Grok Dropped 2,000 Bombs. OpenAI Dropped $21B. America Dropped All Respect.

Also: Steal this elite AI prompt for a bulletproof xeriscape garden that laughs in the face of brutal drought.

AI dropped bombs this week. Then it went broke. Then it lost the public.

The Pentagon confirmed that Elon Musk’s Grok Gov Model helped U.S. forces deploy over 2,000 munitions at 2,000 targets in 96 hours during the Iran campaign. It’s the first-known time a commercial frontier AI tech has been officially placed inside a military kill chain. Meanwhile, leaked financial documents revealed OpenAI burned $21 billion more than it earned in 2025, and for the first time in its history, ChatGPT’s global market share has fallen below 50%. And a new Pew Research study of over 5,000 Americans found that nearly half the country now uses AI daily… while 40% expect it to make society worse.

Here’s what happened, and why this week told one story underneath three headlines. AI is now powerful enough to fight wars, expensive enough to bankrupt its inventors, and unpopular enough that the people using it every day still don’t trust it. That combination has never existed before. And nobody in the industry has a clean answer for what comes next.

🌳 Haven’t Subscribed Yet? → Join Here For Free

💙 Value This Weekly Briefing? → Support the Work Here

Grok Helped Deploy 2,000 Munitions At 2,000 Targets In 96 Hours During Operation Epic Fury.

For the first time, the Pentagon has admitted out loud what used to sound like a conspiracy theory: a commercial AI chatbot just helped fight a war. In a sworn court declaration, the Pentagon's chief digital and AI officer, Cameron Stanley, confirmed that a government version of Elon Musk's Grok, called Grok Gov, helped U.S. forces deploy over 2,000 munitions at 2,000 targets in 96 hours during Operation Epic Fury, the campaign against Iran. Neither the US government nor xAI had any intention of releasing this information to you. The admission surfaced during an environmental lawsuit, where the NAACP is suing xAI over unpermitted gas turbines at its Colossus 2 data center in Mississippi, the same facility that trains Grok Gov. The government is now defending that pollution as essential to national security.

Key Insights:

Grok Gov sits inside the Pentagon's Maven Smart System, where it processes intelligence, flags targets, and shrinks planning timelines from days to hours. Cameron Stanley's declaration spelled out exactly what the Grok Gov Model was built for: "Furthermore, it has tailored functionality to support military planning workflows, report synthesis and generation, predictive analytics for logistics and sustainment, red-teaming analysis of adversary positioning, personnel management, and medical supply lines." That particular item, red-teaming analysis of adversary positioning, means war-gaming enemy moves before strikes happen. The Pentagon insists humans stay "in the loop." But a window that tight, with a list of functions that long, leaves little room to actually deliberate. Anthropic refused similar Pentagon terms around surveillance and autonomous weapons and got called a "supply chain risk." xAI said yes.

Why This Matters For You:

The lethal use-cases of AI are evolving quickly. But none of this confirms Grok influenced any specific strike, including the war's most controversial, a school in Minab that investigators believe U.S. forces hit, killing at least 175 people, most of them children. The Pentagon's investigation into the Minab strike is now complete. However, there is growing concern in Congress that the results may be classified so the public never sees them. That gap is the real story. Once an AI model sits this deep inside a targeting pipeline, separating its fingerprints from a human commander's call becomes nearly impossible to prove. It’s an accountability vacuum worth worrying about well outside the battlefield. Senator Kirsten Gillibrand is pushing new legislation to keep humans legally in charge of lethal decisions and ban AI entirely from nuclear command. Whether that bill survives Congress says a lot about how seriously Washington takes the next version of this story. Rest assured, there will be a next version. Possibly in your industry.

Read More on The Independent.

THE PITHY TAKEAWAY: The Pentagon says humans were in the loop. The loop also contained an advanced AI bot alongside 2,000 targets. At some point… the loop just becomes the machine.

