Show HN: GPT Classifies HN Titles

6 points

3 years ago

With all the "what can GPT do", I went to dabble with it to see if I could do something that plays to its strengths rather than just speculating about "would it be good at that?"

https://gist.github.com/shagie/ff2f67b33bb517f6a297e2da66530cf4

The basis for the prompt is:

    Classify following titles into one or more of {US Politics}, {World Politics}, {Technology}, {Security}, {Current Events}, {Boasting}, {Curiosity}, {Unknown}

The results from the front page as I type this:

    34152137    Technology: Lite XL: A lightweight text editor written in C and Lua
    34151880    Curiosity:  Hydrochloric acid is more complicated than you think
    34156235    World Politics, Current Events: Apple Japan hit with $98M in back taxes for missing duty-free abuses
    34152333    Technology: The Architecture of the Lisa Personal Computer (1984) [pdf]
    34151951    Curiosity:  The Zen anti-interpretation of quantum mechanics (2021)
    34155009    Unknown:    Added 49 volumes of Arkansas, Mississippi, and Tennessee law
    34145665    Technology: FEA Solver that doesn't need defeaturing
    34151107    World Politics, Current Events: Amazon packages burn in India, final stop in broken recycling system
    34155783    Technology: Your Cellphone Will Be a Satphone
    34154778    World Politics, Current Events: Vandals destroy 30k-year-old indigenous cave drawings in Australia
    34146397    Boasting:   Ask HN: Elevator Pitch for a Polymath?
    34151708    Technology: All About USB-C: Illegal Adapters
    34149340    World Politics, Current Events: Southwest cancels 5,400 flights in less than 48 hours
    34154896    Technology: Beaker Browser is now archived
    34145811    Technology: Run Third-Party Scripts from a Web Worker
    34152978    World Politics, Current Events: When Algorithms Rule, Values Can Wither
    34152981    Security:   Experian Vulnerability Shows Any Report with just SSN, DOB, ADR
    34151382    World Politics, Security, Current Events:   US Government demands direct police access to European biometric data [pdf]
    34143272    Curiosity:  Archaeologists devise a better clock for Biblical times
    34149804    Technology: Using eBPF and predefined inspections to minimize “observability tax”
    34155875    US Politics:    Ventura Issues
    34152100    Technology: Anki SRS Algorithm : Spaced repetition explained with code
    34154406    Technology: How RISC-V Timer Interrupts Work
    34146493    Curiosity:  Old-school computing: when your lab PC is ancient (2021)
    34152369    Technology: BorgBackup: Deduplicating archiver with compression and encryption
    34147825    Technology: Show HN: Self-hosted CMS on serverless Cloudflare
    34153648    Boasting:   Brooklyn Bar Menu Generator (2015)
    34117429    Unknown:    “Blue Light” creating capacity for nothing (2007)
    34149239    Curiosity:  Visually symmetric words

There are certainly some that one could argue would be different though I think its fairly correct based on titles alone. I'll also note that there are some different results when I was working in a smaller sample size - the entry "Added 49 volumes of Arkansas, Mississippi, and Tennessee law" was classified as "US politics, Current Events" rather than "Unknown".

I do want to note that isn't that bad.

Running this:

    "prompt_tokens":560,"completion_tokens":222,"total_tokens":782

So, this is just a hair under $0.02 to do.

Thinking about it, some cleanup can be done by removing the year and media type which could trim a few tokens off the request.

4 comments

34152137 Technology: Lite XL: A lightweight text editor written in C and Lua 34151880 Curiosity: Hydrochloric acid is more complicated than you think 34156235 World Politics, Current Events: Apple Japan hit with $98M in back taxes for missing duty-free abuses 34152333 Technology: The Architecture of the Lisa Personal Computer (1984) [pdf] 34151951 Curiosity: The Zen anti-interpretation of quantum mechanics (2021) 34155009 Unknown: Added 49 volumes of Arkansas, Mississippi, and Tennessee law 34145665 Technology: FEA Solver that doesn't need defeaturing 34151107 World Politics, Current Events: Amazon packages burn in India, final stop in broken recycling system 34155783 Technology: Your Cellphone Will Be a Satphone 34154778 World Politics, Current Events: Vandals destroy 30k-year-old indigenous cave drawings in Australia 34146397 Boasting: Ask HN: Elevator Pitch for a Polymath? 34151708 Technology: All About USB-C: Illegal Adapters 34149340 World Politics, Current Events: Southwest cancels 5,400 flights in less than 48 hours 34154896 Technology: Beaker Browser is now archived 34145811 Technology: Run Third-Party Scripts from a Web Worker 34152978 World Politics, Current Events: When Algorithms Rule, Values Can Wither 34152981 Security: Experian Vulnerability Shows Any Report with just SSN, DOB, ADR 34151382 World Politics, Security, Current Events: US Government demands direct police access to European biometric data [pdf] 34143272 Curiosity: Archaeologists devise a better clock for Biblical times 34149804 Technology: Using eBPF and predefined inspections to minimize “observability tax” 34155875 US Politics: Ventura Issues 34152100 Technology: Anki SRS Algorithm : Spaced repetition explained with code 34154406 Technology: How RISC-V Timer Interrupts Work 34146493 Curiosity: Old-school computing: when your lab PC is ancient (2021) 34152369 Technology: BorgBackup: Deduplicating archiver with compression and encryption 34147825 Technology: Show HN: Self-hosted CMS on serverless Cloudflare 34153648 Boasting: Brooklyn Bar Menu Generator (2015) 34117429 Unknown: “Blue Light” creating capacity for nothing (2007) 34149239 Curiosity: Visually symmetric words