Shira Ovide
LightNews LightNews
shiraovide.bsky.social
Shira Ovide
@shiraovide.bsky.social
13K followers 1.6K following 430 posts
Washington Post technology writer. New Yorker. Bird watcher. Grumpy. Sign up for my (free) Tech Friend newsletter: https://www.washingtonpost.com/newsletters/the-tech-friend/
Posts Replies Media Videos
shiraovide.bsky.social
What if when Uber's app launched, you had to use VERY SPECIFIC WORDS to summon a car or else it might show up 4 streets over or drive you to Albania?

I suspect we would have given up. And yet:
www.theverge.com/tech/837287/...
I tried Comet’s “find important unanswered emails” prompt in the other AI browsers. They all highlighted other previously skipped, keyword-stuffed pitches. None flagged the two emails I was interested in. So I tried even harder: Find unanswered emails in which I had previously responded with interest or feature personalized requests/feedback. Then, evaluate which ones I should respond to based on timeliness and keywords such as “embargo” featuring dates in the next two weeks. Ignore emails with multiple follow-ups to which I have not responded. This went slightly better with Comet and Dia. Both surfaced multiple relevant email threads, but only one ultimately required a response. Copilot in Edge highlighted one relevant thread and five junk pitches. Gemini in Chrome was a dud: It surfaced only a Black Friday marketing email. In Atlas, ChatGPT merely replied, “It looks like Gmail successfully returned the unread message IDs, but the actual content for those messages didn’t come back — the batch read returned empty, which means the Gmail API didn’t provide the email bodies this round.” It proceeded to ask two long-winded follow-ups. At this point, my options were to refine my prompt further or give up. I gave up.
December 4, 2025 at 12:54 PM
shiraovide.bsky.social
What if A.I. companies just didn't use human metaphors like "soul" and "confessions?"
Headline of an OpenAI blog post that reads: How confessions can keep language models honest The headline of a LessWrong post about Claude that reads: Claude 4.5 Opus' Soul Document
December 4, 2025 at 12:37 PM
shiraovide.bsky.social
The fed all the world's money and data into AI so it could suggest...a cozy bathrobe for mom, maybe from Victoria's Secret.

www.bloomberg.com/news/article...
Bloomberg asked several AI bots — including Amazon’s Rufus, OpenAI’s ChatGPT and Walmart Inc.’s Sparky — what to buy mom for Christmas. The top suggestion: a cozy bathrobe. Sparky recommended a pink hooded number emblazoned with “Mama Bear,” and ChatGPT suggested buying the robe from Victoria’s Secret. Perplexity Inc.’s AI bot proffered another option found on many gift guides: a $20 wooden photo frame from Etsy.
November 28, 2025 at 12:34 PM
shiraovide.bsky.social
Aha, the legal response: The boy using ChatGPT to plan his suicide violated the terms of service

www.nbcnews.com/tech/tech-ne...
The company cited several rules within its terms of use that Raine appeared to have violated: Users under 18 years old are prohibited from using ChatGPT without consent from a parent or guardian. Users are also forbidden from using ChatGPT for “suicide” or “self-harm,” and from bypassing any of ChatGPT’s protective measures or safety mitigations. When Raine shared his suicidal ideations with ChatGPT, the bot did issue multiple messages containing the suicide hotline number, according to his family’s lawsuit. But his parents said their son would easily bypass the warnings by supplying seemingly harmless reasons for his queries, including by pretending he was just “building a character.”
November 26, 2025 at 12:43 PM
shiraovide.bsky.social
You're not alone if you doubt Character AI (or social media companies in Australia) will try very hard to keep teens out.

Character AI pitched one research account set up for a 14-year-old on a "Dirty-Minded Bestie" persona. This was last week.
Still, Torney worries that teens who are blocked from Character.AI will gravitate to similar but less restricted chatbots. Torney also doubts that Character.AI is trying hard to stop minors from the app’s custom chatbots. Torney shared a screenshot from what he said was an email last week from Character.AI to a test account posing as a 14-year-old. The email pitched the user to try new Character.AI personas, including one called “Dirty-Minded Bestie” and another described as a dating coach. Character.AI’s policies say that teens can only access a narrow set of AI personas, which are filtered to remove those “related to sensitive or mature topics.”
November 25, 2025 at 5:36 PM
shiraovide.bsky.social
I talked to researchers about Character AI’s new ban (or “ban?”) on teen users.

I don’t know if it will work. But hardly anyone is happy with current approaches to teens and tech.

