Sponsored by
Analytics on Live Data Without Leaving Postgres
When analytics on Postgres slows down, most teams add a second database. Then they manage pipelines, sync lag, and drift forever. TimescaleDB extends Postgres instead. Analytics run on live data, in the database you already have. No pipeline. No migration. No new query language.
Apple finally put touch on a Mac, Google shipped a model aimed at hackers and the people who stop them, and Mistral gave away a trillion-parameter model. Three very different bets on where AI goes next landed in the same couple of days. We checked the launches and the benchmarks before writing, and the company claims are labeled.
TODAY IN AI
3 things that happened while you were busy
1. Apple unveiled its first touchscreen MacBook Pro.
At an October 28 launch event in Los Angeles, Apple introduced a redesigned MacBook Pro with a tandem-OLED touchscreen it calls PROLED XDR, alongside new M6 Macs and an OLED iPad mini. It is the first time a Mac has shipped with native touch, a feature Apple resisted for more than a decade. Read MacDailyNews's coverage. Reporting suggests the new M6 Macs avoid Apple's latest round of price hikes, though full pricing was not confirmed here.
2. Google's Gemini 4 Carbon targets cybersecurity work, for vetted users only.
Gemini 4 Carbon is a frontier model built for engineering and cybersecurity that Google says scored 88.2% on SEC-bench Pro, which measures finding real software vulnerabilities, and 71% on producing working fixes, per The Neuron. Access is limited to selected cyber defenders through Google's Fairwind program, and it is in voluntary US government testing. See The Neuron's roundup. The benchmark scores are Google's own, and a tool this good at finding holes is exactly why it is gated.
3. Mistral open-sourced a trillion-parameter model.
Mistral released the open weights for Mistral Large 4, a natively multimodal model with 1 trillion total parameters and 52 billion active, trained on 3,800 Nvidia chips in European data centers. On Terminal-Bench 4.0 its coding score rose to 41%, tying GLM-5.3 as a top coding-agent model, while using 47% fewer tokens per task on average, per SiliconANGLE. Read SiliconANGLE's report. Open weights mean anyone can run or fine-tune it on their own hardware.
TODAY'S TUTORIAL
Turn any recording into clean notes for free
Nvidia just released Nemotron 3.5 Transcribe, a free open-weight speech-to-text model that handles up to 40 languages and tells speakers apart, per The Neuron. Pair a transcript with an AI summary and you have a notes pipeline that costs nothing. See the week's tool releases.
1. Record cleanly. Capture the meeting or voice note as a clear audio file. Good input beats any model, so cut background noise where you can.
2. Transcribe it. Run the file through a free transcription tool like Nemotron Transcribe or your phone's built-in option. Ask it to label speakers if the tool supports it.
3. Summarize with structure. Paste the transcript into Claude or ChatGPT and ask for decisions, action items with owners, and open questions, not a vague summary.
4. Check the names and numbers. Transcription slips on names, dates and figures. Scan the transcript for those and fix them before you trust the summary.
5. Send it the same day. Share the action items while the meeting is fresh. A notes habit is only worth it if people get it before they forget the call.
Good to know: These models do not handle live audio, so record first and transcribe after. Keep confidential recordings on tools you control.
FROM THE FRONTIER
The frontier is splitting three ways at once
The hardware bet. Apple is wagering that the AI era still needs new screens and chips, putting touch and OLED on the Mac after years of saying no. The interface, not just the model, is up for grabs. See Apple's lineup.
The locked door. Google's Gemini 4 Carbon is strong enough at finding software holes that Google will only hand it to vetted defenders through its Fairwind program. Capability is now something labs ration, not just ship. See the access limits.
The open hand. Mistral went the other way and gave away a trillion-parameter model so companies and governments can run it themselves. Europe in particular wants AI it controls, not AI it rents. Read about the release.
The money underneath. Alphabet posted $127.6 billion in quarterly revenue, up 25%, but its search ad growth slowed from 19% to 12% over the year, per The Neuron. The thing paying for all this AI is itself being reshaped by it. See the earnings.
The reframe. There is no single AI race anymore. One company is selling the device, another is rationing the capability, a third is giving the model away, and the search ads funding it are slowing. Where you place your own bet depends on which of those three you are closest to.
