Halloween Drop: Nostalgia Hooks, AI Deskilling Confessions & the 'Lost Thing' Format

What creators are doing

TikTok77–195 HN points

The 'lost thing resurfaces' hook. Creators frame a forgotten or buried piece of culture — a lost novel, an OS nobody runs, a 14-year-old audio clip — as something that was just rediscovered. The format opens with 'Nobody talks about this' or 'This disappeared and I need to know why', then delivers a tight 60-second explainer. Halloween makes the 'buried and risen' framing feel native today.

Why it works: Curiosity gap plus low prior awareness means viewers watch to the end to find out if they missed something real — completion rate drives distribution.

Copy this

Open with: 'A Jules Verne novel sat hidden for [X] years — here's what's in it and why they buried it.'

Make this today

Make the 'Lost Thing' TikTok or Reel in Under 60 Minutes

1. 1

   Pick your lost thing5 min

   Choose one item from today's pack: Jules Verne's lost novel, BeOS running in a browser, or the NESformer (a working NES emulated inside 1GB of Qwen AI weights). The last one is the most visual and has the clearest surprise reveal. Look up one fact about it you can state in under 10 seconds.

2. 2

   Write your hook and script10 min

   Hook (spoken on camera or as text overlay): 'Someone built a fully working Nintendo inside an AI model. Not emulated — inside the weights.' Then: 'The project is called NESformer. It runs in about 1GB of Qwen weights. You can play it right now on GitHub. And yes, it actually plays.' Close: 'The line between software and neural network just got very weird — follow for more of these.' Total: about 60 words, 25–30 seconds.

3. 3

   Record your take10 min

   One front-camera take, plain background, no B-roll needed. If you can screen-record the GitHub repo or the running demo, cut that in as a 5-second insert after your hook. If not, a still screenshot works. Speak like you're texting a friend who would find this genuinely interesting.

4. 4

   Edit and caption15 min

   Trim to the tightest version — aim for 25–35 seconds. Add the project name as a text overlay so it's searchable. Write your caption with the project name spelled out ('NESformer', 'Qwen') for search. Add 3 hashtags: one niche (#AItools or #machinelearning), one broad (#technology), one cultural (#retrogaming or #Nintendo).

Platform signal: what's worth posting where today

Tools: what these let you make now

Scripts to post today

TikTok

Someone built a Nintendo inside an AI model

“Someone built a fully working Nintendo inside an AI model. Not emulated — inside the weights.”

The project is called NESformer. It runs inside about one gigabyte of Qwen AI weights. You can literally play games inside a neural network right now — there's a link on GitHub. What makes this weird is that it's not a traditional emulator. The AI has learned to simulate the hardware. That means the game isn't running on code — it's running on probability. I don't fully know what to do with that information. But I do know the line between software and a trained model just got a lot blurrier. Follow if you want more of these.

Why it works: Opens with a genuine surprise that takes three seconds to land, then delivers a real fact (1GB, Qwen weights, GitHub link) that gives curious viewers somewhere to go — search traffic and saves both benefit.

Make a video about this

LinkedIn

I let AI do my job for 8 months. Then I tried to do it myself.

“I let AI handle the hard parts of my job for 8 months. Last week I tried to do it without it.”

Profiles trending right now

sgondala2X

Posted a thread on building an AI-native engineering org that generated 110 HN points and 141 comments — more comments than many higher-scored posts today, suggesting genuine debate rather than passive upvotes.

Steal this: The format: a practitioner sharing a real internal decision (not a take, a thing they actually did) with enough specifics that engineers can argue about the details. Specificity is what separates a viral thread from a post nobody replies to.

pbakausGitHub

Their project impeccable hit 2,300 GitHub stars on trending — a design vocabulary layer for coding agents. The repo description is unusually clear: 'catch AI slop before it ships'. That phrase alone is likely driving shares.

Steal this: Write your product description the way you'd text a developer friend: one specific problem, one specific solution, one line that's a little bit funny. 'Catch AI slop before it ships' is doing more marketing than three paragraphs of feature copy.

thedotmackGitHub

claude-mem reached 1,140 GitHub trending stars with a single tight value prop: persistent context for AI agents across sessions. The README leads with the user problem, not the technical implementation.

Steal this: Lead any product post with the friction you're eliminating, not the feature you built. 'Your agent forgets everything between sessions' is a better opener than 'Introducing persistent context management'.

What to do today

• Film the NESformer 'Nintendo inside an AI' TikTok today — the GitHub project is live, you can screen-record the demo, and the story is peaking now. Script is in this issue. (20 min)
• Write the LinkedIn AI deskilling post while the HN thread (500 pts, 378 comments) is active. First-person, one specific task you tried to do without AI, open question at the end. Post before 2pm local time for peak LinkedIn reach. (15 min)
• Build the Tea Leoni / Winona Ryder / Jamie Lee Curtis nostalgia carousel for Instagram — all three are in Google Trends today. A 6-slide 'forgotten films' format with a 'save for tonight' CTA on the last slide. (30 min)
• If you cover tech or dev topics: post a 3-platform security incident carousel today (Northflank + Bunny CDN + Linode). The audience is primed and searching; a plain-English explainer with a 'check your API keys' CTA earns saves. (25 min)
• Open impeccable (https://github.com/pbakaus/impeccable, 2,300 stars) and run it against one AI-generated UI you have sitting in a project — a landing page, a tool, a link-in-bio. Use the output as content: 'I ran a design audit on my AI-built site — here's what it flagged' is a strong LinkedIn or X thread format right now. (45 min)