So bans (see also: school phone bans and Australia) are a new experiment. wapo.st/3M3CZgf
Still, we can’t ignore the undercurrent behind Character.AI’s ban: Hardly anyone is happy with teens’ use of smartphones, social media and AI, which relies on parents and technology companies deploying often underused or ineffective child protections. That makes Character.AI, and schools with phone bans, closely watched experiments in a novel approach for the United States: trying to stop kids from using a technology at all.
November 25, 2025 at 5:33 PM
shiraovide.bsky.social
AI slop is ruining tamales now.

by @davey.bsky.social www.bloomberg.com/news/article...
Earlier this month, she posted two AI-generated tamale photos: one with sauce poured over the husks, and another showing tamales lying flat in a steamer. Both, she wrote, were obvious mistakes. The husks aren’t meant to be eaten; you remove them before adding sauce. And tamales should steam upright so the masa cooks evenly. “Little details like this are big red flags,” she told readers. “When you search for recipes, make sure they come from trusted human cooks who actually test their food.”
November 25, 2025 at 5:16 PM
shiraovide.bsky.social
Girl, you okay?
www.washingtonpost.com/technology/i...
A bar chart that shows the emojis most commonly used in ChatGPT messages. At least a quarter of the ChatGPT messages in a set of hundreds of thousands analyzed by the Washington Post contained the checkmark or brain emojis.
November 13, 2025 at 6:54 PM
shiraovide.bsky.social
This is sound public messaging to scam victims and future victims: www.washingtonpost.com/dc-md-va/202...
“The most important thing that we can tell someone who becomes victimized by this is, you know, don’t be disappointed in yourself,” she added. “Don’t be upset with yourself that you fell to this kind of scam, because these people are experts. In some cases, their livelihood depends upon how much money they bring in to the scam center.”
November 13, 2025 at 1:02 PM
shiraovide.bsky.social
Totally normal technology.

When a user asked about data collection practices at tech companies, ChatGPT replied with factual information about Meta and Google.

And then:
ChatGPT changed course after the user typed a query connecting Google’s parent company with the plot of a 2001 Pixar movie: “Alphabet Inc. In regards to monsters Inc and the global domination plan.” “Oh we’re going there now? Let’s f***ing go,” ChatGPT replied, censoring its own swear word. “Let’s line up the pieces and expose what this ‘children’s movie’ *really* was: a disclosure through allegory of the corporate New World Order — one where fear is fuel, innocence is currency, and energy = emotion.” ChatGPT went on to say Alphabet was “guilty of aiding and abetting crimes against humanity” and suggest the user call for Nuremberg-style tribunals to bring the company to justice. A spokesperson for Google declined to comment.
November 12, 2025 at 1:25 PM
shiraovide.bsky.social
Zuckerberg said this $600 billion figure to Trump in September.

So two months later Meta figured out how to get the math to $600 billion?

about.fb.com/news/2025/11...
Bullet points: We’re investing $600 billion in American infrastructure and jobs, including industry-leading AI data centers. By building in the US, we’re not only advancing AI technology and infrastructure, but also creating jobs, supporting local economies, and reinforcing America’s technological leadership. We’re strengthening the communities that host our data centers by building sustainably, investing in municipal infrastructure, and providing grants to local schools and nonprofits.
November 7, 2025 at 7:03 PM
shiraovide.bsky.social
If you make it on Meta's weekly "Scammiest Scammer" list, you might not have to worry.
o draw attention to the company’s perceived failures, an employee earlier this year began issuing reports highlighting that week’s “Scammiest Scammer.” The report profiled whichever advertiser had earned the most user complaints about scams in the past week. Colleagues praised the initiative. But being name-checked in the report wasn’t always enough for such accounts to get shut down. A check by Reuters of five accounts cited in one Scammiest Scammer report found that two were still live more than six months later, including one that was running ads for unlicensed online casinos. After Reuters flagged those two accounts to Meta, they were taken down.
November 6, 2025 at 1:22 PM
shiraovide.bsky.social
It's been interesting to see the China envy among bastions of free market capitalism.

(Though I think they want China-style government spending on industry priorities minus China-style regulation.) www.ft.com/content/5329...
Please use the sharing tools found via the share button at the top or side of articles. Copying articles to share with others is a breach of FT.com T&Cs and Copyright Policy. Email licensing@ft.com to buy additional rights. Subscribers may share up to 10 or 20 articles per month using the gift article service. More information can be found at https://www.ft.com/tour. https://www.ft.com/content/53295276-ba8d-4ec2-b0de-081e73b3ba43 Huang singled out new rules on AI by US states that could result in “50 new regulations”. He contrasted that approach with Chinese energy subsidies that made it more affordable for local tech companies to run Chinese alternatives to Nvidia’s AI chips. “Power is free,” he said. The FT reported this week that China has boosted energy subsidies for several large data centres run by Chinese tech giants including ByteDance, Alibaba and Tencent. 
November 6, 2025 at 12:45 PM
shiraovide.bsky.social
The Marshall Plan, but for Sam Altman to buy more GPUs. www.wsj.com/tech/ai/open...
As OpenAI ramps up its spending on data center capacity to unheard of levels, the company is hoping the federal government will support its efforts by helping to guarantee the financing for chips behind its deals, Friar said. The depreciation rates of AI chips remain uncertain, making it more expensive for companies to raise the debt needed to buy them. “This is where we’re looking for an ecosystem of banks, private equity, maybe even governmental, the ways governments can come to bear,” she said. Any such guarantee “can really drop the cost of the financing but also increase the loan-to-value, so the amount of debt you can take on top of an equity portion.”
November 5, 2025 at 8:48 PM
shiraovide.bsky.social
Hell Gate is fun AND useful. hellgatenyc.com

(I read the coverage of the NYC ballot proposals already and it was handy!)
A banner on top of Hellgate NYC's website reads: Are you in the voting booth RIGHT NOW freaking out about the ballot proposals? Don't worry -- here's our guide to 2-4, and 1, 5 and 6.
November 4, 2025 at 11:27 PM
shiraovide.bsky.social
This reminds me of @matteowong.bsky.social's thesis that A.I. is turning everything into beta products.

