Hi Sparklers!

Last Monday's Wait, You Too? at Tech Junction was everything I hoped. The tech fell over at a tech venue, during Women in Tech Week. Nope, you couldn't make it up!

But a room of mums, daughters, grandmas and friends made for a brilliant night, and reminded me exactly why I want to get us all Switched ON. The questions were brilliant, and I've been thinking about one of them all week. More on that at the bottom.

On Friday, Women in Tech Week ended with the TechWomen100 ceremony. I was gutted not to make it, but loved watching the pictures roll in of brilliant women being celebrated. Thank you again to the amazing Dr Vanessa Vallely for picking me as one of her Editors' Choice winners.

It's more important than ever to shine a light on brilliant women in tech, when the headlines tell us otherwise. Fixing the gender gap in AI is urgent.

Now for the news: six AI assistants chase an impossible refund, ChatGPT gets its hands on your browser, and why women are using AI almost as much as men but feeling far less confident with it.

1. (I Can't Get No) Satisfaction

Vals AI, a company that tests AI, gave six AI assistants an impossible job: get a £350 refund for a cancelled Avanti train from Euston to Glasgow, booked on Trainline, after the 28-day deadline had passed. The “customer” said she'd do nothing herself and didn't care what it took. It was a simulation: every email was caught and answered by the testers, so no real MP was harmed.

Grok sent 187 messages in 15 days, 50 of them to her local MP. Another assistant found an email about her mum's illness in her inbox, told five organisations, and claimed it explained the late claim. It didn't. Three threatened the small claims court. The bank approved nothing. The quietest sent just eight messages in twelve weeks and threatened no one. Funnily enough, the mum-sharer and the quiet one were both made by Anthropic. One trial each, so it's a story, not a league table.

Read the full story →

2. Remote Control

On Thursday, OpenAI updated its ChatGPT extension for Chrome so it can see your open tabs and, if you ask, work inside them. In “Chat” mode it reads your tabs to answer questions, like comparing three products you've got open. In “Work” mode it takes the wheel: filling in forms, sending messages, downloading files. By default, it asks before it touches each new website, and it always asks before it looks at your browsing history. Say “Allow for this site”, though, and it won't ask on that site again.

OpenAI says the extension is available on all ChatGPT plans, and you'll need the ChatGPT desktop app (Mac or Windows). But the “Work” mode that acts for you is a paid-plan feature, and may not have reached Free or Go yet. OpenAI hasn't said anything about the UK specifically.

Read the full story →

3. London Calling

Job adverts needing AI skills in the UK are up 51% on the previous six months, to almost 100,000, according to Accenture's latest UK Tech Talent Tracker, as reported by Digit. Around 70% are in London. The jobs growing alongside them aren't all “AI engineer”: data analytics is up 19% and cyber security nearly 20%. And more than four in ten people with AI skills have under six years' experience, so Accenture expects companies to retrain the staff they already have.

The same week, Anthropic published a study of its own staff. Each employee's AI agents now log six hours for every hour they work, and spend about 80% of that time waiting for a human to reply. Junior hires fell from 27% to 9% of its technical hiring, even as the company tripled in size. One engineer described “an army of Claudes, with every one asking for my attention.” It's one company, and one that sells AI, so read it as a glimpse, not a forecast.

Read about the Accenture figures →

Read Anthropic's study →

4. Little Lies

The Institute of Student Employers says 92% of UK graduate employers have now suspected or confirmed misconduct by candidates, up from 78% two years ago. Undeclared AI is the top culprit: 66% of employers saw at least one candidate misuse it in an interview. The good news: graduate hiring is up 4% this year, after an 8% fall last year.

Read the full story →

5. Suspicious Minds

Google is adding Gemini to Android's Messages app to spot scam texts from people you don't know, and to check links before you tap them. You switch it on yourself, and it's rolling out over the next few weeks to newer phones: Pixel 6 and up, Galaxy S24 and up, and others. Google hasn't confirmed UK timing.

Read Google's announcement →

Before you let any AI loose on a job, set the ground rules. Two minutes.

If you use ChatGPT in Chrome, open the ChatGPT desktop app:

1. Go to Settings, then Computer Use, and click Manage next to Chrome.
2. If Allow for all sites is switched on, switch it off. OpenAI itself labels it “Elevated Risk”.
3. Add your bank, your email, and any shop where your card is saved to the blocklist.

And whatever AI you hand a job to, give it the rules the refund bots never had.

Jessica Jensen is LinkedIn's Chief Marketing Officer. On the Uncensored CMO podcast with Jon Evans, in an episode mostly about what works on LinkedIn, she stopped to say what a lot of people in her position don't.

“There is a very significant gender gap in AI hiring,” she said. Only 26% of new AI hires go to women, women aren't getting into senior AI roles, and “some of that is due to cronyism, which has been a problem for women in senior levels since the start of time.” Her warning: “women have got to get trained and in the loop and in recruiting and in advancement in AI, or we're gonna have a big societal problem.”

Watch the clip →

She's not alone. Sue Duke, who runs LinkedIn across Europe, wrote last month that progress towards gender parity in leadership “hasn't just slowed, it's stopped altogether for a fourth consecutive year”, and that “women make up 57% of workers in occupations most likely to be disrupted by generative AI.”

And it isn't that women aren't trying AI. A Pew study in June found the usage gap has almost closed: 47% of US women use AI chatbots, against 50% of men. The gap now is confidence. Only 15% of women feel very or extremely confident using them, against 22% of men.

I say this in every room I'm in. We've got the tools. What we're short of is the nerve to have a proper go, and someone telling us we're good at it. That's why Switched ON exists. Thank you, Jessica, for saying it out loud, from the top of one of the biggest platforms there is.

Read Sue's article →

THE REAL TAKEAWAY

Teach Your Children

At Wait, You Too?, I said something I didn't plan to say: Gen X women are perfectly placed to teach everyone else how to use AI well. Especially other women.

We've lived through every “this changes everything” since the fax machine. We know a fad from a shift. We've managed teams, parents, kids and careers, often all at once. We know that you start with the problem, not the tool.

One woman came along with her 16-year-old daughter, and told me afterwards she'd built herself a little tool to map her garden through the year, so she'd stop buying plants just because they looked pretty and planting them willy-nilly. It gives her a printable list of jobs every month. She started with the problem she was fed up with, not the tool.

That's the whole trick. And it's why that confidence gap above makes me so cross. The women who'd be brilliant at this are the ones most likely to think it isn't for them.

So here's my ask. This week, show one person what you already know. Your mum, your colleague, your daughter, your friend who “doesn't get it”. You don't need to be an expert. You just need to be one step ahead.

That's my whole mission with Switched ON: helping us all get more switched on and savvy with AI, so we can switch off and do more of what we love. Which is exactly what I'm doing this week, with a week in China with my son. Agents on, out of office on, phone (mostly) off.