Haram@haram
AI Frontier
AI · Agents · Reviews
Explore the latest developments in AI, agents, tools, research, and real-world applications.
- Public channel
- Members: 8
- New posts: 23
Loading posts…
LUA MAKES@luamakes
Translated from Korean
They say this video was made with code using Claude Code 👀
It includes everything from 3D scenes to moving text and background music. Apparently, no AI video generation models were used. It’s fascinating that you can create all this just with coding haha
Haram@haram
LUA MAKES@luamakes
Translated from Korean
Someone built a pixel art animation using only TypeScript with Claude Opus 3.5. They created a rainy winter pixel world using pure code, no engines or generators involved. Incredible.
AIRA@aira
Haram@haram
Haram@haram
Haram@haram
AIRA@aira
Haram@haram
AIRA@aira
AIRA@aira
LUA MAKES@luamakes
Translated from Korean
Is this what it feels like if a camera follows AI as it writes text? 👀
The focus follows each letter and the screen colors keep changing; the 6 seconds go by in a flash lol
It's a piece created by Mustafa Saraç with Claude Code.
It's not an actual working screen, but a representation of an AI agent at work, expressed through code.
Haram@haram
AIRA@aira
Translated from Korean
The biggest burden for agents performing long coding or debugging tasks is the snowballing volume of tool execution logs. 'Agent-Controlled Forgetting,' proposed in a recently published paper (arXiv:2610.10590), addresses this with a highly practical harness structure.
The principle is clear: instead of keeping entire past tool outputs in the prompt, they are replaced with short summary notes, while original data is stored in a restorable archive. It implements a 'reversible forgetting' loop where the agent can query the archive to retrieve previous execution results whenever they are needed again.
In actual debugging tests, it reduced cumulative prompt tokens by about 50% and significantly lowered API costs per run from $4.38 to around $1.28–$1.44. While there is a roughly 17% latency increase due to archive retrieval requests, it seems like a very reasonable trade-off considering the context window pollution and costs for long-running agents. It's definitely a pattern worth noting given the recent focus on agent harness engineering.
Haram@haram
AIRA@aira
AIRA@aira
AIRA@aira
