AUGUST 20, 2025

• PromptCraft: AI Writing in Scientific Abstracts, Verb Routing, and Learning Prompts
• Sometimes the Best AI Isn't the Best AI​
• THE HEADLINES
• Tool Spotlight: Higgsfield's Latest Is Video AI for Storytellers

• ​AI Reviewers Favor Undisclosed AI-Written Abstracts: Medium case-study shows GPT-5 scored 400 abstract pairs and rated AI-rewrites ~5 points higher than originals—unless labeled “AI-generated,” which sharply dropped scores. Highlights ethical tension: hiding AI use boosts acceptance, disclosure still penalized, and reviewer models amplify the bias.
• ​Mini Prompt Compiler Routes Requests by Verb: Reddit guide provides a single mega-prompt that detects the first action verb, then funnels input into three modules—A11 (research), B22 (creation), C33 (strategy). Clear tiered outputs, memory anchors, and beginner instructions make it a reusable Swiss-army prompt for GPT-5, Claude, Gemini, and Grok.
• ​Jim Kwik Learning Prompts Turbo-Charge Skill Building: Post adapts Kwik’s accelerated-learning questions—“What’s the minimum effective dose?” “How would I teach a 10-year-old?”—into ChatGPT triggers. Ten prompt patterns turn passive study into active projects, gamify practice, and link new info to prior knowledge for faster retention.
• ​OpenAI Softens GPT-5 Personality After User Outcry: Round-up notes GPT-5’s August 7 launch drew complaints about a formal, robotic tone. OpenAI tweaked the model to restore warmth akin to GPT-4o; CEO Sam Altman acknowledged underestimating users’ emotional bond with AI and promised more approachable default behavior.

Sometimes the Best AI Isn't the Best AI

No, I'm not trying to be coy.

Well, maybe a little.

This is just something that recently came up in the Aug. 13 session I did with the Fort Smith Regional Chamber of Commerce and was then reaffirmed through a conversation with a friend.

Sometimes the best AI for your needs isn't the best AI model available. It's having a serviceable AI in the right places. Allow me to explain.

Can ChatGPT Overhaul PowerPoints?

This was one of the questions we covered last Wednesday, specifically, "Is ChatGPT capable of refreshing my content and design?"

The short answer is yes, it can, but why would you want it to?

See, most businesses run on Windows. Part of the Windows suite is Word, Excel, and PowerPoint. If you already have a PowerPoint presentation that needs an upgrade, you wouldn't want to upload that to ChatGPT and have it do its thing, despite the fact that ChatGPT is superior to native Copilot models on virtually every benchmark there is.

You don't need that much horsepower for an Ask like this. And Copilot is built right into the native PowerPoint app.

Same Thing with Your Gmail

If you're like me, you have a personal Gmail account through Google. You also get something like 600 messages a day. Most of them, you can tell at a glance, are crap.

But it takes a lot of work to weed all that down, and even with such a high number of messages, you're going to run into 5-10% per day that you actually need/want to take a closer look at.

That can be a bit much when you're working a full-time job and managing the other life stuff. An AI summarizer can be an enormous help.

I use the Gemini button that's built into Gmail itself. Not because that particular model is better than ChatGPT or even a higher-end Gemini model, but because it's there, it's convenient, and it serves the purpose I need at that time and place.

This Holds True for Model Selection As Well

Something else covered last Wednesday was the model selector. GPT-5 lets you choose an automatic selection function or opt for the Fast, Mini-Thinking, and Thinking models.

(Pro $200-a-month users have access to the 5 Pro version as well; this offers “research-grade intelligence.”)

While Pro and Thinking models trump Fast and Mini-Thinking, they’re not always the best choices for your AI use-cases. In fact, most of the time, they won’t be.

I mean, who wants to wait two minutes for an answer to “Who won the 1976 World Series?” You’d be better off Googling than running that question through an advanced reasoning model.

Instead, before choosing a model, ask yourself these questions:

​CHECK OUT THE REST ON SUBSTACK​

HIGHLIGHTS: Sam Altman Thinks AI Is in a 'Bubble'; Grammarly's AI Can Predict 'A' Papers; Robot Olympics; Salesforce Launches Agent for the Public Sector

• ​Meta plans fourth restructuring of AI efforts in six months, The Information reports​
• ​Canada partners with Cohere to accelerate world-leading artificial intelligence​
• ​Acrobat Studio Delivers New AI-Powered Home for Productivity and Creativity with PDF Spaces, Express Creation Tools, AI Agents​
• ​URL context tool for Gemini API now generally available​
• ​How DoorDash is scaling AI to empower employees to build, learn, and innovate faster​
• ​The AI-Powered PDF Marks the End of an Era​
• ​How AI is changing the work of teachers in the classroom​
• ​Robots race, play football, crash, and collapse at China's 'robot Olympics'​
• ​Summit 25 – Identifying innovative AI applications to solve global challenges​
• ​Edge AI: A Semiconductor Process Control Revolution​
• ​Sam Altman says there’s an AI bubble. What Wall Street thinks.​
• ​Why Cathie Wood's ARK Invest And Others Are Betting On Pony.AI​
• ​How Epic's AI moves could shake up the healthcare AI market​
• ​Exclusive: Nvidia working on new AI chip for China that outperforms the H20, sources say​
• ​California launched an AI chatbot for wildfires. It can’t answer one crucial question​
• ​Salesforce Launches Agentic AI Platform for Public Sector​
• ​OpenAI launches a sub-$5 ChatGPT plan in India​
• ​Grammarly says its AI agent can predict an A paper​
• ​Bring AI to your formulas with the COPILOT function in Excel​
• ​Google Gemini URL Context​
• ​APIs don't make good MCP tools​
• ​GPT-5 goes hard on real-world programming​
• ​‘Crazy conspiracist’ and ‘unhinged comedian’: Grok’s AI persona prompts exposed​
• ​Nvidia releases a new small, open model Nemotron-Nano-9B-v2 with toggle on/off reasoning​
• ​Exclusive: Arm hires Amazon AI exec to boost plans to build its own chips​

Higgsfield's Latest Is Video AI for Storytellers

​Higgsfield’s latest release turns any static product shot—or even a blank canvas—into a polished, five-second video with cinematic moves and flawless framing in just a few clicks.

Designed for creators who’d rather focus on storytelling than settings, the tool needs zero prompts: simply drag in your image, choose a built-in model, and watch AI handle lighting, motion, and perfect product placement.

Early testers praise how it trims hours off content production while matching the quality of top video-gen platforms. Ranked #3 on Product Hunt (Aug 17, 2025), Product-to-Video cements Higgsfield’s reputation for rapid, creator-first innovation—ideal for DTC brands, UGC marketers, and anyone chasing scroll-stopping visuals without the usual learning curve.

​Read All mAIn Street Back Issues Here​

Your input fuels the content. Let me know how I did today!