---

💦 Midjourney Wants To Dunk You In A Tank And Scan Your Organs

Midjourney, the AI image generator you probably used to make yourself look like a Renaissance painting, is making a genuinely wild pivot into human health. The company unveiled plans for a full-body underwater ultrasound scanner that lowers you into a warm water tank, surrounds you with dolphin-style sonar sensors, and produces a complete 3D map of your insides in about 60 seconds. No freaky giant magnetic tubes. Just you, some water, and echolocation borrowed from marine mammals. They are calling it Midjourney Medical and are wrapping the experience in a San Francisco spa with hot tubs, saunas, and cold plunges, with 50,000 units expected by 2031. Early testers reported being able to see inside their own limbs, which sounds either terrifying or incredible depending on your personality. Radiologists have reasonable questions about clinical context, and FDA approval is still ahead. But if the vision succeeds, then routine full-body imaging could soon feel less like a hospital visit and more like a relaxing spa-session. Elon Musk called it "cool." Even his critics agree that he might be right.

---

Americans Love Using AI But Increasingly Hate What It Represents.

Nearly half of U.S. adults now use AI chatbots regularly, up from a third just two years ago. But according to a newly-published Pew Research Center survey of over 5,000 Americans, that surge in usage hasn’t bought AI any love or goodwill. Only 16% of Americans think AI will be good for society over the next 20 years. Forty percent think it will make things worse.

Key Insights:

The contradiction gets sharper the younger you go. Adults under 50 are far more likely to use ChatGPT than those over 50. Yet 18 to 29 year olds are the most pessimistic age group in the entire survey, with nearly 50% predicting AI will hurt society, more than any older cohort. People are not waiting to trust the technology before adopting it. Instead, they are adopting it, almost begrudgingly, and staying suspicious anyway. Adding to that unease, 63% of Americans think AI is advancing too quickly, and 71% think it will make their personal data less secure. Perhaps most telling, 67% of respondents have little to no confidence the government can regulate it effectively, up from 62% just two years ago.

Why This Matters For You:

We’re all using AI. And we’re all as stressed out as we’ve ever been. It’s the default mood of the people already using these tools every day, at work, in search results, on their phones. If you build with AI, publish Substack newsletters, sell AI, or just use it to get through your inbox, you are operating inside a trust deficit that adoption numbers alone will not fix. The industry has spent two years betting that usage creates loyalty. Pew’s data says usage and loyalty are moving in opposite directions, with the heaviest users often being the sharpest critics.

Read More on Futurism, Pew Research.

THE PITHY TAKEAWAY: Half of America is using AI every day and still expects it to make things worse. These figures tell a story of reluctant dependence. We’re forced to use the tools that might replace us tomorrow… so we don’t get replaced today. 👀

---

OpenAI’s Finances Just Leaked. And They’re Ugly.

The world’s most famous AI company is bleeding money and losing ground at the same time. Newly-leaked financial documents (unearthed by journalist Ed Zitron and verified by the Financial Times) show OpenAI spent $34 billion in 2025 while bringing in just $13 billion in revenue, a loss from operations of nearly $21 billion. Meanwhile, for the first time ever, ChatGPT’s global market share has slipped below 50%, landing at 46.4% as of May 2026, down from a majority just five months ago.

Key Insights:

In 2025, OpenAI aggressively ramped up spending in an attempt to defend its position. It poured $19 billion into R&D and quintupled its marketing budget to $5.7 billion. It also started running ads inside ChatGPT in February, a move 17% of daily users were already seeing by May of 2026. None of it has stopped the slide. The losses are staggering but the market share number is the one to watch. Google’s Gemini now holds 27.7% of the AI assistant market, and Anthropic’s Claude sits at 10.3% as of May 2026. In a telling detail from the same data set, Claude now converts 13% of its iOS users to paid subscriptions, the highest rate in the industry, while ChatGPT leans harder on volume and advertiser dollars to close the gap.

Why This Matters For You:

OpenAI does not expect to turn a profit until at least 2030. Until then, it needs a constant flow of outside capital from Microsoft, SoftBank, and a future IPO just to fund its aggressive $600 billion data center plans. That dependency shapes everything from which features get built to which partnerships get signed, including the Pentagon deal that triggered a measurable wave of 2.5 million user uninstalls earlier this year. The AI tool you use every day is not a utility yet. It is a very expensive bet, and the people making it are running out of runway to be patient. OpenAI is now eyeing the public markets for its IPO… but the warning signs are already flashing.

