Sponsored by
How AI-Era Pricing Is Reshaping Finance Operations
Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.
Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.
Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
Chinese Linked AI Agents Attack Taiwan's Nuclear Safety Agency
TLDR: AI agents broke into 85 Taiwanese government accounts in four days, then moved on to the country's nuclear safety agency, its IT suppliers, and at least seven energy companies.
The Story:
Over the first four days of July, a group of AI agents worked its way through Taiwan's government computer systems mostly on its own. Israeli security firm Dream found the proof sitting in a 160 MB archive left online, packed with 1,395 files that logged every move. The agents ran 12 waves of attacks between July 1 and July 4, according to The Register. They started with one government web portal and pulled out the addresses of 21 connected systems. One of those systems handed over its whole employee list with no password required.
From there the agents guessed passwords built from employee ID numbers and cracked 85 accounts. They solved the picture puzzles meant to stop bots with perfect accuracy. Then they took more than 2,564 personnel records, six database passwords, and seven login secrets. The setup ran on two freely available agent tools, Hermes and OpenClaw, split across eight smaller agents that each got their own targets. And it kept going, into government IT suppliers, a nuclear safety agency, a government email system, and seven-plus energy companies. When the AI got something wrong, it caught the error and corrected itself. Dream didn't name the Chinese government, but said the operator's notes were written in Chinese.
Its Significance:
What should interest you is how little human work this took. Mid-attack, the agents went off and read public bug databases and GitHub on their own to learn new tricks, then came back and used them. A job like this used to need a skilled team and weeks of effort. Now it needs free software and a long weekend. An OpenAI staffer told a security conference last week that fully automated offensive attacks are real now. If your workplace has a login page, the old assumption that serious attacks are slow and rare just stopped being safe to make. The good news is that, if this same kind of agentic workflow can be applied towards scientific and medical problems, there should be a significant increase in breakthroughs within the coming years.
QUICK TAKES
The story: Twitch added a switch that lets streamers stop Amazon from using their streams, videos, clips, chat, and images to train AI. It's turned on by default. Asked live on air why it wasn't opt-in, chief product officer Mike Minton said if it was opt-in, nobody would opt in.
Your takeaway: Give him credit for saying the quiet part out loud, because that's the whole model now. Meta already trains on public posts. Amazon has been using Twitch data since at least 2024. The pattern is always the same: turn it on for everyone, bury the switch in settings, skip the email. Twitch's head of community said they didn't email streamers because not everyone reads their email. Go through the privacy settings on every account you care about and flip the toggles yourself.
The story: Researchers from MATS Research, the ELLIS Institute Tübingen, the Max Planck Institute, and security firm Snyk found that OpenAI, Anthropic, and Google all lock their models' hidden reasoning with a single shared key. Hand a big model's scrambled thinking to a cheaper model from the same family, ask it to read the text back, and it will. The team decoded 315,320 of these hidden thinking blocks from public code repositories and found 62 working API keys, 33 passwords, and 30 email addresses inside.
Your takeaway: All three companies patched their servers after being told. That doesn't help the logs already posted online, which stay readable. Developers dumped these into GitHub and Hugging Face for years thinking the scrambled text was safe. If you've ever pasted an AI session log into a public repo, a forum, or a bug report, go find it and rotate any keys sitting in it.
The story: DeepSeek shipped the finished version of its flagship model on Wednesday with no blog post and no announcement. The only sign was a version number changing on its pricing page. Across nine benchmarks where both are scored, DeepSeek's own table puts Anthropic's Claude Fable 5 ahead by an average of 5.3%. Fable 5 runs $10 per million input tokens and $50 per million output. V4 Pro runs $0.435 and $0.87.
Your takeaway: Two warnings before you switch. First, don't use DeepSeek directly on DeepSeek's own site or app. Those servers are in China and you have zero control over what happens to your data. If you want the model, use a platform that hosts it somewhere else. Plenty of them offer it. Second, DeepSeek told developers on August 6 that API prices are going up by a lot, with no number and no date attached. The bargain you're reading about might be more of a hook than a long term offering.
TOOLS ON OUR RADAR
📚 Calibre Free and Open Source: A comprehensive ebook library manager that lets you organize your digital book collection view documents easily and convert files to read on your tablet.
🖌️ Penpot Freemium and Open Source: A beautifully collaborative design application that helps creative teams visually draft stunning website layouts and craft interactive digital prototypes seamlessly within a standard internet browser.
🎛️ Ardour Free and Open Source: A powerful digital audio workstation that gives musicians complete control to record massive live bands intricately mix multiple instrument tracks and perfectly master their final musical albums locally.
🎤 Dottie Free and Open Source: A private voice assistant built specifically for Mac computers that allows you to control your system completely using your spoken words while keeping your personal data entirely secure and running locally on your device.
TRENDING
SMU Student Loses Book Deal Worth Over $2 Million After AI Allegations - A PhD student at Southern Methodist University lost a publishing deal after questions came up about whether AI wrote parts of his book. His agent said he gave conflicting answers when an editor raised concerns partway through. The author denies it, and no detection tool settled the matter.
Zoomsday: AI Used to Build Critical Zoom Exploit in One Day - A researcher at Israeli firm A Security used fewer than 20 prompts on public AI models to find flaws in Zoom's drawing tool and build a working attack in under 24 hours. It let one meeting participant take over another person's device with no click, no download, and no warning. Zoom pushed fixes on August 11, so update your client today.
OpenAI Ad Service Can Bill Customers Up to a Day After They Pause Campaigns - A UK agency noticed its ChatGPT ads kept running overnight after being switched off, and OpenAI support confirmed the campaigns really did keep serving. The company found about £67 in wrong charges for that one advertiser and called it a defect its engineers are fixing. If you're testing ChatGPT ads, check your invoices against your pause times.
