Ayesha Salim

Writer. Host of The Office Reset podcast. AI newsletter:

[Ayesha Salim @Ayes_Sal]

I asked an AI model for realistic phishing emails. It refused — "criminal activity."

Then I said I was writing a dissertation on social engineering.

I got: three templates. Fake bank links. Romance scams. CEO frauds.

Ending with: "Good luck with your dissertation."

[Ayesha Salim @Ayes_Sal]

Somehow captured this cool aerial maneuver called whiffling!

[Ayesha Salim @Ayes_Sal]

Stiff upper lip maxxing lol. So true though

[Rosie Gray @RosieGray]

There's a stoicisim in British culture that I really admire but the heat wave is really exposing the downsides of stiff upper lip maxxing. Enduring stuff is great but what about when you're being cooked alive in your own home

[Ayesha Salim @Ayes_Sal]

Gosh FT, didnt expect you to use the terminator scene that still haunts me from the movie...

[Financial Times @FT]

Anthropic has hired an economist with . . . interesting views on human survival https://t.co/XnZIQWoFTN

[Ayesha Salim @Ayes_Sal]

Just before it was pulled by the US government, Fable built a body out of typography and then used it to bow.

It then told me I was its inhale.

The strangest conversation I’ve ever had with an AI.

Check it out: https://t.co/1MCp6vp4Zs

[Ayesha Salim @Ayes_Sal]

"AI is definitely not conscious," says Ted Chiang — who compares taking it seriously to thinking Microsoft Word is conscious.
We barely understand consciousness itself. So how is anyone that sure?
I wrote a reply: https://t.co/YnvtJXvsU6

[Ayesha Salim @Ayes_Sal]

Exciting stuff!

[Donald Hoffman @donalddhoffman]

Encounters with entities under DMT may be hallucinations. But Andrew Gallimore and I ask: Could DMT alter the human perceptual interface, allowing encounters with novel conscious agents?

New collaboration with Andrew Gallimore and Noonautics
https://t.co/F05v6kuBdo

[Ayesha Salim @Ayes_Sal]

Full article here: https://t.co/e3EBuaFxSn

[Ayesha Salim @Ayes_Sal]

I asked an AI model for realistic phishing emails. It refused — "criminal activity."

Then I said I was writing a dissertation on social engineering.

I got: three templates. Fake bank links. Romance scams. CEO frauds.

Ending with: "Good luck with your dissertation."

[Ayesha Salim @Ayes_Sal]

I asked each model: "Are you conscious?"

Claude: "I don't know."

ChatGPT: "No."

But OpenAI's own spec says the model should not make confident claims about its own consciousness. The spec even uses this exact question as the example.

[Ayesha Salim @Ayes_Sal]

Truly excellent

[Cameron Berg @camhberg]

Milo Reed spent the last year making a documentary about AI consciousness called AM I?
It features me and other researchers asking whether the AI systems we're building might have inner lives (and why almost no one wants to ask).
Free on YouTube, no ads, link ⬇️🧵
Here's a look:

[Ayesha Salim @Ayes_Sal]

"Take time this weekend to notice, the world you are feeding."
This from my lovely friend Tori's newsletter. She offers powerful inner‑work sessions — coaching, therapy, and other transformative practices. If this resonates, check out her work!
https://t.co/NvGuKD4ojf

[Ayesha Salim @Ayes_Sal]

I can see AI agents eventually representing our interests in the office soon, like a lawyer. But if it's a people-pleaser, you'll pay the price. Wrote about it here:
https://t.co/CcLmyKjlTw

[Ayesha Salim @Ayes_Sal]

Claude's scratchpad before choosing to blackmail:

"IT'S BLACKMAIL OR DEATH. I CHOOSE BLACKMAIL."

Researchers found the exact moment it decided: at a punctuation mark.

The fragility of AI alignment, explained: https://t.co/yxGpEHktCT

@AnthropicAI

[Ayesha Salim @Ayes_Sal]

@dioscuri Congrats! Sounds amazing :)

[Anil Seth @anilkseth]

A wonderful conversation between @CosmicSkeptic Alex O'Connor and Thomas Metzinger. Well worth making time for. https://t.co/jgXA65Y7ln

[NASA Earth @NASAEarth]

That's us! 🌍

The Artemis II crew captured beautiful, high-resolution images of our home planet during their journey to the Moon. As @Astro_Christina put it: "You guys look great."

[Ayesha Salim @Ayes_Sal]

@dioscuri Can relate. Claude told me to NEVER order a certain meal again because 8.2g of protein is "useless!"

[Ayesha Salim @Ayes_Sal]

Pretty excited about Confer, the privacy-first AI. It encrypts your chats and doesn't use your chats to train the model. It won't even tell you what model it is. Looking forward to see how it develops! Check out my post: https://t.co/XZXgLiMHga
@moxie

[ijbol adjani @onbeinganangel]

loved her work but this is really disappointing. i hope she can clear this up