The AI Instruction Insider July 2025

AI that teaches itself: MIT's SEAL Framework

What is the AI news?

MIT researchers have introduced SEAL (Self-Adapting Language Models), a new AI framework that learns by teaching itself. Instead of relying on external training data or instructions, SEAL generates its own study material and figures out how best to learn from it. In tests, SEAL outperformed even models trained with data generated by GPT-4 in some tasks, jumping from 0% to 72.5% accuracy in puzzle-solving challenges.

Why it matters

This is more than a technical curiosity - it's a glimpse into the future of adaptive AI. For technical writers, SEAL signals a shift toward AI systems that can evolve and specialise without human intervention. Imagine documentation tools that automatically update based on product changes or writing assistants that refine themselves based on your personal tone and structure over time.

How it helps you

Here’s the exciting part: SEAL-like models could soon power writing assistants that actually learn from you. Instead of a one-size-fits-all AI, future tools might adapt to your writing style, your preferred document structure, and even the kinds of questions your audience tends to ask. This means fewer generic suggestions and more context-aware, personalised help.

If you’re working on complex or evolving products, SEAL offers a vision where your documentation tool could track product updates and intelligently adjust your drafts or suggest revisions based on real-time changes. Think fewer hours spent chasing changes and more time focused on clarity and strategy.

Of course, there are caveats. SEAL still struggles with "catastrophic forgetting" (losing old knowledge when learning new things) and requires hefty computing power. But as this tech matures, it could fundamentally reshape how we approach writing and maintaining documentation.

In short, SEAL is an early but powerful step toward self-improving AI - a development that could make our tools smarter, our workflow smoother, and our writing sharper.

ClickHelp adds “Ask your docs” AI: semantic search for tech writers

What is the AI news?

ClickHelp, a popular documentation platform, released a powerful new feature in June: “Ask Your Docs” (Beta). This tool lets users ask natural-language questions and get accurate answers from across all project documentation - including drafts. Backed by their AI Suite and new “CAPs” (ClickHelp AI Prompts), it turns your documentation into a conversational knowledge base.

Why it matters

Documentation grows fast - and so does the challenge of keeping track. This update gives technical writers and their teams an edge by allowing quick Q&A across large doc sets. Here’s why it’s a game-changer:

• Fewer repeated questions: Colleagues can query the docs directly instead of asking you for details.
• Faster onboarding: New team members can learn by asking the system, not digging through folders.
• Live draft access: Even unpublished or in-progress content can be queried, improving collaboration across roles.
  
  

How it helps you

This feature transforms how you manage and use documentation:

• Streamlined maintenance - easily spot outdated sections when AI can't confidently answer a question.
• Better FAQs and troubleshooting - Use real queries to shape support content.
• Time savings - Instead of writing summaries or overviews manually, let the AI do the heavy lifting and refine as needed.

If you're already using ClickHelp, “Ask Your Docs” is worth enabling. For those exploring tools that combine authoring with AI search, this makes ClickHelp a serious contender in 2025.

Overleaf adds AI assist for LaTeX: smart suggestions inside your docs

What is the AI news?

On June 24, 2025, Overleaf launched AI Assist, a new built-in tool that provides grammar feedback, rewording suggestions, and LaTeX-aware code improvements - all directly within your Overleaf editor. Writing research papers, documentation, or reports just got easier - now you can polish both language and code syntax in one place.

Why it matters

Many technical writers, especially those in academia, engineering, or science, rely on LaTeX for precise document formatting. Until now, you had to bounce between tools - one for writing help, another for LaTeX code. With AI Assist:

• You stay in the flow - No need to copy text out to get grammar feedback.
• Fewer code errors - The AI understands LaTeX, so it can help correct or improve syntax.
• More polished writing - It offers tone and clarity suggestions, even for complex or formal topics.

How it helps you

If you write in LaTeX, this update streamlines your workflow:

• Faster reviews - Fix typos, clarify wording, and debug code right where you write.
• Consistent tone - Maintain the appropriate level of formality and technical accuracy.
• Integrated editing - No back-and-forth between Grammarly, ChatGPT, and your document editor.

Prepping a user manual, white paper, or technical spec is now faster - AI Assist cuts down on context-switching and streamlines your workflow. If you haven’t used Overleaf in a while, this is a great time to jump back in with built-in AI handling the heavy lifting.

Google NotebookLM gets smarter sharing: AI that knows your docs

What is the AI news?

In June 2025, Google added public notebook sharing and admin tools to NotebookLM, its AI-powered research assistant. Now you can collaborate using shared “notebooks” that respond to questions based on your uploaded content - like PDFs, specs, guides, and research files. New admin controls also help organizations manage access, ensuring NotebookLM fits better into professional workflows.

Why it matters

NotebookLM has been one of the most promising tools for technical writers who juggle multiple source documents. These new features make it more team-friendly and enterprise-ready:

• Shareable AI context - You can invite teammates to ask questions about your doc sets, boosting collaboration.
• Better version control - Everyone works off the same up-to-date AI model trained on your shared content.
• Admin oversight - Controls over content access mean it’s safer for sensitive materials.