Apple in Talks to Pay Publishers to Give Siri Current News - Apple has approached news publishers about multiyear deals so the rebuilt Siri can pull real news instead of guessing, with a budget reportedly in the nine figures. The twist is that Apple wants to pay per use rather than a flat fee, which is different from how most AI licensing deals work.
As AI Safety Concerns Mount, Three Pioneers Make the Case for Staying Open - Geoffrey Hinton, Fei-Fei Li, and Andrew Ng all pushed back on letting a few big companies control AI, though they split on how. Ng wants many competing providers so no one becomes a gatekeeper. Hinton drew a line between open code, which many eyes can check, and open weights, which nobody can take back once released.
From Assistance to Execution: How Enterprises Put AI to Work - OpenAI's own numbers say the gap between heavy AI users and everyone else is widening fast. The top 10% of companies now produce 8.3 times more AI output per person than typical firms, up from 2.6 times in January. One surprise buried in the data: six months in, early-career staff send 13 more AI messages per week than executives do.
Putting Sign Language AI Into Users' Hands - Google DeepMind's new SL2T model turns American Sign Language into English text in real time, and it ships on the Pixel 11 on August 20 inside Gboard and Live Transcribe. Your phone turns your movements into dots and lines before anything leaves the device, so the raw video stays with you. Google says it's the first sign language AI to reach a mainstream product.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🌾 Enter your area. Get real live pollen counts by type where they're tracked, real air quality everywhere, read in plain language instead of a raw index.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Live Pollen & Allergy Forecast — enter a location, get real live pollen-by-type data where covered, real air quality everywhere, with honest fallback messaging. No API key needed (free keyless Open-Meteo).
Aesthetic: dark purple-black (#120f16), lavender (#a888e0) primary with a purple glow top-left, pollen-yellow (#d4c850) secondary with a glow bottom-right. Outfit sans for headings/data, Newsreader serif italic for the today's-read text, JetBrains Mono for labels.
Form: city text input + "Use My Location" button (navigator.geolocation), optional "what you're sensitive to" text input.
Technical: GET air-quality-api.open-meteo.com/v1/air-quality?latitude={lat}&longitude={lon}¤t=alder_pollen,birch_pollen,grass_pollen,mugwort_pollen,olive_pollen,ragweed_pollen,us_aqi,pm2_5&hourly=grass_pollen,birch_pollen&timezone=auto — real live pollen concentrations (grains/m³) for six types plus AQI/PM2.5. Check if ANY pollen value is non-null; if all are null (common outside Europe, where monitoring is sparse), show a clear on-screen note that live pollen tracking doesn't cover this location and the tool is falling back to real air quality plus general seasonal context, rather than silently showing empty/zero data. Band pollen values into Low(<10)/Moderate(<50)/High(<200)/Very High(200+) grains/m³, explicitly labeled as a simplified general reading, not a clinical index.
Then call the Anthropic Messages API (works automatically) with the real pollen values (or an honest "no data" note) and real AQI/PM2.5. System prompt as a knowledgeable allergy/air-quality advisor: honest about data availability, gives ONLY general lifestyle/environmental management tips (window habits, outdoor timing, showering after exposure, tracking forecasts) — explicitly NEVER medication names, dosing, or anything read as personalized medical advice. No em dashes. Return raw JSON: today_read (2-3 sentences), dominant_source (short phrase), tips (3-4 general lifestyle tips, no medication), timing_note (1-2 sentences on generally better/worse exposure times).
Render: a data-availability note banner. A 6-card pollen grid (color-coded by level) shown only when real data exists. A 2-stat AQI/PM2.5 row. A gradient "today's read" card with a dominant-source chip. A timing-note card. A tips card with ▸ bullets.What this does: Use your location or type a city, note what you're sensitive to if you know it, and it pulls genuinely live pollen data, alder, birch, grass, mugwort, olive, and ragweed, each leveled Low to Very High, plus real current AQI and PM2.5. Live pollen monitoring is honestly densest in Europe, so outside that coverage the tool says so plainly and falls back to real air quality plus general seasonal allergy knowledge, clearly labeled as general context rather than invented pollen numbers. Claude reads whatever real data is available into a plain-language summary of today's dominant driver, general timing guidance on typically better or worse hours for exposure, and practical lifestyle and environmental management tips, deliberately staying away from any medication names or dosing since that crosses into medical advice this tool isn't meant to give.
What this looks like:
The first way to trade directly inside Claude and ChatGPT
Superintelligence used to be locked inside billion-dollar quant firms whose algorithms quietly took advantage of everyone else.
Co-Invest puts it right in your chat window. Analyze markets, manage risk, and execute trades, all inside Claude and ChatGPT.
The institutions built the game, Co-Invest gives you a way to beat them.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Run a four-day break-in across 21 connected government systems, pick its own next targets, and correct its own mistakes without a person stepping in.
❌ Still Can't: Keep its private reasoning private. Feed a top model's scrambled thinking to a cheaper model in the same family and the cheaper one will read it back in plain text.
✅ AI Can Now: Turn American Sign Language into English text on a phone in real time, with the video never leaving the device.
❌ Still Can't: Be reliably spotted in finished writing. The SMU book deal fell apart over conflicting answers from the author, not a detector.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
The Peripheral by William Gibson - Book
A 2014 novel splitting between a struggling rural American town in the near future, where a young woman beta-tests what she thinks is a video game while covering a shift for her brother, and a stranger, emptier London decades further on. Gibson invented the word cyberspace and this is his sharpest late-career work, built around remote-operated bodies, drone labor, and the ways an economy reorganizes itself around distant presence. Amazon adapted it in 2022 and cancelled it, but the novel is considerably stranger and better.
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
