thegdsks

I used to use em dashes earlier in my chats or titles for articles or Social Media Posts

But since last year the ChatGPT & CLAUDE content started flooding online.... and aggressive usage of em ( — ) dashes by AI in response

Kinda scared to use it now, cause real content or legit posts also get tagged as AI written as em dash shows up

Same with emojis also... earlier Github commit used to use feat() and now all repos cleaning their readmes and commit rules removing emojis from their old traditions...

Forced to use hyphen now , with the unsaid norms....

For anything that starts as a side project, SQLite covers it. The moment you need to ask this question the answer is usually "not yet." Ship first, migrate when it actually hurts

This is why CC0 and MIT matter for projects people depend on. Once you build on something with a restrictive license this is always a risk.

The CLI tool space has changed so much since then. Everything is getting rewritten in Rust or Go now.

How does this compare to less with syntax highlighting? I've been using bat as a pager (bat --paging=always) and it covers most of what I need. Curious what the advantage is for larger files.

This resonates. I still have a Pi 3B running pihole and it's been up for years. No updates needed, just works. The newer boards trying to compete with mini PCs feels like a different product category entirely.

Agreed. In practice I just ask "am I storing the same fact in two places?" & fix it if yes. Never once sat down and thought "let me check if this is in 4NF specifically."

Looks like they are slowly getting the OpenClaw features here in Cowork Already seeing the 5 limit per day in usage bar now..

Good Luck Remembering all those icons.. Amazing

Not seriously... I too love the sharp edges but this is scary.. Lol

Finding brand logos as clean SVGs is unreasonably painful. Press kits behind login walls. Inconsistent quality. No dark mode variants. No way to import them into a project without manual work.

I built theSVG to fix this. One search bar, 3,847 brands, copy as SVG/JSX/Vue/CDN URL.

Hey HN,

  I built AskVerdict AI — a multi-agent debate engine that
   argues both sides of a question before delivering a 
  structured verdict with confidence scores.

  How it works: You ask a question ("Should I migrate from
   REST to GraphQL?" or "Should we hire or outsource?").
  The engine spins up 2-5 AI agents with distinct personas
   (Advocate, Critic, Economist, Engineer, Devil's
  Advocate, etc.) who argue across multiple rounds. They
  build an argument graph — claims, rebuttals, evidence,
  concessions — tracked with survival scores. A consensus
  detector watches for convergence. Then a synthesizer
  analyzes everything and produces a verdict with:

  - Clear recommendation + confidence %
  - Per-option strengths/weaknesses
  - Strongest dissenting argument
  - Decision matrix with weighted criteria
  - Full narrative analysis report

  The technical bits:

  - Engine is provider-agnostic: Gemini, DeepSeek, Claude,
   Groq, OpenRouter, Ollama (local). Default preset uses
  Gemini 2.5 Pro for synthesis + DeepSeek V3 for debaters
  = ~$0.08/debate
  - Argument graph with claim statuses (standing,
  rebutted, conceded, refined), evidence tracking, and
  fallacy detection
  - Consensus detection via multiple signals: confidence
  convergence, concession rate, position shifts,
  Bayesian-inspired scoring
  - 4 debate modes: instant (<5s), fast (2 agents/2
  rounds), balanced (3/3), thorough (5/5 with web search)
  - BYOK (bring your own keys) — plug in your own API keys
   and pay providers directly. No markup
  - Full stack: TypeScript monorepo, Hono + Bun API,
  Next.js 15 frontend, Drizzle + Postgres

  Why I built this: I kept asking ChatGPT for advice and
  getting agreeable, non-committal answers. "It depends on
   your use case." Thanks. I wanted AI that actually
  commits to a position, shows its reasoning, and honestly
   represents the counterargument. The multi-agent
  approach forces genuine tension — agents can't just
  agree because they have adversarial roles.

  Free tier: 3 debates. BYOK plan: unlimited with your own
   keys. Pro: $4.99/mo.

  Would love feedback on the engine architecture and
  verdict quality. The engine package is ~4K LOC
  TypeScript if anyone wants to nerd out about consensus
  algorithms or argument graphs.

  ---
  Tips for posting:

  1. Post between 8-10 AM ET on a weekday
  (Tuesday-Thursday is best for Show HN)
  2. Don't ask for upvotes — HN will penalize that
  3. Be ready to answer technical questions in comments —
  HN loves implementation details
  4. If someone asks about the consensus algorithm or
  argument graph, go deep — that's your differentiator
  5. Have a compelling demo ready (a specific question
  with a good verdict) people can try immediately

Hey HN, I built FeatureDrop — an open-source alternative to Pendo/Appcues/Beamer for product adoption.

The core idea: instead of server-side analytics pipelines that need weeks of data, FeatureDrop runs a behavioral engine entirely in the browser (localStorage). It tracks session patterns, detects notification fatigue, and picks the right format (badge, toast, modal) per user — from the first session.

v3 just shipped with:

- AdoptionEngine (4 kB) — BehaviorTracker, TimingOptimizer, FormatSelector, AdoptionScorer - 11 framework integrations (React, Vue, Svelte, Solid, Preact, Angular, Web Components + SSR for Next.js, Remix, Astro, Nuxt) - shadcn/ui registry — 5 components installable via `npx shadcn add` - MCP server so AI editors (Claude Code, Cursor, VS Code) understand the library

Core is 3 kB gzip, engine is 4 kB gzip, zero runtime dependencies. 479 tests. MIT licensed.

The tradeoff vs server-side tools: you get instant decisions with no data collection, but you lose cross-device persistence and team-wide analytics. For most indie/small SaaS apps, that's a good tradeoff.

Docs: https://featuredrop.dev

GitHub - https://github.com/GLINCKER/featuredrop

Happy to answer questions about the architecture or the behavioral heuristics.

Hi HN. I built AskVerdict after a week at Anthropic's Claude Code Hackathon.

The core idea: instead of one LLM giving you one answer, spawn multiple agents with distinct stances that debate the question through structured rounds. Forced cross-examination, argument status tracking (standing/challenged/rebutted/conceded), and a synthesized verdict with confidence scores.

Under the hood it's a TypeScript monorepo: Hono + Bun for the API, Next.js frontend, Postgres with Drizzle ORM. The debate engine uses a multi-tier model router — heavier models for orchestration and synthesis, lighter ones for debate rounds and classification. Keeps costs under $1 per debate.

BYOK only right now. Bring your own API keys for Claude, GPT, DeepSeek, Groq, or OpenRouter. No vendor lock-in.

Opening to the first 100 beta users. Feedback welcome, especially on the debate protocol design — should agents be allowed to concede gracefully or should the system force them to defend until the end?