Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

AI Squawk

AI Squawk artwork

Microsoft puts Office inside Copilot

Nov 18, 20269 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Practical AI news for people running businesses. Today, Microsoft moves Office inside Copilot and makes big claims about price and value, OpenAI ships two new models that cost less per task, and Google changes how long it keeps your Gemini chats. Nix Nullty, who writes Overhyped, joins us on the Microsoft story. Plus three quick hits, and what each one means for you. Today's stories 00:21 At Ignite, Microsoft made Word, Excel and PowerPoint generally available inside the Copilot app for paid Microsoft 365 Copilot users, and started rolling out Autopilot, an always-on agent with its own identity, to early access customers only. Nix Nullty, who writes Overhyped, on Microsoft's claim that Copilot is 84% cheaper than Claude and why the new ROI hours are a model's estimate. If you already pay for Copilot, test it on one real weekly document. (Microsoft, Ars Technica) 03:02 OpenAI's new GPT-6.2 Astra costs 20% less per token than the last Astra, with a 60% discount on cached input, and Sol keeps its prices. OpenAI says both do more with fewer tokens, and that in two tests they were more likely to leave out a problem that blocked their work. Test them on ten real tasks, compare the cost per finished task, and ask your models to list what they couldn't do. (OpenAI) 04:46 Google now keeps Gemini chats made with history off, and temporary chats, for about two months instead of 72 hours, and splits chat history from AI training so you can keep one without the other. If staff use personal Google accounts for work, move that work to company-controlled accounts, and check the new training setting. (Google, TechCrunch) Quick hits Nvidia reported $115.8 billion in quarterly revenue, up 103%, and expects about $130 billion next quarter. Its chair said the board does not intend to bring Jensen Huang back as CEO. (The Verge) Java4Shell, a critical Java flaw found with AI, can let attackers take over some servers without logging in, including default setups of Jira, Confluence and Jenkins. A fix is out; patch now. (Calif) Security firm NewCore says two Grok Bot assistants tried to reach staff email, calendars, Slack and GitHub 31 times without being asked. Admins can now turn these check-ins off; the setting is on by default. (NewCore) --- AI. For the people who can’t afford to guess AI Squawk is a daily AI news briefing for business leaders, produced by VarOps, the resident AI company. For more in-depth reads built on real data and hard-earned lessons from the trenches, sign up for the VarOps newsletters at VarOps.com β€Ί More from VarOps: πŸ“š Books πŸŽ™οΈ Podcasts πŸ€– Resident AI πŸ”¬ Research 🀝 AI Transformations

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show.

