Hi, I am a game dev, artist & gamer. Check links for more content. Links: https://teapaw.org/t.co/YdQYWDXbJ6 https://teapaw.org/t.co/MRw3hw6wLZ
Ottawa,Canada
Joined July 2014
- Tweets1.1K
- Following442
- Followers143
- Likes358
Kalponic Games(Commissions Open) retweeted
We cracked isometric walking spritesheets. Fully.
And we're giving away the entire recipe: prompts, skills, everything. Make your own in seconds.
scenario.com/explorations/is…
Kalponic Games(Commissions Open) retweeted
Two Hong Kong students just made Karpathy's loop 5x better - dropped 18-page PDF
The twist: the loop got 5x better the moment you put another loop on top of it
here's the whole method, step by step:
step 1 → Karpathy's loop gets stuck - the LLM keeps reproposing the same changes, falling back to its priors
step 2 → so they add an outer loop that reads the inner loop's code and finds where it's stuck
step 3 → the outer loop writes new search logic as Python and injects it live - 5x better, same model
how to steal this for your agents:
step 4 → write a second agent whose only job is to read the first one's logs and find where it's stuck
step 5 → let it rewrite the rules - workflow, skill, prompt - not just retry the task
step 6 → auto-revert every rewrite on failure, so a bad change never breaks your pipeline
the result: 5x better than Karpathy's loop alone - same LLM, no smarter model, it's the architecture
this 18-page PDF is what comes after the Karpathy loop
read it now - the full build workflow is in the article below ↓
Kalponic Games(Commissions Open) retweeted
UN PROGRAMADOR ACABA DE LOGRAR LO QUE GOOGLE LLEVA AÑOS PASANDO POR ALTO
Desarrolló un navegador en rust creado específicamente para automatizar procesos, hacer scraping web y potenciar agentes de IA
> Solo consume 30MB de RAM
> Las páginas cargan en apenas 85ms
> Bloquea automáticamente más de 3.500 trackers
> Elimina anuncios, analíticas y scripts de rastreo
Se llama Obscura
Y tiene algo que Chrome nunca va a poder ofrecer
Cada sesión genera una huella completamente distinta. GPU, canvas, audio, batería… todo se randomiza
Ningún detector lo identifica porque se comporta exactamente igual que un Chrome real
Es el reemplazo directo de Puppeteer y Playwright
Sin Node.js. Sin dependencias. Un solo binario
Ya supera las 16k estrellas en GitHub. 100% open source. Totalmente gratis
Guárdalo antes de que se te olvide, es una joya 📄
Kalponic Games(Commissions Open) retweeted
Generates educational videos via code
github.com/showlab/Code2Vide…
Kalponic Games(Commissions Open) retweeted
ASASR is now open-sourced.
performs x4 image SR with a FLUX.1-dev and dual-LoRA inference: upscaling prior, OminiControl + DPO LoRA
huggingface.co/wafer-bob/ASA…
SUPIR upscaler is outdated. ASASR- turns blurry, low-quality photos into sharp, high-res images. Prevents the fake hallucinated details.
- improves OCR
- high segmentation accuracy
- based on FLUX.1 dev
this looks sweet
github.com/wafer-bob/ASASR
Kalponic Games(Commissions Open) retweeted
需要跟同事讲解项目系统架构，光说不画图效果有限，自己动手画又费时间还画得不好看。
archify，一个能装进 Claude Code、Codex CLI 和 opencode 的 Agent Skill，把一段大白话描述直接变成一张架构图。
能画系统架构图、工作流程图、时序图、数据流向图和生命周期状态图这五种技术图，深色浅色主题一键切换。
GitHub：github.com/tt-a1i/archify
生成的是单个自包含的 HTML 文件，不装额外依赖打开浏览器就能看，图能直接复制粘贴到 Slack 或 Notion 里。
也能导出到 4 倍分辨率的 PNG、JPEG、WebP，或者矢量 SVG。
经常需要跟同事讲清楚架构、写技术文档配图的朋友，用 Claude Code 顺手就能画，比手动画图省不少事。
Kalponic Games(Commissions Open) retweeted
Use Fable 5 as orchestrator and Opus + Codex to execute (to save fable usage):
Fable 5 (max reasoning) = orchestrator
Opus = deep reasoning subagent
Sonnet = mechanical work subagent
Codex = peer Sr. engineer, different perspective
Setup:
1. Set Fable 5 as your main model In Claude Code: /model → Fable 5 → reasoning /effort to max
2. Create 2 subagents with /agents In Claude Code:
deep-reasoner → pinned to opus "Use for reasoning-heavy phases, architecture, debugging complex issues, algorithm design. Think thoroughly, return a concise conclusion the orchestrator can act on."
fast-worker → pinned to sonnet "Use for mechanical tasks, boilerplate, tests, formatting, simple edits. Execute efficiently."
