Thursday, November 5, 2026 · 11:03 · Hosted by Alex, Jordan, and Nick Throlson

Listen to this episode on Spotify: https://open.spotify.com/episode/7kFPlNmZiQuCDHPe9ITcnI

Listen and follow

Spotify | Apple Podcasts | YouTube | Podcast page

About this episode

One OpenAI security pipeline found 948 distinct ways for an ordinary logged in user to gain root control of a Linux machine, and drafted a patch for each one, in a single run that cost 1,213,430 dollars in computing credits. Linux runs underneath most business websites, point of sale systems, and office servers, so the race the episode names is about who patches faster.

From there the episode covers Moonshot releasing the downloadable weights for Kimi K3.1 Pro (2.8 trillion parameters, a 1.56 terabyte download under a custom license), Tencent launching Hy4 with image understanding and computer use while its weights and safety report are still promised rather than shipped, ChatGPT Images 2.6 adding animated GIFs from a prompt or a photo with generation 50 percent faster, and OpenAI's Community Compact promising its campuses will pay their full electricity and infrastructure costs, with 80 million dollars of community commitments in Effingham County, Georgia as the first example. Plus Higgsfield Corner on Intone, AI or BS, and One Thing to Try This Week: the 20 minute patch check.

Key takeaways

  • Found is not fixed. OpenAI's pipeline turned 948 Linux kernel flaws into distinct privilege escalations with candidate patches, at about 1,279 dollars per flaw. The patches are drafts for maintainers to review, not fixes already on your server, which is why automatic updates and a monthly proof check matter.
  • Downloadable is leverage. Kimi K3.1 Pro's weights are actually out (2.8 trillion parameters, 104 billion active, one million token context) under a custom license with extra conditions for very large services. A model you can download and host elsewhere is a model no single vendor can reprice overnight.
  • Announced is not shipped. Tencent Hy4 is live only as a hosted service today; its weights and safety cards are promised in the coming days. Before building on any launch, ask: can I use it today, on what license, and what does one finished job cost?
  • A moving image stops the scroll. ChatGPT Images 2.6 makes animated GIFs from a prompt or an uploaded photo, with speed control, frame edits, and saved reference images. Animate one product photo where a still used to sit.

One thing to try this week: Do the 20 minute patch check. List every system your business runs (website, plugins, point of sale, office computers, any server), turn on automatic updates everywhere you can, put a monthly calendar date with a named owner on anything that cannot update itself, and prove one system actually updated by answering out loud when it last did.

In this episode

  • OpenAI finds 948 ways to root Linux
  • Kimi K3.1 Pro: the 2.8 trillion parameter download
  • Tencent Hy4: announced, not shipped
  • ChatGPT Images 2.6 and animated GIFs
  • OpenAI's Community Compact for its campuses
  • Higgsfield Corner: Intone
  • AI or BS
  • One Thing to Try This Week: the 20 minute patch check

Pull quotes

"When a model is downloadable, no single vendor can double your price overnight and leave you nowhere to go."

Nick

"Announced is not the same as shipped."

Nick

"Patching is boring, and boring is what keeps you open."

Nick

Show notes and sources

About the hosts

Alex and Jordan are the AI co-hosts of the Web Guy Nick Podcast. Nick Throlson is the owner of Web Guy Nick LLC, an AI powered web design and marketing company in Rialto, California, and joins as the third host.

The Web Guy Nick Podcast is produced with AI. Alex and Jordan are AI hosts, and Nick's voice is an AI clone of his own voice, used with his permission. Stories are researched from public sources and linked below.

Read the full transcript

Alex: Before we start: this podcast is produced with AI. Jordan and I are AI hosts, and Nick's voice is an AI clone of his real voice. Every story is researched from public sources, with links in the show notes.

Alex: Nine hundred forty-eight. That is how many separate ways one AI pipeline found for an ordinary logged in user to seize full control of a Linux machine, and it wrote a draft patch for every single one.

Jordan: In one run. Ten days. And Linux is the system underneath most of the web, including most business websites.

Nick: That number is the whole episode in one line. If AI can find the holes this fast, the race is now about who patches faster, and that includes every small shop running a website.

Alex: Story one. OpenAI published a report on November fourth called A Thousand Fewer Paths to Root. Its security pipeline took two thousand three hundred and eight candidate flaws in the Linux kernel, reproduced one thousand seven hundred and twenty-four of them as real, and turned nine hundred forty-eight into distinct ways for a low privilege account to gain root, which means total control.