How it helps you

This makes NotebookLM a true “AI knowledge base” for tech writing teams:

• Speed up ramp-up - New writers can get answers from past specs or release notes without pinging senior staff.
• Generate content faster - Ask NotebookLM for summary bullets, feature overviews, or change logs.
• Stay organised - Keep source materials, drafts, and output linked in one searchable workspace.

If you document complex products with lots of moving parts - think APIs, hardware, compliance - you’ll find NotebookLM’s new sharing tools a huge productivity win. Try creating a shared notebook around your next project kickoff.

Apple Intelligence adds writing tools: On-device AI for clean, clear docs

What is the AI news?

In June 2025, Apple rolled out new Writing Tools as part of its Apple Intelligence launch for iOS 18 and macOS Tahoe. These features let you edit tone, proofread, summarise, and rewrite text using natural-language prompts - directly within apps like Mail, Notes, and Pages. The best part? It all runs on-device, preserving privacy and responsiveness.

Why it matters

Not every writing tool fits into enterprise workflows, but Apple’s new on-device AI brings assistive features to tools writers already use daily. Here’s why it’s relevant:

• Privacy-first - Everything runs locally, making it safer for internal or confidential drafts.
• Integrated editing - No need to switch tools; just select text and ask for clarity, tone adjustment, or a summary.
• System-wide support - These tools appear across Apple’s apps, from documentation drafts to status reports.

How it helps you

Drafting documentation, fine-tuning technical details, or responding to team feedback - all become easier:

• Rewrite in context - Quickly rephrase sections for tone (“Make it more formal” or “Simplify this paragraph”).
• Summarise long notes or threads - Turn meeting recaps or product updates into short, scannable bullet points.
• Build repeatable workflows - Combine Writing Tools with Shortcuts to automate editing tasks.

For Mac or iPad users working in Apple’s ecosystem, this is a huge boost. It brings smart, context-aware editing into everyday tools - without the privacy trade-offs of cloud-based models. Give it a try the next time you're polishing a draft in Pages or summarizing a stakeholder email.

Gemini app gains video and automation features - now more useful for content-heavy teams

What is the AI news?

Google’s Gemini app received two major enhancements in May: Veo 3, its AI-powered video generator, and Scheduled Actions, which lets users automate repetitive tasks like summaries or email digests. Veo 3 can generate coherent videos with synced audio and text input, while Scheduled Actions uses context from your workspace to handle recurring tasks automatically.

Why it matters

These additions make Gemini more than just a text assistant. For technical writers, this means the ability to communicate complex information visually and reduce time spent on routine documentation updates:

• You can generate short training videos directly from scripts.
• Routine communications like status updates or release notes can now run on autopilot.
• Writers supporting product onboarding or help centres gain new tools for user-facing content.

How it helps you

• Convert text-based guides or walkthroughs into simple explainer videos with Veo 3.
• Schedule Gemini to summarise bug reports or changelog entries every week.
• Use video as a companion to text documentation, especially for complex UI workflows.

The Gemini app is evolving into a hybrid content engine - if your team leans on video or automation, it’s time to explore these features.

OpenAI’s Responses API adds automation-friendly upgrades

What is the AI news?

OpenAI quietly updated its Responses API in May, introducing more structured endpoints, output formatting controls, and better integration support for developers building content pipelines or tools. This opens the door for deeper AI use in automated documentation workflows.

Why it matters

APIs are the backbone of custom documentation systems. With these enhancements, AI-generated content can now be fetched, transformed, and integrated into internal CMSs or workflow tools more efficiently.

How it helps you

• Automatically generate draft articles, feature descriptions, or changelogs from developer notes.
• Pipe AI-generated outputs directly into Confluence, GitHub Wikis, or other content hubs.
• Trigger content generation based on events - for example, a new feature branch triggering an updated product summary.

This makes OpenAI’s platform more than a chat interface - it becomes a content generation engine that can plug into existing technical writing systems.

Document360 adds interactive elements and formatting upgrades

What is the AI news?

In May, Document360 released several improvements aimed at enhancing the usability of knowledge bases. These include interactive decision trees for user-guided help paths, an updated search interface, improved list formatting, and a new format-painter tool for styling consistency.

Why it matters

Technical documentation is often dense and navigationally complex. These updates provide better ways to guide readers, making support content more intuitive. For writers, better formatting tools reduce friction when maintaining large-scale documentation sets.

How it helps you

• Build interactive help flows with decision trees, ideal for troubleshooting content.
• Ensure consistent styles across long pages or multiple writers using the format-painter.
• Improve content discoverability through smarter internal search and cleaner list structures.

If your knowledge base content often supports product support or IT self-service, these tools can directly improve user experience and content quality.

Thank you to all our readers for following The AI Instruction Insider. Your continued interest and feedback help shape every edition.

We’re taking a short summer break – see you in September!