deepseek – Ridley Park

Nitpicking DeepSeek

August 4, 2025Posted in Artificial Intelligence, Book, ChatGPT, Claude, critique, Generative AI, Ridley Park, Software, WritingTagged addiction narrative, ai, AI hallucination, AI-assisted editing, Artificial Intelligence, author editing process, books, ChatGPT, ChatGPT editing, Claude, cognition, deepseek, DeepSeek review, editing, excerpt, fiction, fragmented prose, Gemini, Generative AI, Grok, language, literary fiction, Meta, Needle’s Edge novel, novel, perception, raw storytelling, Sarah Stacey character, stream-of-consciousness narrative, technology, Thinking, trauma fiction, unconventional narrative style, visceral writing style, Writing, writing with AILeave a comment

How do you use AI to assist your writing or editing process? As I’ve shared before, I use it extensively for research, where previously, I would have used Google, a library, or personal books or journals. I use them for proofing, editing, and alpha and beta reading. Today, I’m editing.

I’m still editing the manuscript for Needle’s Edge. When I finish a scene or a chapter, I run it through several AI platforms – the usual suspects: ChatGPT, Claude, Grok, Meta, DeepSeek and Gemini – with a simple prompt:

how does this scene read?

Usually, I copy-paste the content, but this manuscript also relies on presentment, so I share a screen capture instead. Rather than share the pages, I share the line I am discussing here:

Image: Excerpted line from Needle’s Edge

From this, DeepSeek shared this as part of its “thinking” process.

The writing style is very visceral and fragmented – short sentences, intentional misspellings (“c-noses” for cash, “waigh” for might), and stream-of-consciousness elements. This creates an immersive, uncomfortable reading experience that mirrors Sarah’s physical and emotional state.

Interestingly, there were no intentional misspellings. DeepSeek’s OCR failed and mistook ‘c-notes’ for ‘c-noses’, and then tried to interpret the results. I couldn’t determine where it got ‘waigh’. There is a word ‘might’ per its comment, but I don’t see how it confuses that. No matter.

Still, its assessment feels close enough, and a reminder that grammar and syntax are less important than one might (waight?) believe.

Overall Impression:

The scene is powerfully distressing. It doesn’t glamorize or moralize; it simply shows the brutal, day-to-day reality of addiction and survival on the margins. It’s effective in conveying:

Physical Horrors: The relentless pain, discomfort, and bodily damage.

Psychological Toll: The desperation, numbness, fleeting panic, and dangerous denial.

Social Environment: The hostility, degradation, danger, and shared, traumatic knowledge of the street life.

Hopelessness: The cycle of seeking relief only to encounter worse consequences, underscored by the memory of others’ fates and the resigned “never again” that likely won’t hold.

It’s a difficult read due to its unflinching portrayal of suffering and decay, but it’s compelling in its raw authenticity and visceral impact. The style perfectly matches the content, immersing the reader in Sarah’s painful, precarious world.

AutoCrit Challenges

March 11, 2025Posted in Artificial Intelligence, AutoCrit, ChatGPT, Claude, critique, Generative AI, Short Story, Software, WritingTagged ai, AutoCrit, book review, book-reviews, books, characters, ChatGPT, ChatGPT 4, Claude, closure, consistency, critique, deepseek, editing, emotional quotient, emotions, eq, fiction, Generative AI, human condition, narratives, real life, reality, resolution, stories, variability, WritingLeave a comment

I don’t hide the fact that I rely on AI for early editorial feedback. Once a story is complete, I break out AutoCrit. This programme works well for typical stories that follow standard practices with common tropes. It gets quite confused when I feed it intentionally awkward stories, not the least of which is to advise me to eliminate the awkwardness.

This is a challenge with AI more generally. In this particular story, I leave a lot of loose ends and misdirects, as it’s a commentary on the conspiracy-driven culture we inhabit. The advice, is along the lines of, “You forget to close this lopp. What happened to so and so.”

But this is life. We don’t always know the full story. We drive past an multi-car accident where cares are overturned and in flames, but we never find out what happens – even if we scour the newspapers and internet. Who was that? What happened? What caused it?

We often never find out. In most books and movies, we find out everythung, and it all comes packaged with a nice bow. This is what AI expects. It’s the diet it’s been fed.

Some stories subvert these notions here and there, but by and large, this is not typical American fare. Readers and viewers need to be spoonfed without inconsistencies.

Speaking of inconsistencies addressing one scene, AutoCrit said that a character should act impulsively in one situation and reserved moments later. This was flagged as an iinconsistent character.

In the scene, a woman stops her car immediately to help an injured man on the roadside, but as she gets out of her car an approaches her, she shows caution.

This was a red flag. Why would she have always been rash or always been cautious?

My response, because that how real people act. She acts on instinct but quickly considers that she’s a vulnerable woman alone with a man miles from anywhere.

I don’t suspect a human reader would find this surprising. This is the intelligence absent from Artificial Intelligence — cultural intelligence, a cousin of EQ, emotional quotient.

I know how I want the character to act. I do want AutoCrit to inform me that character A is wielding a pistol but then stabs another character, or that character B is a teetotaler and is getting drunk or that character C has a shellfish allergy but is downing lobsters like they’re going out of style. And I certainly what to be shown continuity errors.

The biggest challenge I have with AutoCrit that is less promonent with other AIs is that I can preface my content with a note explaining my intent. I can even do this after the fact.

If I feed ChatGPT, Claude, or DeepSeek a story of segment to critique without a preface, the responses may be similar to AutoCrit, but when I follow up with some meta, the response may be, “Now it makes sense, but why is John wearing lipstick?” Perhaps he’s metrosexual or non-traditional. Perhaps it’s an oversight.

I dont meán to demean AutoCrit. I’m just advising that if you are writing stories not compliant with 80 per cent of published works, take the advice with a grain of salt, or reserve AutoCrit for more standard fare.