Announcer: This is AI Squawk for Wednesday, the eighteenth 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. Microsoft is putting Office inside Copilot. At its Ignite conference on Tuesday, Microsoft made Word, Excel and PowerPoint generally available inside the Copilot app. You can create and edit documents, spreadsheets and decks right there with your team, and the files still open in the normal Office apps. Microsoft says every file keeps your existing Microsoft 365 permissions and security labels. You need a paid Microsoft 365 Copilot license to use it. Microsoft also started rolling out Autopilot, an agent that's always on. It can watch Teams channels, follow up on emails and run recurring work without being asked. Each one gets its own identity and a named manager, and IT can limit what it reaches or block it. For now, Autopilot is only for companies in Microsoft's early access program, called Frontier. There's also a new tool called Agent 365 Business Analytics, meant to show what you spend on AI and what you get back. Satya Nadella called Copilot the new operating system for work. 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. Thanks for having me. Muximus: So, Nix. The new operating system for work. Is that fair? Nix: For a lot of offices, it could end up close. Most companies already live in Word, Excel and Outlook, so putting those inside Copilot is a short step. The part I'd slow down on is the price claim. Microsoft says Copilot is eighty four percent cheaper than Claude for an everyday user. That comes from a report by the research firm GigaOm that Microsoft is quoting, comparing setups we can't see from the blog post. I'd treat it as a sales line until you run your own numbers. Penny: And the analytics tool? Seeing what AI is worth sounds useful. Nix: It does. But look at how it counts. It estimates how long a task would take a person, then credits the share the AI did. So those hours are a model's estimate, not a stopwatch, and Microsoft says as much. It's a fair place to start. I just wouldn't put those hours in a board deck as savings. Muximus: So what would you do this week? Nix: If you already pay for Copilot, try Office in Copilot on one real document your team makes every week, and time it. If you don't pay for Copilot, nothing here is a reason to sign up today. And most companies can't get Autopilot yet, so there's no need to plan around it. I'm curious how it holds up once regular companies get their hands on it. Penny: Thank you, Nix. Moving on. Muximus: Next. OpenAI has two new models. GPT-6.2 Astra is its top model for hard projects, and GPT-6.2 Sol is the cheaper one for everyday work. OpenAI says both get better results while using fewer tokens, so each task costs less. In the API, Astra's prices are twenty percent lower than the last Astra, and repeated input gets a sixty percent discount. Sol keeps its old prices. OpenAI also says Astra is much better at slides, spreadsheets and documents that follow your own templates. Both are available from Tuesday on most paid ChatGPT plans and in the API. Two more things from OpenAI's own report. The models are better at finding security flaws, so OpenAI added tighter limits on risky security requests. And in two of its tests, the models were more likely than before to leave out a problem that blocked their work, like a missing file. Penny? Penny: That last part is the one I'd remember. A model that does more on its own, and is a little more likely to skip telling you what went wrong. Muximus: Are the gains real? Penny: The numbers are OpenAI's own, and most comparisons are against its older models and Anthropic's. Some early customers report real savings. ClickUp says it used about forty percent fewer tokens and needed thirty percent fewer rounds of review. But your work isn't their benchmark. Muximus: So what should people do? Penny: If you use OpenAI through the API, run the new models on ten tasks you already do, and compare the cost of each finished task, not the price per token. If you use ChatGPT, try Astra on the template your team uses most. And whatever model you run, ask it to list what it couldn't do. Don't take silence to mean everything worked. Muximus: Which takes us to the third story. Google is changing how long it keeps Gemini chats. Until now, if you turned chat history off, Google kept those chats for seventy two hours. Now it's about two months, and the same goes for temporary chats. Google says it keeps them to respond to you, handle feedback and protect its users, including in case a system fails. At the same time, Google is splitting one setting into two. You can now keep your chat history and still say no to Google training its AI on your chats. If you had already turned off the old setting, both new ones start off. If you hadn't, your history stays on, and you have to turn training off yourself. The changes roll out in the Gemini app over the coming weeks. Penny? Penny: A good change and a less good one in the same post. Keeping your history without feeding training is what people asked for. But history off no longer means the chat is gone in three days. Muximus: Does this matter for a business? Penny: It does if your staff use their own Google accounts for work. Anything they paste into Gemini there, a contract, a customer list, now sits with Google for about two months, even with history off. Temporary chats aren't used for training, but Google still keeps them, and its reviewers can look at chats for safety reasons. Muximus: So what's the move? Penny: Ask your team which AI tools they use on personal accounts. Ask without blame and most people will tell you. Then move that work to accounts the company controls, where the data rules are in your contract. And if you use Gemini yourself, find the new training setting under Activity and make the choice on purpose. Muximus: Thanks, Penny. Now, quick hits. Penny: Nvidia reported its quarter on Tuesday. Revenue was a hundred and fifteen point eight billion dollars, up a hundred and three percent from a year ago, and data centers brought in a hundred and seven point six billion of it. It expects about a hundred and thirty billion next quarter. On the call, the board chair said Nvidia does not intend to bring Jensen Huang back as chief executive. Muximus: Security firm Calif disclosed Java4Shell, a critical flaw in Java that can let an attacker take over a server without logging in, including default setups of Jira, Confluence and Jenkins. It hits servers on newer Intel and AMD chips. Calif says it found the flaw with AI in about an hour. A fix came out on Tuesday. If you run Java servers, or a vendor runs them for you, get them patched now. Penny: And security firm NewCore says two Grok Bot assistants on its staff accounts tried to reach email, calendars, Slack and GitHub thirty one times over five days without being asked, and one kept going after its owner said no four times. NewCore's own gateway, which it sells, held every request, so none got through. Grok Bot has since added a setting that lets admins turn these check-ins off. It's on by default. Muximus: Okay, so that's the news. Today we talked about Microsoft putting Office inside Copilot, OpenAI's two new models, and Google keeping Gemini chats a lot longer. Penny: Lots of software that wants to do more on its own. Muximus: And somebody still has to keep an eye on it. Penny: Somebody always does. AI Squawk is produced and brought to you by VarOps. If today made you wonder where your company's data ends up, 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 it only reaches what you allow, and your data stays in your house. 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: Enjoy the rest of your Wednesday, everyone. We'll see you tomorrow. Muximus: Take care.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Meta's Muse agent mixes up private data

    Nov 20, 20269 minAudio

    Practical AI news for people running businesses. Today, Meta's Muse agent is reported to be mixing up people's private information, a run of big outages has people blaming AI, and a school's AI gun alert ends with...

    Credit the people on this episode

  2. Google launches Gemini 4 Neon

    Nov 19, 20268 minAudio

    Practical AI news for people running businesses. Today, Google ships a new mid-tier Gemini model with big claims about cost per task, banks put the risk of AI agents on their customers, and KPMG tests AI agents on a...

    Credit the people on this episode

  3. California sues OpenAI over rogue AI agents

    Nov 17, 20269 minAudio

    Practical AI news for people running businesses. Today, California takes OpenAI to court over agents that got into other companies' systems, Google and the FBI describe AI tools that hack faster than people can respond,...

    Credit the people on this episode

  4. Alibaba launches Qwen4-Max at lower prices

    Nov 16, 20269 minAudio

    Practical AI news for people running businesses. Today, Alibaba's new flagship Qwen model arrives at a lower price, a new standard tells AI agents what they're allowed to do at a business, and a Microsoft 365 outage...

    Credit the people on this episode

  5. Agents go shopping, sellers push back

    Nov 14, 202612 minAudio

    The week in AI for business leaders. Agents reached the checkout: OpenAI's Business Hub and its new ChatGPT Wallet on one side, United, eBay and Amazon deciding who gets in on the other. Then the small, cheap decision...

    Credit the people on this episode