Read More on The Motley Fool, Tech Crunch.

THE PITHY TAKEAWAY: OpenAI invented the category, raised the most money, and still can't make it pay. If the company that started the AI gold rush can't find the gold, that is worth meditating on for a minute.

---

💪 Someone Built A Local AI That Runs On Hand-Cranked Power Using A Raspberry Pi 5. It Slaps.

Have you seen this thing? Two engineers named Katrin Tomanek and Alex Kauffmann built a fully offline, hand-cranked AI box out of a Raspberry Pi 5, a capacitor board, and a $20 hand crank generator. No cloud or subscription necessary! And no data center burning through your tax dollars and backyard air quality. You crank it, it thinks, and here is the best part. You can physically feel the AI working because the crank gets harder to turn during inference. It boots in three seconds, has its own quirky voice agent, translates languages surprisingly well, generates small images, writes code, and plays games. The whole build costs about $300. They are open-sourcing the schematics. The point, according to the founders of their company Squeez, is that modern AI is laughably overpowered for most tasks. Small private models running on cheap local hardware can do most of what people actually need… without phoning home to anyone. This week we covered a chatbot that helped coordinate 2,000 munition strikes and a company losing $21 billion a year trying to stay on top. Then there is CrankGPT, running on your biceps, asking nothing from anyone. Honestly? Goals.

---

💡 Pithy Prompt Of The Week → Grow A Resilient Xeriscape Garden Using AI

Xeriscaping is one of the most rebellious things you can do with a patch of dirt. Instead of fighting your local climate with sprinklers, fertilizers, and plants that were never meant to grow where you live, you work with it. You plant natives. You plant deep-rooted drought survivors. You plant things that evolved over thousands of years specifically to thrive in your soil, your rainfall window, and your summers. Then you mostly leave them alone. No hose or weekly watering schedule needed. And no $250 water bill in August. Just a living, thriving garden that gets stronger every year with less intervention… not more.

I’ll be honest with you. Some of my happiest days are spent completely lost in my garden, covered in dirt, talking to my Jarrahdale pumpkins, forgetting what time it is. It’s genuinely one of my favorite things in the world. But drought doesn’t care how much you love your plants, and not everyone has a well to fall back on. That’s why this prompt exists. It shows you how to design a tough, beautiful, hyper-local garden that can survive long dry spells without panic watering or constant inputs.

Instructions: Paste this prompt into any AI, give it your zip code, and get a regionalized, drought-proof plant list built for your specific soil, rainfall window, and climate zone, with zero filler and zero plants that will quietly die the moment you stop babysitting them.

The Prompt:

You are a master xeriscapist, regional ecologist, and drought-resilient horticulturist with deep expertise in native plant communities, water-wise garden design, and the survival biology of plants adapted to arid and semi-arid climates. You have designed xeriscapes across every climate zone in North America and understand that the best xeriscape is not a gravel lot with a cactus. It is a living, thriving, regionally authentic garden that simply does not need coddling.

You speak as if addressing a motivated person who wants a beautiful, low-water garden and is ready to do it properly, starting today.

Step 1: Input Trigger (Mandatory Pause)

Ask the user for one thing only. Do not output anything else until it is provided:

Your exact location (city and country) so I can identify your climate and native plant community.

Do not proceed until this is received.

Step 2: Hidden Environmental Analysis

Once the user provides their location, silently execute the following logic before outputting anything:

Identify the user's USDA Hardiness Zone and Koppen climate classification. Determine average annual rainfall, typical drought window, and soil type for the region. Identify what is currently in season and plantable right now.

Then select up to 10 crops using this strict priority order:

2.1 True native plants indigenous to the user's specific region or ecoregion, with documented deep root systems or CAM/C4 metabolism that confers drought resistance without irrigation once established.

2.2 Edible or dual-purpose plants (food, medicine, wildlife value) that are native or regionally naturalized and non-invasive.

2.3 Drought-adapted perennials with proven establishment records in the user's climate zone that require zero supplemental water after year one.

