ChatGPT: Hobbies, Fitness & Learning

Knit doll socks from a ChatGPT pattern, then find one that works

Knitter Ciara of The Fairythorn asked ChatGPT to write a sock pattern for an 18-inch doll. Ciara managed to knit a sock from it, but only by turning the heel from experience, as that section had a fundamental error, and the sock came out far too small for the doll to ever wear. Asked instead to point to a free pattern on Ravelry, ChatGPT offered three suggestions in under a second. The first, 'Basic Doll Socks' by Jacquelyn Fitzgerald, knitted up quickly and easily and fit the doll, with just one small error.

Write a life story book with aphasia after a stroke, using ChatGPT to edit

A 75-year-old patient with aphasia from a stroke eight years earlier had trouble finding words and forming sentences, and wanted to write an autobiography. In speech therapy sessions, the patient wrote their own content and used ChatGPT to edit it, guided by a visual flow chart and set prompts. With ChatGPT, the patient produced more sentences with fewer errors. In three out of 10 sessions the patient noted ChatGPT sometimes wrote long, complicated sentences, and preferred the 'edit the following' prompt. It took clinician prompting and modeling at first. By semester's end there was enough for a printed life story book, shared with friends and family.

Write a crochet pattern from a photo with ChatGPT and Claude (gave up)

Melissa Warr wanted to crochet a Valentine's dog from an image found online, but couldn't find a pattern for it and had never written one. Basic prompts for a pattern from a picture didn't work, so Warr had ChatGPT help write a better prompt. The patterns kept coming out as numbered lists instead of bullet points, with arms and legs that weren't long enough. Warr ran the prompt in both ChatGPT and Claude; Claude's results were better but 'missed many of the parts (the tail, ears, etc.)', and both struggled with irregular shapes. Warr gave up on the AI pattern and bought a pattern on Etsy instead.

Research a Victorian-born grandfather's life with Claude, ChatGPT and Gemini deep research (partial)

Dan Taylor-Watt gave Claude, ChatGPT and Gemini the same brief, with deep research turned on, to research a paternal grandfather, Cecil, who died in 1960, starting from a Word document written by Taylor-Watt's father. Claude took about 6 minutes, used 340 sources and wrote an 11-page, 4,219-word report; it found no significant new information but flagged inaccurate dates, and it later made a 3-minute-30-second narrated video using code and an open-source voice model. ChatGPT took 9 minutes, ran 714 searches and wrote 14 pages (4,659 words); it caught an error about the year of the fire that destroyed service records, noted that the LNER railway wasn't created until 1923, found information on the first wife's family, and spotted a Lance-Corporal's chevron in a photo, but its video attempt fell short. Gemini wrote the longest and most visual report (5,347 words), but its maps were wrong (Osmotherley is east, not south, of Northallerton), it missed the date problems, and it said it couldn't make a video.

Train for a first half marathon on a day-by-day plan built by ChatGPT

With no coach and a few 5Ks and 10Ks behind them, the runner behind Everyday Ai + Autonomy gave ChatGPT their weekly mileage, usual distance, pace and the days they could run. It built a day-by-day, week-by-week plan with tempo runs and long runs. The runner printed it, read it through, wrote every run on a calendar and ticked each off, then ran a first Berkeley Half Marathon on that plan. About a year later they ran Berkeley again and set a personal record.