From Portable Spell-Checkers to LLMs: How AI Finally Lets Me Write

Rebuilding the Heart of My Classic: 4.2L V8 Engine Restoration

From Chat to Star Wars: A Simple POC Integrating GraphQL with LangGraph

Welcome to Roving Code: Two Distinct Drives

I still remember the first time I clutched one of those portable spell-checkers. It felt like a small miracle — a cheap, plastic companion that rescued me from the most embarrassing typos. Later, when word processors became ubiquitous, I felt hopeful all over again: red squiggles, instant backspacing, copy-and-paste salvation. Those tools mattered. They saved time, blunted some frustration, and made typed work legible.

But there was a limit. Spell-checkers and conventional grammar tools are, by design, literal: they match patterns and suggest replacements. If I typed a correctly intended but phonetically odd word, or if my sentence rhythm was off because I’d mentally jumped two steps ahead, the tools often missed the point of what I *meant*. They corrected surface errors but not the deeper structural problems that dyslexia can cause awkward phrasing, uneven flow, and the occasional sentence that simply didn’t capture the idea I had in my head. Worse still they oftern 'corrected' my words to something completely different.

That’s where large language models (LLMs) have felt different for me. They don’t just flag a misspelling; they help me reformulate, reorganise, and research. Instead of wrestling alone with a paragraph, I can give the model a few notes or bullet points and ask it to draft a few versions. I then pick the one that sounds most like me, tweak it, and the result reads as my voice just clearer. That’s not outsourcing my ideas; it’s giving them a translator that understands both my intent and the conventions of good writing.

I would frequently write using simpler words then my thoughts after getting frustrated with spell checkers correcting my phonetic spellings to words that were not what I intended, and failing to help get me to the words I meant.

## Why older tools often failed me

- **Literal corrections**: Spell-checkers would suggest the nearest dictionary word, not the sense I wanted to convey. If my phonetic spelling looked like a different valid word, the suggestion often veered me off-course.
- **Poor structural help**: Grammar checkers rarely offered meaningful rewrites for poor sentence structure; they flagged but didn’t reconstruct.
- **Research overload**: When writing technically, I’d spend hours trawling forums, papers and manuals. The sheer volume of text often poorly organised could switch my brain into overload and shut the writing down.

Those pain points are pretty well documented in accessibility and human-computer interaction research: tools that operate at the surface level help, but don’t fully address the cognitive differences in planning, organising and producing text for many neurodivergent writers. Recent prototypes and studies explicitly targeting dyslexic adults show promising directions beyond simple spell-checking.

## What LLMs do differently (and why it helps me)

From my experience — and from emerging studies and prototypes LLMs help in three concrete ways:

1. **Intent capture and paraphrasing**  
   I can type a rough, messy sentence and ask the model to paraphrase it in different tones. The model’s ability to infer intent from fragments and produce fluent alternatives saves the “rewrite until readable” slog. Research prototypes designed for dyslexic adults show this capability is useful for real email and short-form writing tasks.

2. **Structure and scaffolding**  
   When my thoughts are out of order, I tell the model roughly what I want to say and ask for a clear outline or a few structured paragraphs. That scaffolding mirrors how a supportive editor might help only it’s instantaneous. Qualitative studies of neurodivergent communities show people using LLMs for summarisation, rephrasing and step-by-step instruction building.

3. **Focused research and summarisation**  
   Instead of wading through dozens of pages, I can ask an LLM to pull together key points from studies or explain a concept in plain English. There’s increasing academic work showing LLM based workflows can help students with dyslexia engage with learning materials and produce better written outputs when combined with guided prompting.

## What the research says (a quick, referenced tour)

Some deeper research, here are some of the research papers I found which might be of interest:

- **Goodman, S. M., Buehler, E., & Clary, P. (2024).** *LaMPost: Supporting Dyslexic Adults with Everyday Writing Tasks Using Large Language Models*. ACM. [https://dl.acm.org/doi/10.1145/3626952](https://dl.acm.org/doi/10.1145/3626952)  
  *Summary:* Introduces **LaMPost**, an LLM-enabled email writing prototype for adults with dyslexia; a study of 19 participants showed features like “rewrite” and outline support reduced writing friction.

- **Carik, B., Ping, K., Ding, X., & Rho, E. H. (2024).** *Exploring Large Language Models Through a Neurodivergent Lens: Use, Challenges, Community-Driven Workarounds, and Concerns*. ACM. [https://dl.acm.org/doi/10.1145/3701194](https://dl.acm.org/doi/10.1145/3701194)  
  *Summary:* Analyses 61 neurodivergent community discussions, categorising 20 use cases for LLMs (from emotional support to productivity), while highlighting challenges like overly “neurotypical” responses and reliance risks.

- **Fung, K. Y., Fung, K. C., Lee, L. H., Lui, R. T. L., Qu, H., Song, S., & Sin, K. F. (2025).** *A Study on Using ChatGPT to Help Students With Dyslexia Learn Chinese and English Writing*. SAGE Journals. [https://doi.org/10.1177/20965311251358269](https://doi.org/10.1177/20965311251358269)  
  *Summary:* Finds that a ChatGPT-based system (CHATTING) improved behavioural and emotional engagement, as well as motivation, in dyslexic students, though gains in cognitive engagement were more modest.

Those papers don’t claim LLMs are a complete solution, rather they show LLMs as a *new* class of assistive technology that can address higher-level cognitive and rhetorical difficulties that earlier tools could not.

## Important caveats (because nuance matters)

I’m enthusiastic, but not naive. The literature and ongoing coverage highlight real concerns:

- **Bias and hallucination**: LLMs can generate inaccurate statements confidently. For research or clinical claims, always verify original sources. It's important to set guardrails and create a verification process.
- **Privacy and data handling**: Feeding drafts with sensitive personal information to cloud services has implications. Choose tools with clear privacy policies if that’s a concern.
- **Not a substitute for professional support**: LLMs help with expression and research, but they don’t replace specialist teaching, therapy or reasonable workplace adjustments when those are needed.

## How I actually use LLMs in my workflow

A practical run-through of what works for me:

1. **Brain-dump**: I open a short doc and write rough bullet points or a couple of messy sentences — mistakes and all.  
2. **Prompt for options**: I tell the LLM: “Turn these notes into three short paragraph versions: casual, technical, and concise.”  
3. **Pick and personalise**: I take the version that best matches my voice, edit a little (I still like to tinker), and sometimes I'll add a bit more detail.
4. **Verify any facts**: Review the article, check the facts stated, verify the references and citations.
5. **Publish as me**: The final text is my idea, shaped and polished by the model.

This combination keeps my voice authentic while removing the mechanical barrier of translating thought → text.

## Final thoughts

If you grew up with spell-checkers and word processors, you’ll understand the small victories those tools brought. LLMs aren’t a replacement for traditional assistive tech; they’re a new layer, a partner that helps with the *meaning* behind the words, not just the surface errors.

The academic evidence is nascent but growing: prototypes and studies indicate real benefits for dyslexic and other neurodivergent writers, while reviews urge caution about the risks. For me personally, LLMs have unlocked an ability to share ideas that used to get stuck between thought and sentence. That’s not magic, it’s an assistive technology that finally helps me to express my ideas, no longer held back by the frustration of not being able to convey my thoughts.