3. Add OpenAI's official Codex plugin (install codex cli in your computer first), In Claude Code type:
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/codex:setup
4. Drop this in your CLAUDE.md in your folder:
## Orchestration workflow
You (Fable) are the orchestrator. Plan, decompose, synthesize.
Reasoning-heavy phases → deep-reasoner
Mechanical work → fast-worker
Codex (/codex:rescue --background) is a cracked engineer on par with deep-reasoner, from a different perspective. Treat as a peer, not a reviewer.
High-stakes decisions: task Opus + Codex on the same problem in parallel, synthesize the best of both, without showing either the other's answer. Keep your own context lean.
5. Then prompt Fable like a tech lead: "Goal: [what you want] Context: [files, constraints] You're the lead. Delegate reasoning to deep-reasoner, grunt work to fast-worker, fresh-perspective problems to Codex. Show me your plan first, then execute."
That's it.
Kalponic Games(Commissions Open) retweeted
50 USEFUL GITHUB REPOS
1. iFixAi — AI misalignment testing
→ github.com/ifixai-ai/iFixAI
2. public-apis — free APIs for everything
→ github.com/public-apis/publi…
3. build-your-own-x — learn by building
→ github.com/codecrafters-io/b…
4. developer-roadmap — learn any tech skill
→ github.com/kamranahmedse/dev…
5. free-programming-books — thousands of free books
→ github.com/EbookFoundation/f…
6. system-design-primer
→ github.com/donnemartin/syste…
7. coding-interview-university
→ github.com/jwasham/coding-in…
8. the-art-of-command-line
→ github.com/jlevy/the-art-of-…
9. project-based-learning
→ github.com/practical-tutoria…
10. you-dont-know-js
→ github.com/getify/You-Dont-K…
11. the-book-of-secret-knowledge
→ github.com/trimstray/the-boo…
12. tech-interview-handbook
→ github.com/yangshun/tech-int…
13. awesome-selfhosted
→ github.com/awesome-selfhoste…
14. javascript-algorithms
→ github.com/trekhleb/javascri…
15. 30-seconds-of-code
→ github.com/Chalarangelo/30-s…
16. github gitignore templates
→ github.com/github/gitignore
17. ollama — run LLMs locally
→ github.com/ollama/ollama
18. langchain
→ github.com/langchain-ai/lang…
19. n8n automation
→ github.com/n8n-io/n8n
20. openclaw — local AI assistant
→ github.com/openclaw/openclaw
21. dify — AI app builder
→ github.com/langgenius/dify
22. langflow
→ github.com/langflow-ai/langf…
23. mem0 — AI memory layer
→ github.com/mem0ai/mem0
24. browser-use
→ github.com/browser-use/brows…
25. crewAI
→ github.com/crewAIInc/crewAI
26. MetaGPT
→ github.com/geekan/MetaGPT
27. AutoGen (Microsoft)
→ github.com/microsoft/autogen
28. aider — AI coding assistant
→ github.com/Aider-AI/aider
29. markitdown (Microsoft)
→ github.com/microsoft/markitd…
30. open-webui
→ github.com/open-webui/open-w…
31. maigret — OSINT tool
→ github.com/soxoj/maigret
32. TradingAgents
→ github.com/TauricResearch/Tr…
33. stagehand
→ github.com/browserbase/stage…
34. firecrawl
→ github.com/mendableai/firecr…
35. transformers (HuggingFace)
→ github.com/huggingface/trans…
36. vLLM
→ github.com/vllm-project/vllm
37. llama.cpp
→ github.com/ggerganov/llama.c…
38. llama_index
→ github.com/run-llama/llama_i…
39. nanoGPT (Karpathy)
→ github.com/karpathy/nanoGPT
40. RAGFlow
→ github.com/infiniflow/ragflo…
41. supermemory
→ github.com/supermemoryai/sup…
42. awesome-claude-skills
→ github.com/ComposioHQ/awesom…
43. Bumblebee (Perplexity AI security tool)
→ github.com/perplexityai/bumb…
44. ComfyUI
→ github.com/comfyanonymous/Co…
45. DeepSeek (official repo varies by model release)
→ github.com/deepseek-ai
46. Lobe Chat
→ github.com/lobehub/lobe-chat
47. freeCodeCamp
→ github.com/freeCodeCamp/free…
48. system design + interview prep extras (bonus)
→ github.com/jwasham/coding-in…
49. AI agents ecosystem (extra)
→ github.com/langchain-ai/lang…
50. automation + AI workflows (extra)
→ github.com/n8n-io/n8n
Save this before you forget it exists.