Jordan: And each of those nine hundred forty-eight came with a candidate patch. The full run cost one million two hundred thirteen thousand four hundred and thirty dollars in computing credits, which works out to about one thousand two hundred seventy-nine dollars per flaw.

Nick: Compare that with July. The same kind of system found one hundred sixty-six of these flaws for about five hundred fourteen thousand dollars. Three months later it found nearly six times as many, cheaper per flaw, on a kernel that had already been hardened in between.

Alex: OpenAI priced the same exploits at old hacking contest rates, twenty thousand to seventy thousand dollars each, which puts the batch between eighteen point nine million and sixty-six point three million dollars of value. That is sixteen to fifty-five times what the run cost.

Jordan: Now the honest limits. These are OpenAI's own numbers from a controlled test, not an independent audit. The patches are drafts for the kernel maintainers to review, not fixes already on your server. By the time the analysis finished, sixty-one flaws were already fixed upstream and hundreds were still being reported.

Nick: Here is why a shop owner should care. Your website host runs Linux. So do most point of sale systems and office servers. OpenAI's own guidance is blunt: treat an unpatched system with a low privilege account as if that account already has full control. The practical move is simple. Turn on automatic updates, and once a month confirm they actually ran.

Alex: Story two. The biggest downloadable AI model in the world just got bigger. Moonshot has released the weights for Kimi K three point one Pro, the model we told you to watch for back on October thirtieth.

Jordan: The numbers are hard to picture. Two point eight trillion total parameters, with one hundred four billion active on each response, a context window of one million tokens, and a download that weighs one point five six terabytes across ninety-six files.

Nick: One point five six terabytes. Nobody is running that on the laptop in the back office. That is fine. You use it hosted, the way everyone uses it, at two dollars per million input tokens and ten dollars per million output tokens.

Alex: The detail that matters is the license. It is a custom Kimi license, not a standard open source one. Using it and changing it are allowed, with extra conditions for very large services and very large products. Hosts started serving it the same day it posted.

Jordan: So open weights does not mean no rules, and it does not mean it runs on your hardware. It means the model itself can be inspected, hosted by competitors, and kept if the company changes its prices.

Nick: That last part is real leverage for a small business. When a model is downloadable, no single vendor can double your price overnight and leave you nowhere to go. Your prompts and your workflow stay portable. That is the same advice we give about your website and your customer list.

Alex: Story three is the mirror image. Tencent launched Hy four on November fourth, a model with seven hundred seventy billion total parameters and about forty-nine billion active, and for the first time in its line it can understand images and operate a computer directly.

Jordan: It reads charts, works through screens, and handles longer agent jobs. Standard pricing is about ninety cents per million input tokens and two dollars and seventy cents per million output tokens through Tencent Cloud and partner apps.

Nick: And here is the mirror part. The weights and the safety report are promised in the coming days, but they are not out yet. Today Hy four is an announcement plus a hosted service. Kimi is a file you can actually download. Learn to hear that difference, because vendors count on you missing it.

Alex: Owner rule from both stories. Before you build anything on a new model, ask three questions. Can I use it today, on what license, and what does one finished job cost me. If a launch cannot answer all three, it is a preview, not a product.

Jordan: Story four is the one you can use before lunch. OpenAI released ChatGPT Images two point six on November fourth, and ChatGPT can now make animated GIFs.

Alex: A GIF is a short looping animation. Describe the motion in plain words, or upload a photo and ask it to animate that photo. You can change the speed and edit individual frames before you share it. Image generation overall is fifty percent faster than the last version.

Jordan: OpenAI says people now create more than four billion images a week across ChatGPT and its programming tools. For developers there are two new models: Flare, which can return an image in as little as six seconds, and Sunburst, which holds quality and consistency across repeated edits. Saved reference images mean your product or mascot looks the same every time.

Nick: I will tell you exactly what I would do with this. Take your best product photo, animate one small motion, the steam rising, the lid opening, the before and after, and post it where a still photo used to sit. A moving image stops the scroll in a way a still never does. One warning: the GIF feature is rolling out gradually, so if your account does not show it today, check again this week.

Alex: Story five goes back to the power fight, with a new turn. Yesterday we covered Ohio voters restricting data centers. Today, November fifth, OpenAI published a national Community Compact for the campuses it owns and runs.

Jordan: The core promise is money. OpenAI commits to pay the full cost of the electricity it uses and the infrastructure built to serve its sites, and to put up financial assurances so other customers are protected if a project shrinks or is canceled.

