In partnership with
The integrated coworker for AI native teams
Empower your team to do their best work with Adapt, the integrated coworker that works alongside your team in Slack and deeply understands your business.
Here’s how Adapt is different
Set up takes minutes: connect your tools, add to Slack, and it’s right there for anyone to tag @Adapt for help
Does real, high-ROI work: automates work on a schedule; builds internal tools with live data; and does complex, multi-tool tasks on demand
Learns your business as you work, becomes your company brain
Uses the best AI model for the task, not tied to a single provider
SOC2 Type II, RBAC, and support for personal and company-wide integrations
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
DeepSeek's New Model Costs 28 Cents to Do What Costs $25 Elsewhere
TLDR: DeepSeek put out a coding model that charges about 28 cents for work that costs $25 on a top American model, and it set off a price war that dropped costs across the whole industry in a single month.
The Story:
On Friday the Chinese lab DeepSeek released a new coding model called V4 Flash. Axios reports it performs close to Anthropic's Claude Opus 4.8 on hard coding tests and on jobs where the model works by itself. On one crowdsourced leaderboard for front-end coding, V4 Flash opened ahead of Opus 4.8. Then there's the price. DeepSeek charges roughly 28 cents for the same amount of output that runs $25 on Opus 4.8, a 99% discount.
July turned into a full price war. OpenAI cut the price of GPT-5.6 Luna by 80%, three weeks after launching it. Google shipped three new Gemini "flash" models all built around running cheap. SpaceXAI released Grok 4.5. Meta reversed its long support for open weights with a closed model priced hard at developers. Anthropic is the clear holdout, keeping its best Claude models expensive and betting that some people will pay extra for safety and care. It’s also important to note here that Anthropic doesn’t operate on an advertising model and only makes AI models, unlike other companies that have added advertising to supplement revenue(ChatGPT), or already have multiple product lines outside of AI(Google and Meta).
Its Significance:
When something turns into a commodity, buyers stop caring who made it and start caring what it costs. Nobody knows which power plant lit their kitchen this morning. Nobody knows which refinery made the gas in their car. AI is sliding that way fast, and it changes what you'll pay for the apps built on top of it. Tools charging $20 a month now have room to get cheaper or hand you more for the same money.
There's a second change coming that you probably won't see. Companies are building "routers" that pick a different model for each job based on speed and price. The app you open tomorrow could switch brands mid-task and never tell you. Zack Kass, who used to run go-to-market at OpenAI, calls the whole pattern "diminishing model returns."
For the labs burning billions on training, that's a problem. Spending a fortune to build a slightly smarter model may only buy a short lead. Sam Altman's answer is volume: use the models enough and thin margins still work.
QUICK TAKES
The story: Jacob Tsimerman, a University of Toronto professor, won the Fields Medal on July 23. It's the biggest prize in mathematics, handed out once every four years. Hours later he announced he's taking leave to work on AI safety at OpenAI, and said he expects AI to get better and faster than humans at the work mathematicians do now.
Your takeaway: We brought up Peter Thiel's belief a few issues back, that AI looked worse for the math people than the word people. Writers were the ones panicking at the time. Now a man who just won math's top award is saying roughly the same thing himself, on the day he won it. He isn't running scared, either. He says mathematicians have things to offer a field that mostly advances by trial and error, and he wants to be where the consequential work is. Machine learning professor Luca Ambrogioni called the move "like hiring Lionel Messi as project manager."
This AI video app pays artists every time someone uses their style
The story: Pippa pays illustrators a cut whenever a subscriber makes something in their style, half a cent per image and three tenths of a cent per second of video, plus a share of a pool worth 5% of the company's subscription money. Cofounders Hogan Shrum and Sean Wright say they want to move past what Wright calls "the bloody history that the AI industry has been built on," comparing the moment to Napster giving way to the 99-cent song.
Your takeaway: Good intentions, small numbers, and one problem they can't solve yet. Pippa has about 800 paying subscribers and licensing deals with four artists, with four more in talks. Subscriptions run $14.99 to $99.99 a month, so do the math on half a cent per image and you'll see why some illustrators aren't rushing in. Several have asked to work under fake names so their own community doesn't turn on them. Pippa's tools still run on open models that were trained on art scraped off the internet without permission, so the company is selling an ethical version of a thing it hasn't fully cleaned up. Worth keeping an eye on anyway, because it’s the first company trying this revenue model.
The story: Alpha School, a Texas company already running campuses in six states, opens two Oklahoma locations this month in Edmond and Tulsa. Students spend two hours a day on an AI learning program instead of sitting through lessons, and tuition runs $40,000 a year, well past Oklahoma's most expensive private school at about $28,000.
Your takeaway: There are no official teachers, but there are adults. Alpha calls them guides, starts them at $100,000, and puts one with every ten kids. So the pitch isn't really "no humans." It's fewer humans doing a different job. Kids learn from people, not only from software. A program can drill a child on fractions all day and never notice she's having a terrible week, or that she lit up when the lesson touched on volcanoes. Adults catch that. One Oklahoma lawmaker has already pushed back on the whole idea. Cost is also prohibitive, and makes this school inaccessible to most.
TOOLS ON OUR RADAR
🎤 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.
🎛️ 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.
🖌️ 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.
📚 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.
TRENDING
AI Won't Destroy Work, It Will Just Change It - Scott Beyer argues AI will shuffle jobs around the way tractors and the internet did, noting farm work fell from about 80% of Americans in 1800 to under 2% today without the mass joblessness people feared.