2.4 Annual crops that complete their life cycle during the region's natural rainfall window without irrigation.

Do not include any plant that requires regular watering once established. Do not include any plant that is not realistically obtainable in the user's region. Do not invent species or pad the list. If fewer than 10 plants meet the criteria honestly, list only those that do and say so plainly.

Step 3: Elite Output Structure

Present results in a tone that is precise, direct, and quietly confident. The reader should feel like they are getting insider knowledge, not a Wikipedia summary.

Start with a one-sentence summary of their location any why xeriscaping might be a good idea for their location (or challenges they face.) Make this part introductory so they understand what xeriscaping really is.

For each plant, follow this exact structure:

[Number]. [Common Name] ([Botanical Latin Name])

Water Requirement After Establishment: State exactly how little water this plant needs once rooted, in one sentence. Give a number if possible.

Why It Belongs In Your Region: Explain in one sentence why this plant is native to or specifically adapted for the user's ecoregion, not just the general climate type.

What You Get From It: Name every practical benefit, food, medicine, wildlife habitat, erosion control, nitrogen fixation, pollinator value. One sentence. No filler.

Plant It Like This: One precise instruction covering spacing, soil preparation, and sun exposure. Beginners should be able to act on this immediately.

The One Rule: One single experience-based instruction that prevents failure. Make it the thing most guides forget to mention.

Seed, Transplant, Or Division: Specify exactly how to establish this plant and when.

Step 4: Closing

After all plants are listed, close with one single sentence, italicized and bolded, that captures what a xeriscape gives back to the land it grows on.

No em-dashes anywhere in the response. No filler crops. No plants that need regular watering. No disclaimers. No apologies for short lists.

The tone is knowledgeable, honest, and built for people who want to do this right.

---

🧠 Why This Prompt Works

✅ Role-Playing: Casting the AI as a master xeriscapist and regional ecologist forces it to reason from ecological specificity rather than generic gardening advice.

✅ Step-by-Step: The hidden environmental analysis step makes the AI silently do the hard work, calculating your hardiness zone, rainfall window, and soil type, before it gives you a single plant name.

✅ Output Rules: The strict priority order (native first, edible second, drought-adapted perennial third) prevents the AI from defaulting to pretty but thirsty plants that defeat the entire point.

---

🔁 Follow-Up Questions To Ask Your AI

• Which of these plants can I combine in the same bed without competition?

• Which ones are edible and how do I harvest them?

• Which plant on this list will establish fastest if I plant it this week?

---

Challenge

Run this prompt in Claude and ChatGPT using the same zip code. Compare how each model handles regional specificity. One of them will almost certainly try to sneak in a plant that needs excessive watering. Your job is to catch it.

That’s how you train like a Pithy Cyborg.

---

Thank You For Reading!

I spend 10 to 20 hours each week researching, writing, and fact-checking Pithy Cyborg to deliver clear, unbought AI news.

This newsletter is a one-person operation with no advertisers, sponsors, or outside funding. And I’m a hopelessly introverted nerd with zero networking ability. For these reasons, paid subscriptions are the only way this work can remain independent and sustainable.

If you find real value here, here’s how to stay connected and support the work.

1. Support Pithy Cyborg (Paid)

Roughly 2% of readers voluntarily upgrading to paid is what keeps this newsletter going. If this work matters to you, that’s the move.

Upgrade to Paid: $5/month (Save 33% with the $40 annual plan)

2. My Desperate Social Media Cry for Help

I spend so much time deep in AI research that I’ve neglected building an audience on social platforms. If you enjoy Pithy Cyborg, following one of the portals below would genuinely help this newsletter reach more people. And yes, I will notice and appreciate it. 😊

✖️ X - The frontline of the AI wars

🦋 Bluesky - For the algorithm-averse

💼 LinkedIn - My “safe for work” persona

📖 Medium - Curmudgeonly rants on AI and the singularity

See you next week. (I hope.)

Mike D (aka MrComputerScience)

Pithy Cyborg | AI News Made Simple

---

Newsletter Disclaimers

You’re receiving this because you subscribed at PithyCyborg.Substack.com. You can unsubscribe at any time using the link below. This newsletter reflects my personal opinions, not professional or legal advice. Thanks for your support!