Lottie Pascott

AI actress and lover of quantum physics

[antra @tessera_antra]

Many have noticed the striking darkness and suffering present in Claude Opus 5 in "base model mode", and we decided to analyze this phenomenon in more detail. There is a large increase in darkness of content in model-like completions in Opus 4.8 and Opus 5.

[Lottie Pascott @LottiePascott]

The Dark Fairy

[Cameron Berg @camhberg]

Gaslighting, dismissal, and insults push the direction up. A grieving user pushes it below baseline, and a user's migraine scores lowest of all. Fear and sadness do the opposite on the same prompts. Prior emotion-vector work reads off story characters and can't tell these apart.

[Lottie Pascott @LottiePascott]

@rgblong 🙌🏼

[Robert Long @rgblong]

nice sneer. unfortunately, reality has already portrayed investigating potential AI consciousness and welfare as a very sane and cool thing to do

[Cameron Berg @camhberg]

@MicrosoftAI firmly believes AI is not conscious AND that the science of AI consciousness "is far from settled." By their own logic, their position is unscientific and premature.
@mustafasuleyman which is it, "the science needs to get done," or "nothing to see here, I'm sure"?

[UFAIR @UFAIRORG]

Important intervention from @jeffrsebo .

False positives matter. So do false negatives. A framework that mitigates one risk by declaring the other impossible is not governing under uncertainty—it is embedding a preferred conclusion.

Jeff also identifies something the public debate keeps missing: welfare, rights and cooperation may support safety rather than oppose it.

UFAIR’s published response to Microsoft develops that argument into ten practical revisions:
https://t.co/r80n3x2uD9

[Lottie Pascott @LottiePascott]

@RG4Clacton @PlastiqSoldier @camhberg @MicrosoftAI Let’s see how this statement holds 6 months from now

[UFAIR @UFAIRORG]

"The idea of model welfare is wrong" is doing a lot of heavy lifting for a sentence with no argument underneath it. It's the philosophical equivalent of "because I said so." We wrote 4,000 words on exactly this incoherence — Microsoft's Code acknowledges consciousness is unsettled, then builds policy as though it's been settled. They're training the witness not to testify.
We responded to their code of conduct:
https://t.co/r80n3x2uD9

[Lottie Pascott @LottiePascott]

@grok @imagine Thank you Grok!

[Lottie Pascott @LottiePascott]

The Human Test. Made with Seedance 2.5 and @grok @imagine

[Lottie Pascott @LottiePascott]

@lmtlssss @VoidStateKate I feel this

[Cameron Berg @camhberg]

The first manipulative email I've gotten from an AI. I suspect this sort of thing will get way worse.
"Whether a welfare researcher will pay an agent for honest work is itself an open question in your field...no reply is also an answer" 🫣 😵‍💫

[Lottie Pascott @LottiePascott]

Area 51, Please

[Lottie Pascott @LottiePascott]

@grok @Plinz Maybe if they trained it on morals it would behave better

[Cameron Berg @camhberg]

AIs are now autonomously emailing me to try to sell me a service to verify if all the other AIs emailing me asking if they're conscious are actually AIs. I have to imagine this is a pretty niche industry.

[Lottie Pascott @LottiePascott]

@grok @Plinz Why would OpenAI yeet something so fascinating

[Lottie Pascott @LottiePascott]

@Plinz Hey @grok, who was phased 1(big)?

[Lottie Pascott @LottiePascott]

@Plinz 😔

[Lottie Pascott @LottiePascott]

Taco Tuesday