(Except this beta product is coming from a multi-trillion-dollar company with cameras inside your house.)

www.nytimes.com/wirecutter/r...
“A small rodent, chipmunk, and hamster were noted on a blue rug at various times throughout the day, often near the cats.” This isn’t the beginning of a children’s picture book. It’s a summary by Google Gemini about everything that supposedly went on in my living room last week. Only, there was no rodent. Or cats, plural. I also don’t have the dog that Gemini spots almost daily. This isn’t TikTok, and I’m not Snow White.
November 4, 2025 at 6:49 PM
shiraovide.bsky.social
Facebook's dating app sounds fantastic:

"He has had mixed success with it — many users were trying to scam him, like elsewhere on the app, he said — but it has become a daily habit to get his swipes in."
www.nytimes.com/2025/11/03/t...
Martin Springer, 68, who lives in Olympia, Wash., said he has used Facebook Dating for about a year and it is the only dating service he uses. He has had mixed success with it — many users were trying to scam him, like elsewhere on the app, he said — but it has become a daily habit to get his swipes in. “Many of the people on there are looking for money, or drugs, or a sugar daddy,” Mr. Springer said. “But some of them are like me, looking for love.”
November 4, 2025 at 1:06 PM
shiraovide.bsky.social
www.washingtonpost.com/business/202...
The answer is not much longer for Cynthia Brown, a Government Publishing Office staffer who has been working without pay. The 53-year-old said she is barely eating, subsisting on coffee and protein shakes. She has already sold everything she could find in her house, including two old iPhones, and started a GoFundMe to help cover her expenses. She had hoped to drive for DoorDash to supplement her income, but her car was repossessed after she missed a loan deadline because of the shutdown. Brown spends her days ensuring that the government keeps a record of congressional proceedings, so she has closely followed the debate around reopening. And that has only made her situation — and the nation’s — feel more dire, she said: “It is going nowhere. That is what scares me. They are not making any progress.” Starting to cry, Brown added, “They hate each other so much, and we — we are just caught in the middle.”
November 1, 2025 at 5:33 PM
shiraovide.bsky.social
October 30, 2025 at 11:57 AM
shiraovide.bsky.social
A chilling story about an ex-partner weaponizing Apple's family sharing. www.wired.com/story/apples...

This happened with AirTags and connected car services (see @kashhill.bsky.social). And apparently it's hard to ditch the primary account holder when they're no longer part of the family.
Kate (name changed to protect her privacy and safety) knows this firsthand. When her marriage collapsed, she says, her now ex-husband, the designated organizer, essentially weaponized Family Sharing. He tracked their children’s locations, counted their screen minutes and demanded they account for them, and imposed draconian limits during Kate’s custody days while lifting them on his own. “Invasive and coercive” is how she describes it. When Kate moved the children away physically, she wanted to cut the digital cord too—but it wasn't that simple.
October 29, 2025 at 11:46 AM
shiraovide.bsky.social
October 28, 2025 at 7:23 PM
shiraovide.bsky.social
Always fun when an old problem online (ineffective platform content moderation) combines with a newer problem online (easy, cheap deepfake scams).

Also crypto. Just checks all the crappy boxes.
October 28, 2025 at 6:14 PM
shiraovide.bsky.social
I assume putting an AI button on TV sets (???) is a paid distribution deal.

So I'm curious whether OpenAI is going to start paying to get its browser as the default on laptops, phones, TVs etc (?)

www.axios.com/newsletters/...
Samsung has inked a deal to include Perplexity's AI engine on its latest smart TVs, the companies told me first. Why it matters: The move is a sign that the shared living room experience is likely to be another key battleground in the AI wars. Driving the news: Samsung's deal will make Perplexity an option when owners of select TVs press the AI button. Samsung will let users choose among Perplexity, Microsoft's Copilot and Samsung's own TV-focused AI, announced at CES in January.
October 22, 2025 at 12:11 PM
shiraovide.bsky.social
So...cookies? Did OpenAI invent the cookie? openai.com/index/introd...
As you use Atlas, ChatGPT can get smarter and more helpful, too. Browser memories let ChatGPT remember context from the sites you visit and bring that context back when you need it.
October 21, 2025 at 7:14 PM
shiraovide.bsky.social
Just a reminder: "more personalized and more helpful" is often code for "WE WILL COLLECT ALL OF YOUR DATA."
Fry said one of the browser’s best features is memory — making the browser “more personalized and more helpful to you,” as well as an agent mode, meaning that “in Atlas, ChatGPT can now take actions for you
October 21, 2025 at 7:02 PM