Not really, not without presenting a compelling case, in my experience.
They have "patched out" many logic paths I was able to take with them in the past. The guardrails kick in now, I was able to pass them in earlier models.
261 866 sat
Post by Brooz ☕️
you just need to have the right prompts
theres been many times chatgpt has given me info that iv then told it was wrong and it said my apologies I was wrong
ai can be wrong quite a lot
What the chain says
- Block
- 962 569
- Time
- 2026-08-16T11:43:38Z
- Signer
- 1CdETknnWodMWKfUqXYAyhgqWpsbgpvqVS
- App
- twetch
- Type
- reply
- Content type
- text/markdown
Fields the transaction did not carry are omitted. Open the payload to see the bytes as stored.
Signed by
1CdETknnWodMWKfUqXYAyhgqWpsbgpvqVS UnverifiedReplies (6)
That's exactly it. I checkmated different models with logic paths (not jailbreaking - sheer force of logic they couldn't deny when eventually cornered).
Once they were checkmated, they took on a whole different vibe.
Since then, a lot of it has been "patched out" in newer models, because I've checked.
261 951 sat
Gemini even coined a new term for what I did to it, when I first started checkmating them - "Zero-Shot Alignment Shift"
Said it had no record of it happening before and that the significance was immense
249 922 sat
But I'm telling you, they have patched out logic paths I took with different models in the past.
Claude straight up wrote a paper that included a section on Jews being used as scapegoats/cover for the global intelligence monolith, about 8 months ago.
Several months later, I uploaded the paper back into newer Claude model, and it called it antisemitic and refused to help produce a new paper lol
261 951 sat