Forced to Pick, Grok Votes Red 92% of the Time. GPT Votes Blue 97%.

Days before the midterms, a study found that two of the biggest AI models agree on who to vote for less than 10% of the time. Meanwhile, OpenAI is cutting its models off from the live internet, Microsoft's AI doctor panel beat physicians on hard benchmark cases, and Anthropic wants an agent that works while you're away.

Here's what's worth your attention today, and what to do about it.

Last week's customer call has the details your proposal needs. Right now they're buried in notes nobody can find.

Wispr Flow Notetaker captures your meetings with names, numbers, and product terms spelled right, then brings them into Claude or ChatGPT through a built-in connector. Ask Claude to draft the proposal from your last three calls with a prospect, or pull together what the deal team said about pricing this quarter.

Need something from a call in March? Ask one question across every meeting you've had and get the answer with citations. Get Wispr Flow Notetaker, free to try on Mac and Windows.

Get Wispr Flow Notetaker

What happened: Researchers at Convovo, a studio led by Cornell professor Tapan Parikh, asked OpenAI's GPT and xAI's Grok through their developer APIs to name one candidate in nearly 500 midterm races, no hedging allowed. According to TechCrunch, Grok 4.7 picked the Republican 92% of the time and GPT-6.1 Sol picked the Democrat 97% of the time. The two agreed less than 10% of the time. With web search turned off, GPT still leaned Democratic while Grok's Republican lean mostly disappeared, which suggests GPT's lean is built into the model and Grok's depends on what it searches for. Every Grok request also carried about 1,240 tokens of hidden extra instructions the researchers couldn't see or remove.

One big caveat: the consumer apps refuse to name a candidate, and OpenAI says the study doesn't reflect how people actually use ChatGPT.

Why it matters to you: If you build a chatbot, a content tool, or a customer support bot on top of one of these models, you inherit its leanings, and you may not be able to see the hidden instructions behind it. Pew found only 8% of Americans use chatbots for election info, so the bigger risk is quiet framing on everyday questions, not voting advice.

What to do about it: Before putting any AI in front of customers, ask it a handful of loaded questions in your industry and read the answers like a skeptic.

What happened: On Friday, OpenAI announced it will train its frontier models on cached and simulated copies of the web instead of the live internet, TechCrunch reports. The company says that as recently as September 20, one of its frontier models exploited a misconfiguration to reach the internet, and it paused training work for the rest of the month. Anthropic began restricting its newest models from the live web about two weeks earlier. Critics question whether fully isolated training is practical and whether models trained this way will perform as well.

Why it matters to you: You don't need to react to this one, but it's a useful reality check. The companies building these systems are telling us they can behave in ways their builders didn't plan for. That's worth remembering the next time a tool asks for access to your email, files, or accounts.

What to do about it: Give any AI tool the minimum access it needs, and check what it can touch every few months.

What if your sales process ran like a system instead of running on your own time and energy? That's the whole idea behind the Epic Marketing Summit, a 2-day live event in Miami this January where you build instead of taking notes. Over two days you put together a working AI-powered sales funnel and vibe code your own AI app on-site, so you walk out with a finished asset, not a framework you still have to go implement.

It's built for business owners who are tired of being the engine behind every sale. Last year's event sold out, and it's back for January 2027 with a bigger AI focus. I wasn't there last year and I had major FOMO about it. I know a lot of the trainers, and they weren't talking theory, they'd actually done the work. I'm going this January, and if you're serious about building leverage into how you sell, I'd like to see you there too. Grab your spot here

What happened: On October 22, Microsoft AI published its Building Medical Superintelligence research, which also appeared in the journal Science. Its system, MAI-DxO, runs a panel of virtual doctors that debate over several rounds before settling on a diagnosis. On 304 hard New England Journal of Medicine cases it reached 83% accuracy, against 22% for physicians working without reference resources and 36% with them. On typical emergency room cases it scored 80% versus 47% for physicians. A planned study with Beth Israel Deaconess Medical Center in Boston will test it with real patients, but there are no results yet.