You'll come back to at least ten of these.
Kalponic Games(Commissions Open) retweeted
Introducing GameBlocks: open-source building blocks that help coding agents prototype any 3D games with @threejs
3D behavior is hard to specify in natural language. GameBlocks turns fragile spatial logic into inspectable implementations with clear semantics, so agents can build from known-good patterns.
Github: github.com/xt4d/GameBlocks
#gamedev #indiedev #threejs #codex #claude
Kalponic Games(Commissions Open) retweeted
Tencent 🔥: They open-sourced a long term memory system for AI agents that cuts token usage by 61.38% and it runs fully local with zero external API dependencies.
TencentDB Agent Memory builds a 4-tier semantic pyramid:
> L0: Raw conversation logs
> L1: Atomic facts extracted from those logs
> L2: Scene blocks grouped from related facts
> L3: Full user persona distilled from everything above
The agent reads the persona first and drills down to raw logs only when it needs to verify a specific detail.
- Upper layers carry judgment.
- Lower layers carry evidence.
For short-term memory, it compresses heavy tool logs into Mermaid symbol graphs. Instead of thousands of tokens of verbose output sitting in context, you get a lightweight node map. The agent navigates it using node IDs, pulling the full raw text only when an error occurs.
The benchmark numbers from the WideSearch test: task success rate up 51.52%, token consumption down 61.38%. On PersonaMem, accuracy jumped from 48% to 76%.
These were measured over continuous long-horizon sessions running 50 consecutive tasks each not isolated single-turn evals.
github.com/TencentCloud/Tenc…
Kalponic Games(Commissions Open) retweeted
how to improve your /goal by designing the loop before you run it
a poorly designed loop burns tokens and hands you slop fast, that´s why it's important to spend time designing the loop harness
instead of writing a /goal and hoping it works, you run it through LOOPER skill first and it:
> critiques your loop and sharpens it against a library of best practices
> builds a verification rubric, so the loop actually knows what done means
> lets you add an llm as judge, codex and non-claude models included
> sets guardrails and a token budget, tight or generous, so it doesn't run away
> exports the whole thing as a portable artifact you can repeat and edit
it's a design layer that helps you build a loop that produces quality, then hands it to Claude Code or other models
this pairs with the LLM council setup I posted last week. design the loop in looper, run it with a second model checking the work
nice work @KSimback
Feeling a bit loopy? Introducing Looper - your loop design coach
It’s been made clear that we shouldn’t be promoting agents, we should be designing loops
You can use /loop or /goal but if your loop is poorly designed then it won’t produce good output
Even worse, a poorly designed loop will burn a ton of tokens
So I created Looper - a tool to help me design better loops, and decided to open source it
Here’s what it does:
> /looper invokes the loop design wizard
> critiques your goal/loop and helps you sharpen it based on a library of best practices
> builds a verification rubric so it knows what "done" should be
> allows you to add an LLM as judge, including non-Claude models like Codex
> adds guardrails and helps you set a budget with tight/generous options to avoid token waste
> generates portable artifacts to make it easy to repeat / edit the loop
> gives you a visual diagram for your loop (see replies for examples)
Looper is intentionally a design layer, not a heavy orchestration framework - it helps you create a loop worth running, then hands it to Claude Code or your own runner/orchestrator
Please share your feedback, this was v0.1 so I expect it to have room for improvements, and feel free to submit issues or PRs
Note: Looper was inspired by @mvanhorn's "WTF is a loop" article and then fine-tuned with some of the tips in the great loop architecture article by @djfarrelly
github.com/ksimback/looper
Kalponic Games(Commissions Open) retweeted
A senior Google engineer dropped a 424-page doc on agentic design patterns.