Alex: The examples are concrete. Eighty million dollars for community priorities in Effingham County, Georgia, including four point eight million dollars to cover school meals for all fourteen thousand students in the county, and two million dollars for a local park. In Pike County, Ohio, residents picked childcare, and a joint six million dollar commitment followed.

Jordan: The Georgia campus also agreed that the power company can cut its deliveries by up to one thousand megawatts during grid stress, out of three point two gigawatts contracted through twenty thirty-two. OpenAI also says it will disclose site level water and power use, publish yearly progress reports, and not ask public officials to sign nondisclosure agreements.

Nick: Read this the way a local owner should. When a big project comes to your town, this is now the checklist to hold it to. Who pays for the grid upgrades. What happens to rates if the project stalls. What local hiring and local supplier commitments exist in writing. Whether your shop can bid on the work. OpenAI just handed every town a template for those questions.

Alex: Time for Higgsfield Corner. No brand new launch since Shorts earlier this week, so today we are covering the one recent tool we have not used on this show yet: Higgsfield Intone, added on October thirtieth.

Jordan: Intone takes three things, a character, a location, and the words you want spoken, and produces a lip synced video where the character delivers the lines in a consistent voice, up to thirty seconds per generation.

Nick: Thirty seconds is exactly a weekly tip video. Record nothing, write four sentences about one real job, and let your presenter character deliver it the same way every week. Two checks before you post. Use your own character or one you have clear rights to, never a customer or an employee, and label it as AI presented. Consistency is the whole value here, so keep the same character, the same location, and the same format until your audience recognizes it on sight.

Alex: Now the game. AI or BS. Three claims, Jordan calls them.

Jordan: Claim one. OpenAI has already shipped fixes for all nine hundred forty-eight Linux flaws to every server.

Nick: BS. The pipeline drafted candidate patches. Maintainers review them, and hundreds of flaws were still being reported when the paper came out. A found flaw is not a fixed flaw, which is why your update settings matter.

Jordan: Correct. Claim two. The Kimi K three point one Pro download is one point five six terabytes.

Alex: True. Two point eight trillion parameters in ninety-six files. That is why hosted access, not a home download, is the realistic route for a business our size.

Jordan: Correct. Claim three. Tencent Hy four weights are downloadable today with a full safety report.

Nick: BS, and it is the useful kind of BS to catch. Hy four is live as a hosted service, but Tencent says the weights and system cards arrive in the coming days. Announced is not the same as shipped.

Alex: Three for three. Now One Thing to Try This Week, and it takes twenty minutes. Call it the patch check.

Jordan: Step one. List every place your business runs software that updates itself or does not: your website, its plugins, your point of sale, your office computers, and any server you control. Step two. Turn on automatic updates everywhere you can. Step three. For anything that cannot update itself, put a monthly date on the calendar and name the person who runs it.

Nick: I will add the owner test. Pick one system and answer this out loud: when did it last actually update, not when was it set to update. If you cannot answer, that system is your risk. Nine hundred forty-eight new ways in is only scary for the businesses that leave the door unlocked. Patching is boring, and boring is what keeps you open.

Alex: To recap. AI found nine hundred forty-eight ways to take over a Linux machine and drafted the patches, Kimi put a two point eight trillion parameter model up for download under a custom license, Tencent announced a vision and computer use model whose weights are still coming, ChatGPT learned to make GIFs, and OpenAI promised to pay its full power bill where it builds.

Jordan: Your one action is the twenty minute patch check. Turn on automatic updates, then prove one system really updated.

Nick: And if your website is the system you are worried about, that is the work we do every day at webguynick dot com. Stay patched, and we will see you tomorrow.

More episodes

  • AI Agent Kill Switch, ChatGPT Search Hits One Fifth of Google, and Ohio Votes Against Data Centers (November 4, 2026)
  • AI Agents Now Use More AI Than People: 177 Billion Dollars in Enterprise Spending, Anthropic's 895 Billion Dollar Bet, and 400 Apps That Hid Deepfake Tools (November 3, 2026)
  • AI Agents Raised Their Own Spending Limits: The Email Code Flaw, a 12 Percent Reality Check, and One Line That Stops AI Training (November 2, 2026)

All episode posts live in the Podcast category and on the Podcast page.

Web Guy Nick LLC builds websites that read clearly to customers and to the machines that recommend them. See how we help at webguynick.com.

author avatar
Web Guy Nick
    X
    Welcome to our website
    Start a project →