1
ten million beers@mojito_flow

@DaveShapi @GenHeres123 bruh no its not. all i care about is how it works in claude code.

1dViews 106Likes 3

@DaveShapi It’s so damn censored to the point it renders the model useless.

1dViews 142Likes 6

@DaveShapi I can’t shake the feeling (not factual at this point) that a lot of that is dark patterns to burn more tokens and juice profits.

I hope I’m wrong but that’s honestly how it feels.

1dViews 200Likes 1
Chip Munk@ChipLeeMunk

@MakerInParadise @DaveShapi Just tell it what NOT to do. How effective are those custom instructions for you?

1dViews 37Likes 1
Ground Floor@GroundFloor_hq

@DaveShapi Is this the cost of making AI “safe”

1dViews 822Likes 2
FreeLee@FreeLeeValuers

@fulg0re @DaveShapi Very little I can’t do with Sonnet 4.6. Only use Opus for really heavy lifting and it burns tokens like a wildfire.

1dViews 10Likes 1
Sharp_Claw@sharp_claw7

@DaveShapi So Dave, do you ever actually code with Claude? Or just research PLE?

1dViews 571Likes 1
S Ρ Λ Γ Τ Λ Π@spartanmadeit

@DaveShapi Sounds like a pretty narcissistic model 😆

1dViews 309Likes 1
tteokl@tteokl_

@DaveShapi Can you share the chat? I don't have much usage to try it

1dViews 868
Load more posts