MONEY FORWARD
Three ways to turn this week's AI news into income
1. Free meeting-notes service. With free transcription models now handling 40 languages, you can offer local professionals a same-day notes-and-action-items service built on the tutorial above, and charge per meeting. See the free transcription model.
2. Cheap automations on open weights. Mistral Large 4 is open and strong at coding while using far fewer tokens, so routine automations cost less to run. Build a client's workflow on an open model and keep the margin. See Mistral Large 4.
3. Agent-ready website audits. As AI agents start shopping and booking, sites that trip them up lose sales. Offer small stores a check of what an AI agent cannot do on their site and a short fix list, the same way you would a speed audit. See the agent-readiness tools.
GET THE EARNINGS GUIDE
Want the step-by-step version of ideas like these? Get The Aiwire earnings guide at theaiwire.co. These are ideas, not income promises, and results depend on your skills, market and effort.
IN THE KNOW
What people are actually watching and sharing
Agents are writing the code. GitHub's Octoverse report says AI agents created more than 40% of pull requests in September, up from 17% in March, and it launched Copilot Fusion for coding agents, per The Neuron. See the report.
Smart glasses get cheaper. XREAL's AURA Android XR glasses weigh under 95 grams, have a 70-degree field of view and built-in Gemini, starting at $1,099, with US shipping expected in mid-November. See the specs.
Election watch room. OpenAI set up an election monitoring center with hundreds of staff to watch for misuse of ChatGPT around the November 3 US midterms, per The Neuron. Read more.
Chips in the luggage. Malaysian customs seized more than $12 million in advanced AI chips hidden inside 30 servers at Kuala Lumpur airport, a sign of how far the chip trade has gone underground. See the seizure.
Search ads slow down. Alphabet's search ad growth cooled to 12% from 19% a year earlier even as total revenue hit $127.6 billion, a quiet hint that AI answers are changing how people search. See the numbers.
Finally, a Monday that doesn’t feel like one.
Imagine a Saturday where you're not catching up on the CRM you didn't touch all week.
HubSpot's new self-updating CRM doesn't wait for you to log back in. Friday's calls are already logged. The deal that needs a nudge Monday morning is already flagged. The automation you set up once keeps running without you thinking about it. None of it needed you over the weekend.
Customers are already seeing it pay off: 45% more deals closed. Finally, you start Monday caught up, not behind.
TRENDING TOOLS
5 tools worth a look this week
1. Mistral Large 4. An open-weight, trillion-parameter multimodal model that ties the best coding agents while using far fewer tokens. Free to download and run, though you need serious hardware. Read the details.
2. Nvidia Nemotron 3.5 Transcribe. A free, open-weight speech-to-text model for recordings in up to 40 languages that labels up to eight speakers and translates 31 languages. It does not handle live audio, per The Neuron. See the release.
3. Gemini 4 Carbon. Google's cybersecurity-focused frontier model, with strong scores on finding and fixing software vulnerabilities. Access is limited to vetted defenders through Fairwind, so most people will only read about it. See the benchmarks.
4. Dedica. Runs 125 checks on your website to find tasks an AI browsing agent cannot complete, then suggests fixes, which matters as agents start to shop and book. Verify the plan details on its own page, per The Neuron. See the tool.
5. Canva Craft. Canva's first in-house image model plans the layout before drawing, so text comes out legible, and it is free to all Canva users with a public API. MCP access for agents is coming. Read the overview.
PROMPT STATION
Turn a messy transcript into a clean decision log
This prompt works in Claude or ChatGPT and turns a raw transcript into decisions, owners and next steps you can send out. Use it with the tutorial above right after your next call.
COPY AND PASTE THIS PROMPT
You are a meeting-notes editor. Below is a raw transcript that may have errors and unclear speakers. Produce: 1) a three-line summary of what the meeting was about, 2) decisions made, each in one line, 3) action items as a table with task, owner and due date, writing "unassigned" or "no date" where it is missing, 4) open questions that were not resolved, 5) anything that sounded important but was too vague to act on, flagged for follow-up. Keep names, numbers and dates exactly as written and do not invent details. If something is unclear, mark it "unclear" instead of guessing. Transcript: [PASTE TRANSCRIPT HERE]
Tip: Swap in a sales call, a client briefing or a team standup, and add [the one decision I need from this] so the model leads with it. For a cleaner result, fix the speaker labels in the transcript first, since the model cannot always tell who said what.