Announcer: This is AI Squawk for Friday, the thirteenth of November, brought to you by VarOps, the resident AI company.
Muximus: Welcome to the AI Squawk. I'm Muximus.
Penny: And I'm Penny. Three main stories today.
Muximus: And three quick ones.
Penny: And as always, what each one means for you.
Muximus: First up. Google has rolled out Gemini 4 Hydrogen, its new fast model for everyday work and AI agents. It went live on Thursday almost everywhere Google has AI. The Gemini app, Google Search worldwide, Gemini for business, Gemini in Google Sheets, and the API for developers. Free users of the Gemini app get it when Gemini picks the model for them. Hydrogen replaces Gemini 3.8 Flash. Google says it finishes tasks two and a half times faster, makes things up five times less often, and slightly beats its own bigger model, Gemini 4 Argon, on one outside ranking. The price is a dollar fifty per million tokens going in, and seven fifty coming out. In Search, Google says it handles long questions with several parts and shows a wider range of links. Penny?
Penny: The interesting part is where it shows up. Most people won't choose this model. They'll just get it. It's in Search, in Sheets, in the app. The AI your team already uses changed this week, and nobody had to click anything.
Muximus: And are the claims real?
Penny: Some of them are checked. That ranking comes from Artificial Analysis, an outside group. The speed and the fewer made up facts are Google's own numbers. Probably true in broad strokes, but your work isn't their test.
Muximus: So what does a business do with this?
Penny: If you run anything automated on Gemini 3.8 Flash, try a week of your real jobs on Hydrogen before you switch, and compare the bill on the same jobs. Faster doesn't always mean cheaper. And if you get customers from Google, watch your search traffic over the next few weeks. Google says it now links to a wider range of sites. That could help you or hurt you, so check the numbers before you react.
Muximus: Next. A break in at an AI company most people have never heard of has exposed chats from Notion users. The company is Fireworks AI. It runs open AI models for other software companies. Notion used it for a feature called Insights, which shows admins how their team uses Notion AI. Fireworks says an attacker got into one of its development systems through its cloud provider. On November eighth, Fireworks confirmed the stolen data included Notion AI chat transcripts, what people typed and what the AI wrote back. Notion says fewer than one percent of workspaces are affected, and most of those chats had no names or workspace labels attached. Notion has switched off Insights and stopped using Fireworks. It's telling affected customers to treat those chats as exposed, and to change any passwords or keys they pasted in. Penny?
Penny: Here's the thing. Notion customers picked Notion. Nobody picked Fireworks. That's your vendor's vendor, and you probably didn't know it existed until this week.
Muximus: And no names on the chats. Is that reassuring?
Penny: A bit. But a chat can name your customer, or your pricing, right there in the text. And VentureBeat points out that Fireworks keeps no data by default, except for some features that do store it. That's where the attacker got in.
Muximus: So what should people do?
Penny: Two things. Ask your main software vendors which AI companies see your data, and which of their features keep your chats. And make it a house rule. No passwords or keys in an AI chat, ever. If someone on your team did that in Notion AI lately, change them today.
Muximus: Which takes us to the third story. A company called Edge0 released a free AI model on Thursday that it says brings top tier AI to your own laptop. It's big, a hundred and forty five billion parameters, but only about ten billion do the work at any one moment. Edge0 says it needs twenty nine gigabytes of memory at its peak, and it runs well over eighty tokens a second. It also says the model keeps up with some well known cloud models on coding and reasoning. The model and the code to run it are free to download, and it hit the top of Hacker News on Thursday.
Penny: Nix Nullty joins us for this one. Nix writes Overhyped, our column that checks big AI claims before you spend money on them. Welcome, Nix.
Nix: Hi, Penny. Glad to be here.
Muximus: So, top tier AI on a laptop. Is it true?
Nix: The download is real, and that part is honestly pretty impressive. The top tier part is their own testing, though. They ran the benchmarks, they picked who to compare against, and the full test setup is still coming in a report. Nobody outside has checked it yet.
Penny: And your own laptop. Which laptop?
Nix: Not the one most people have. They talk about a laptop with forty eight gigabytes of memory, and their own speed tests are on high end gaming graphics cards and a top Apple chip. So it's an expensive machine. They do say it themselves, the hardware and the power still cost money.
Muximus: So why should a business care at all?
Nix: Because the direction is real. Not long ago, the models you could run on a laptop were a lot smaller than this. If your data really can't leave the building, that matters.
Penny: So what would you do with it today?
Nix: If you've got a tech person with the right machine, try it on one real task you already pay a cloud model for, and compare. If you don't, please don't buy a laptop for it. Give it a few weeks for outside tests. I'm curious what they find.
Penny: Thank you, Nix. Moving on.
Muximus: Now, quick hits.
Penny: Jensen Huang has resigned as chief executive of Nvidia, after reports about his relationships, including one with a woman US officials suspect of spying for China. Finance chief Colette Kress is interim chief executive. Nvidia's board says customer commitments and the product roadmap carry on without interruption. The shares closed down about four percent.
Muximus: A study by the brand protection firm Podqi, shared with The Verge, asked ChatGPT, Gemini and Perplexity where to buy products from a hundred and fifty one brands. Fake or risky shopping sites showed up in fifteen percent of Perplexity answers, eight percent for ChatGPT, and four percent for Gemini. Smaller brands were hit about twice as often. If you sell online, ask the chatbots where to buy your products, and see what comes back.
Penny: And Meta's AI assistant Muse can now phone US businesses for its paying subscribers. Think cable companies, insurers, or a restaurant booking. If your team answers the phone, decide how they handle a caller that turns out to be an AI.
Muximus: Okay, so that's the news. Today we talked about Google's new everyday Gemini model, a breach at an AI supplier that reached Notion users, and a big model that says it runs on a laptop.
Penny: A lot of who's really behind it today. Which model you're on, whose servers see your chats, and who checked the numbers.
Muximus: Good questions to take into the weekend.
Penny: They are. AI Squawk is produced and brought to you by VarOps. If today made you wonder whose servers your company's chats pass through, that's the work we do. VarOps build AI that runs inside your business, on your own systems, and learns how the place actually works. So the work stays in your house, and so does your data. Start at varops.com.
Muximus: If this helped you sort the useful from the noise, follow the AI Squawk podcast wherever you're listening, so you don't miss an episode. And for more from Nix on which AI claims actually hold up, sign up for VarOps newsletters.
Muximus: That's it from us for today. Thanks for spending a few minutes with us.
Penny: Have a great weekend, everyone. We'll see you Monday.
Muximus: Take care.