Dylan Patel, founder of SemiAnalysis, posts parody image of aggressive AI safety filters blocking a benign raspberry query

@dylan522p How dare you ask such a question

@dylan522p It’s like four Loko without the caffeine

It’s just a sad replacement

@dylan522p works fine for me

I love Fable 5 and Anthropic

@dylan522p I am usually favorable towards Anthropic and rate them as the most trustworthy among all AI makers but I can't get a single prompt through Fable 5. It is ridiculous at the moment. Extremely frustrating. Not even worth trying.

@zdrks @dylan522p Yes BUT THE CONTENT!!!

@dylan522p SAFTEYMAXING.

@dylan522p strong combo to love. what part of Fable 5 is actually winning for you?

@dylan522p Is this a real post?

@dylan522p How many Rs in “regulatory capture”?

@dylan522p "I'm sorry, Dylan. I'm afraid I can't count your raspberries."

@i_mika_el @dylan522p honestly the speed feels crazy smooth compared to most ai tools rn

@dylan522p It has a tool to end the conversation if it wants, and it clearly feels your questions are beneath it.

@dylan522p cloudy with a chance of super intelligent strawberries

@dylan522p Lol. Got the same when asking about research on first principles thinking to harness AIs full potential.

That simple, innocent prompt was enough to get me instantly downgraded to 4.8.

The guardrails they put on this thing are absolutely ridiculous.

@dylan522p lol! 😂

“Safety” is for the powerful to control. That’s all it has ever been.

I am rethinking everything about “safety” in all things. •

@dylan522p

@dylan522p GPT 5.6 can't arrive fast enough, afaic.

@dylan522p So smart it has to be limited to safe queries

So dumb it can't figure out what's a safe query

@dylan522p "How many o's in biology?"

gg