Cole Medin
@cole_medin
All about entrepreneurship, AI, and coding ๐ป
Combining my passions to build an awesome business and teach others my secrets ๐จโ๐ซ
89 Following 2.6K Followers
Your AI agent's memory is probably stuck inside one app.
Walrus Memory (@WalrusProtocol) fixes that with an MCP that creates memories as you chat, then can recall them anywhere - Claude Code, Pi, etc.
Portable and actually owned by you!
Show more
Karpathy's LLM compiler idea: raw docs -> LLM -> wiki articles -> agent queries at runtime.
Everyone's applying it to external data.
But your Claude Code session logs are the same thing. Raw folder already exists. Nobody's compiled it yet...๐ค
Show more
Monday Streams are play with subs time so join The Game Room and letโs do this!!
I just started a Discord for anyone who likes gaming, hanging out, and non-toxic lobbies ๐ฎ
Still small right now but weโre building a fun community.
Come join us ๐พ
SWE-bench Verified is contaminated. OpenAI just published the proof.
Top models: 70%+ on Verified, ~23% on SWE-bench Pro.
All frontier models can reproduce original fixes from memory. 59% of hard tasks have flawed tests.
The benchmark everyone was citing? Meaningless now.
Show more
Best coding agent tip I've tried recently: just add "use red/green TDD" to your prompts. Agent writes tests first, confirms they fail, then implements. Catches broken code and unnecessary code. Two words, pretty big difference in output quality.
Show more
Microsoft confirmed a Copilot bug has been summarizing confidential emails since late January, bypassing DLP policies.
Not a prompt injection. Not an attack. Just a bug that gave AI access to data it shouldn't have processed.
Every AI assistant that goes deeper into your data increases the blast radius of a single misconfiguration.
Show more