424 pages.
Most engineers bookmarked it and never opened it again.
I read the whole thing.
Here are the 15 patterns that actually matter — explained in plain English, with exactly when to use each one ↓
Kalponic Games(Commissions Open) retweeted
🚨 @Karpathy predicted the power of the "LLM Wiki." Google just formalized it.
Meet Open Knowledge Format (OKF): a vendor-neutral standard for giving foundation models the curated context they need.
I can genuinely see this replacing Notion, Obsidian, or traditional wikis for developer teams, and the reason comes down to bookkeeping.
Traditional wikis fail because humans inevitably abandon the tedious work of updating them.
As Andrej Karpathy pointed out recently, LLMs don't get bored.
They don't forget to update a cross-reference, and they can touch 15 files in a single pass.
OKF standardizes the interoperability layer so agents can actually do that heavy lifting autonomously.
Because the format is minimally opinionated, it doesn't dictate what you write, it just dictates how it's structured. You get:
→ Human-readable documents that live right alongside your code in version control
→ Cross-links that map out complex entity relationships without needing a graph database
→ A system that survives moving between different tools and organizations
There is no complex compression scheme.
No central registry.
If you can cat a file, you can read it.
If you can git clone a repo, you can deploy it.
This is how we stop rebuilding context pipelines from scratch every time a new model drops.
Announcement + spec file in 🧵↓
Kalponic Games(Commissions Open) retweeted
Qdrant is a production-ready vector search engine and database built in Rust, designed for high-performance similarity search with extended filtering support.
- Written in Rust for speed and reliability under high load.
- RESTful API with convenient client libraries for Go, Python, and more.
- Supports advanced filtering, quantization, sharding, and hybrid search.
- Available as a self-hosted Docker image or fully managed cloud service with a free tier.
Kalponic Games(Commissions Open) retweeted
How to effectively run autonomous long-running coding agents?
This is one of the most exciting discussions on agents I've ever had.
I recorded it and am making it freely available.
(bookmark it)
The idea of autonomous long-running agents is a real thing.
We talk about lots of things like /goal, /loop, and dynamic workflows, and what comes next.
One interesting discussion was around how to make the agent run for longer while ensuring it stays on track.
Most models today will struggle to coordinate work effectively. They sometimes pause the work early. Lots of mistakes happen, and lots of weird shortcuts (reward hacking).
What helps is to be extremely clear about the goals it needs to achieve. To clarify the dos and don'ts clearly. Eliminate any assumptions you think the model would make. Deep expertise matters so much in this.
But you can get far through careful planning. My formula currently is to use Opus 4.8 for planning carefully and GPT-5.5 for all executions. For the evaluator (via /goal), I am often using something like Deepseek or the latest models from Qwen, Kimi, and MiniMax, etc.
Another insight we discussed to enforce goals is to provide strong visual cues for the agent to compare with. I found that a multimodal goal is a much stronger goal than a plain text one. And use agents to help you set clear goals.
Watch here: academy.dair.ai/events/cmplo…
Kalponic Games(Commissions Open) retweeted
传统的RAG已经不够看了，别再用那种“每次都要从头检索”的低效模式了。
GitHub上这个LLM Wiki项目已经突破1万星，它让AI帮你增量构建结构化的知识维基——知识只编译一次，越用越聪明，这才是真正意义上的第二大脑。
几个真正硬核的点：
① 四信号知识图谱，自动建立高质量关联
② Louvain社区检测，帮你发现自己都没意识到的知识盲区
③ Chrome一键剪藏，好网页直接丢进去
④ 完美兼容Obsidian，颜值在线
搞研究、做笔记的朋友，这波必须冲。
🔗 github.com/nashsu/llm_wiki
Kalponic Games(Commissions Open) retweeted
You don’t need to spend weeks reading 50 PDFs.
Upload them to NotebookLM, then use Claude to turn that information into insights you can actually use.
Here are 8 prompts that can help compress 200+ hours of research into a single Sunday afternoon.
Bookmark this thread 🧵👇