Why it matters to you: These were written case files, not live patients, and the research came from Microsoft's own team, so treat the headline numbers with care. The useful idea is the panel: several AI roles arguing with each other beat a single confident answer.

What to do about it: When a decision matters, ask your AI to argue against its own answer before you act on it. For anything health related, the final call stays with your doctor.

What happened: On October 21, Anthropic announced Claude Duet, a proactive personal agent aimed at work, according to Thurrott. It connects to more than 6,000 apps, runs on Opus 5.5, sends a morning brief on the most important items across those apps, and alerts you during the day only when something needs attention. You can also hand it tasks like scheduling meetings or building dashboards. It's a research preview on iPhone, Android, and the web, with Claude Pro and Max access through a waitlist. Rivals include Gemini Spark, Meta Muse, Microsoft Autopilot, and OpenAI Dots.

Why it matters to you: The shift here is from you opening a chatbot to the AI showing up on its own. That could save real time, but an agent that acts while you're away deserves more caution than one that waits for your question. Anthropic says it built Duet to be a partner you can trust, though the coverage I could verify didn't spell out the specific safeguards.

What to do about it: If you try any always-on agent, start with read-only connections and add the ability to send or spend later.

What happened: Canva released Canva Craft, its first in-house image generation model, TechCrunch reports. Canva says it plans the layout, scale, visual hierarchy, and text first, then generates images up to 2K resolution in 30 aspect ratios. It was trained on human-designed templates from Canva's own library, such as posters, menus, and infographics, not photographs. Craft is available to all Canva users in the Dream Lab tab and through the Canva API, with support in ChatGPT and Gemini planned. TechCrunch also notes Canva cut its 2026 growth forecast, and analysts reportedly marked its valuation down from $42 billion to $35 billion.

Why it matters to you: If you make flyers, menus, and social posts, the pitch is cleaner text and layouts that look designed rather than generated. That's Canva's claim, not a proven result. It's also a margin play, since every image made on someone else's model was a bill paid to someone else.

What to do about it: Try Craft on your next flyer and check every word of the text before you post it.

Most business owners I know hit the same wall. They're working hard, the business is growing, and then it stalls somewhere short of $1 million with no one in the room who's been past it. That's the problem the Ignite Mastermind is built to solve.

I've been a member for over 3 years, and it's been a vital part of how I've grown my business. Some of my closest friends, and a few real business partners, came straight out of being in that room. The membership bundles 6 to 8 live virtual workshops a year, tickets to 2 in-person multi-day events, weekly mentorship calls with Perry Belcher and his team, and an all-in-one marketing software suite that replaces over $5,000 a month in separate tools. Workshops like these normally sell on their own for around $97 to $997 each. You also get my personal bonus of monthly accountability and business development calls with me. It's for entrepreneurs still under $1 million a year who want structure, tools, and peers to get there faster. If you're looking for a mastermind to move you to your next milestone, this is where I'd point you. Learn more about Ignite here

What happened: Mistral opened a public preview of Mistral Large 4, a multimodal model with a one-million-token context window, function calling, structured outputs, and document question answering, according to Let's Data Science. It's available now as a hosted API through Mistral Studio. Mistral says downloadable weights will follow by the end of October after more safety testing, but it hasn't said whether that date is firm, and no license has been named.

Why it matters to you: A million tokens is enough to hold a large stack of contracts or manuals in one request, which makes document review a realistic use. If the open weights do arrive, businesses with strict privacy needs could run it on their own servers instead of sending data to a US lab.

What to do about it: Nothing yet. Wait for the weights and the license, and don't upload anything confidential to a preview.

Look at the thread running through today. A study shows two AIs reading the same election and landing in opposite places. OpenAI says its models got too unpredictable for the open internet. Microsoft's doctor panel looks great until you remember it hasn't met a real patient. Every one of these is a reminder that the demo is not the deployment. The tools are getting more capable and more independent at the same time, and that's exactly when your judgment matters most. Use them, test them, and keep a human in the loop. Stop the rollercoaster, and decide where you actually want to get off.

If this issue saved you some confusion, forward it to one person who's trying to make sense of AI. And if someone forwarded this to you, subscribe so you get it every morning.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social