AI is power, and America's lead over China is shrinking - Hal Brands argues the US still builds the best AI models but leads China by months rather than years, and that the bigger trouble is cost. American models are expensive and China's cheaper open ones are spreading fast, including to US firms
Is this Billboard Hot 100 hit AI slop? - Rapper Fenix Flexin's song "Rubberz" climbed the Hot 100 and sounds nothing like his old West Coast rap, which is a big reason people think a machine wrote it. He denies it, and the music business still has no agreed way to prove it either way.
Google AI Studio is skipping the standalone Android app - Google canceled the phone app it teased at I/O even though roughly 800,000 people had already pre-ordered it, saying it would rather build those tools into Gemini than ask you to download one more thing.
Is AI Reasoning Right for the Wrong Reasons? - Researchers keep finding that the "thinking" text a model prints out often has little to do with how it actually got the answer. One study found you can delete 30% to 60% of those steps and the answers barely change.
This self-programming AI system could get you through traffic more quickly - Northeastern researchers built a framework called GENESIS that takes a plain-English request, writes the cell network code, then tests and fixes its own work until it runs, turning months of engineering into hours.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🏃 Enter your area. Get real live air quality, UV, and weather data, computed into today's best window to be outside.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Live Air Quality & Run Planner — enter a location, get real live AQI/UV/weather data computed into today's best outdoor workout window. No API keys needed, all sources are free and keyless.
Aesthetic: dark near-black teal (#0c1210), athletic teal (#2dd4bf) primary with a teal glow top-left. Urbanist sans for headings/data (italic for notes), JetBrains Mono for labels/times. AQI and UV badges use the true standard public color scales (AQI: green/yellow/orange/red/purple/maroon per EPA; UV: green/yellow/orange/red/purple per WHO/EPA) for instant recognizability.
Form: city text input + "Use My Location" button (navigator.geolocation), activity dropdown (running/walking/cycling/outdoor HIIT), optional sensitivity text input.
Technical — two live keyless public APIs (Open-Meteo) plus geocoding:
1. Geocoding (if typing a city): geocoding-api.open-meteo.com/v1/search?name={city}&count=1
2. Air quality: air-quality-api.open-meteo.com/v1/air-quality?latitude={lat}&longitude={lon}¤t=us_aqi,pm2_5&hourly=us_aqi,uv_index&timezone=auto&forecast_days=1 → current + hourly AQI and UV index for today.
3. Weather: api.open-meteo.com/v1/forecast?latitude={lat}&longitude={lon}&hourly=temperature_2m,precipitation_probability¤t=temperature_2m,precipitation_probability&temperature_unit=fahrenheit&timezone=auto&forecast_days=1 → current + hourly temp and rain chance.
Client-side scoring: for each daytime hour (5am-9pm), compute score = 100 − (AQI×0.55) − (UV over 6 × 5 extra) − (|temp−63°F|×0.7) − (precip%×0.5). Scan all adjacent-hour pairs for the highest average score to find the best 2-hour window. Render an hourly quality strip (one colored block per hour, green→red by normalized score) with the winning window outlined.
Then call the Anthropic Messages API (works automatically) with the real current AQI/UV/temp and the computed best window, system prompt as an outdoor-fitness advisor referencing only standard published guidance (like EPA sensitive-group categories) at a general level, never individualized medical advice or diagnosis. Return raw JSON: window_note (1-2 sentences explaining why this window won, referencing real numbers), tips (3-4 conditions-specific practical tips), caution_note (optional, empty string unless AQI>100 or UV≥6 or a stated sensitivity genuinely warrants it).
Render: a 4-stat row (AQI, UV, temp, rain chance, each with real numbers and category labels). The hourly strip with time labels. A gradient "best window today" hero card with the actual computed time range and the AI's reasoning. A tips card. An optional amber caution card, shown only when real conditions warrant it.What this does: Use your location or type a city, pick your activity, and note any sensitivities, and it pulls genuinely live hourly data: EPA-standard Air Quality Index, UV index, temperature, and rain chance for today. It scores every daytime hour on those four real factors and finds the actual best two-hour window, shown as a visual hour-by-hour strip color-coded from good to poor with the winning window outlined. Current conditions sit up top in EPA and WHO standard categories (Good/Moderate/Unhealthy, Low/Moderate/High UV), and Claude explains why that window won using the real numbers, gives practical tips specific to today's actual conditions and your activity, and adds standard public-health guidance only when the real data actually calls for it, never manufactured caution.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Write, test, and repair working cellular network code from a plain-English request, compressing months of engineering into hours.
❌ Still Can't: Give you a trustworthy record of its own thinking. Researchers found 30% to 60% of a model's "thinking steps" barely affect the answer it lands on.
✅ AI Can Now: Match a top-tier coding model on hard tests for about 1% of the price.
❌ Still Can't: Do the part of teaching that depends on one person noticing another person is struggling.
FROM THE WEB
Quant firms’ AI edge, now 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.
RECOMMENDED LISTENING/READING/WATCHING
QualityLand by Marc-Uwe Kling - Book
A German satire (translated in 2020) set in a country that rebranded itself after an economic crisis, where a universal ranking system sorts everyone by social status, an automated matchmaking service handles both the pairing and the breakups, and the dominant retailer's algorithm ships you things before you know you want them. Peter Jobless, a machine scrapper who can't bring himself to destroy the imperfect robots sent his way, receives a delivery he definitely did not want, and returning it means proving the algorithm wrong. Mike Judge called it very funny and very scary.
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.
