|
|
Hi Sparklers!
The people building AI are warning us about the AI they're building. One of them reckons there's more than a 10% chance it kills us within a decade. Read on for more on this.
It wasn't lost on me that, in a week of fear-mongering, I was on stage at the launch of Trinny London's new FEARLESS Skin Tint Drops. The brand's mission is “to give everyone the tools they need to fearlessly feel their best”. Trinny interviewed me, Stacey Duguid and Lorna Luxe about our stories of being fearless, and a make-up launch turned into a conversation about AI and ageism.
Same night, at The F World by FFinc, I argued against the motion “Will our daughters have it harder than we did?” The room started 50/50 and finished 70/30 against. More soon on why I'm so sure they won't. Then two days in a field with no Wi-Fi at Ideas Fest, talking AI from every angle going.
The headlines said be afraid. I went with the tint.
Over to the news.
|
|
THIS WEEK
Switch ON to AI News
01 · THE LEAD
Stop Right Now, Thank You Very Much
AI researcher Jacob Coxon has quit, warning that both his former employers, Anthropic and OpenAI, are racing irresponsibly towards superintelligence. Anthropic's own alignment lead, Evan Hubinger, puts his odds of AI wiping us out at over 10% within a decade. Anthropic boss Dario Amodei wants everyone to slow down. Nobody has yet.
My take: the only thing that will save us here is regulation. Independent regulation, and control over what gets made. I'm glad California has just signed laws to set up independent AI auditors. It's a start.
Read more in my Substack
02 · THE BREAK-IN
Gone in 26 Seconds
Security firm GreyNoise spotted one attacker sending hundreds of AI agents through holes in PaperCut's print software. They broke into 395 organisations. Only the US was hit harder than the UK. Eleven fell in 26 seconds.
My take: even the agents have worked out that teamwork makes the dream work. So as we hand our own agents the keys, let's watch what they can get into and how they connect with each other.
Read it
03 · THE LABEL
Say My Name
Anthropic now hides an invisible watermark in text from its newest Claude models, worldwide. It survives copy-paste and light edits, but a heavy rewrite can remove it. There's no public detector yet.
My take: the label doesn't bother me. What reads it might. Will AI search engines start reading the mark and leave marked work out of their answers? Too early to tell. Machines get this wrong already: one decided I didn't write my own essay.
Read it
04 · THE ROOM
Both Sides Now
A Talent500 survey of women in India's tech sector found 52% have seen gender bias or stereotypes in AI's answers. Yet 64% say AI is speeding up their path to senior roles.
My take: AI skills get you better jobs. PwC found roles that need them pay a 62% premium. We need far more women building those skills, and fixing the bias from the inside.
Read it
05 · THE TABLE
Whisper Words of Wisdom
Common Sense Media surveyed 1,204 US kids aged 9 to 17. Of those using AI, 37% have told it about their feelings. A quarter of those say it sometimes understands them better than most people. Over 4 in 10 say they've never had the AI safety talk from a parent.
My take: as a mum of twins, ouch. AI got to the feelings conversation before a lot of us did. Skip the big sit-down. At dinner, try: “What's the weirdest thing you've asked AI? I'll go first.” Then go first. Nobody confides in a lecture.
Read it
|
|
SPOTLIGHT
Spark Switched ON
Mostly Harmless: The AI Doom Warnings, Translated
Remember the 10% in story one? He hasn't quit. He still works there. And last year his boss, Anthropic's co-founder and CEO Dario Amodei, put the chance of things going “really, really badly” at 25%. No airline would get away with that.
I grew up under the mushroom cloud, so I've watched an arms race turn into treaties before. My new Substack translates the doom: are we really facing extinction, why do the loudest warnings come from people on their way out, and what's in our hands?
Read it on Substack
|
|
TRY IT THIS WEEK
Switch this ON
I'm loving Higgsfield, an AI video and image studio. I've been playing with photographs, making all sorts of things, and using them for some really cool new carousel ideas. That's me above: I asked Higgsfield to turn one flat photo of me into a 360.
What makes it fabulous is its brand-new connection with OpenAI's GPT-6 Astra: Astra does the thinking, and Higgsfield makes the pictures. That's where the magic comes from. I'll share more soon, but for now, give it a go.
Only upload people who've said yes. The free plan watermarks everything. Start with no faces at all:
THE PROMPT · COPY IT AS IT IS
“Turn this photo of [my product / my shop window / my desk] into a five-second video. Slow push-in, soft morning light, nothing else moving.”
Then tell it what you'd change and go again. The back-and-forth is the skill.
Try Higgsfield
|
|
THE REAL TAKEAWAY
Fear makes great headlines and terrible decisions.
A week like this makes anyone want to shut the laptop. Shame, because the people shutting it have the most to gain.
Being fearless with AI means knowing what it can touch. Pick one AI tool you work with and check three things: what it can open, what it keeps, and what it can do without asking you. Want the whole system? That's the Session, below.
Forward this to someone the headlines have scared off. Especially a parent: story five is your dinner-table opener.
|
|
SEPTEMBER
Come and find me
Don't miss this one
The Switched ON Session: Live
Wednesday 23 September · King's Cross, London · workshop with food and drinks
Let's make it really simple. If you want to set up AI, get the best out of it and build agents, you really don't want to miss this one. Most of us have AI in bits. This is one evening to turn it into your own system: one that runs, stays safe and doesn't fall apart when the tools change. Booking closes Monday 21 September.
Look back to move forwards · HAT at 50 · Wednesday 16 September · The IPA, London
This week I'm wearing one of my other hats, and this one's called HAT. I'm President of the History of Advertising Trust, which turns 50 on Wednesday. Not much AI in this one, but it matters: you can't shape where advertising creativity goes next without knowing where it's been. About HAT
|
|
HOW WAS THIS WEEK'S SPARK?
with love and sparkle,
Nish x
P.S. Extinction, I can think about calmly. Jelly shoes coming back next summer, I cannot.
WRITTEN BY MOUTH
My words in this newsletter started out loud, and Wispr Flow did the typing. My keyboard has never had it so easy.
Get 3 months free: wisprflow.ai/nishma. Full disclosure: I'm an investor and ambassador, and that's my link.
LinkedIn ·
Instagram ·
Substack ·
YouTube
|
|
|
