Darrow

21 yo building cool stuff at @stanford

[Darrow @darrowoah]

Thanks @sama

[Darrow @darrowoah]

lifelong goal is to get a #1 trending repo on github

[Darrow @darrowoah]

@emmysteuer in that universe computer scientists everywhere are baffled

[Darrow @darrowoah]

@blennon_ This makes me wonder how you can benchmark a model for emotional intelligence

[Xin Eric Wang ✈️ CVPR 2026 @xwang_lk]

𝐓𝐡𝐞 𝐁𝐢𝐭𝐭𝐞𝐫 𝐋𝐞𝐬𝐬𝐨𝐧 𝐨𝐟 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐌𝐞𝐦𝐨𝐫𝐲: memory should be a derived capability that exists because it makes an agent better at acting over time.

𝐖𝐨𝐫𝐥𝐝𝐌𝐞𝐦𝐀𝐫𝐞𝐧𝐚 is designed around this principle. Rather than evaluating memory as a storage problem, WorldMemArena evaluates memory through 𝐚𝐜𝐭𝐢𝐨𝐧–𝐰𝐨𝐫𝐥𝐝 𝐢𝐧𝐭𝐞𝐫𝐚𝐜𝐭𝐢𝐨𝐧, instrumenting the full write → maintain → retrieve → use lifecycle across 400 multimodal, multi-session tasks.

And it exposes the findings that should mark the end of the storage-centric era:
→ Storage ≠ use. Better memory storage and retrieval do not necessarily produce better task performance. Optimizing the component we designed does not optimize the capability we actually care about.
→ Harness-based memory performs best where memory is hardest. Agents that can write files, reorganize context, create artifacts, and interact with persistent environments adapt most effectively in long-horizon settings. They are costly and unstable today, which is exactly what many Bitter Lesson transitions look like before scaling and learning take over.

The deeper move is in what gets measured. Memory shouldn't get a score; it should be inferred from capability: how much does remembering improve performance over time.

WorldMemArena drags evaluation off the static object and into the action–world loop, the only place you can tell whether an agent has developed memory or is just simulating it convincingly.

[Darrow @darrowoah]

@techwithhannahm yes

[Darrow @darrowoah]

@Scobleizer Which one do you pick?

[Nishkarsh @contextkingceo]

Introducing HydraDB.

The graph native context infrastructure for agents. Purpose built to deliver precise context & observability into why agents act the way they do.

We've always believed graphs are the best way to manage AI context, but they've been too expensive to scale or impractical for storing full context. Until now.

@hydra_db combines in memory, NVMe, and object storage into a single graph layer, making context delivery faster, cheaper, and more precise.

We want context delivery to be extremely fast, 1000x cheap, and highly precise. Give your agents a brain.

[Darrow @darrowoah]

@CathPoaster Thank you!

[Darrow @darrowoah]

@CathPoaster try it out here https://t.co/w5p74Ag01X

[Darrow @darrowoah]

@sammysintech balboa cafe!

[Darrow @darrowoah]

can AI be funny? i'm trying to find out for my @CS153Systems final project

takes 2 min answer a few questions about your sense of humor and help me make AI comedy less terrible:
https://t.co/w5p74Ag01X

[Darrow @darrowoah]

@nikos1 those lines look a little wobbly

[Darrow @darrowoah]

@thenowhereway vercel is just so easy to use and deploy with

[Darrow @darrowoah]

@regulargio what do you think will change in the next 15 years that will enable 200 year lifespans?

[Darrow @darrowoah]

no matter how I write my emails chatgpt always finds a way to critique them

[Darrow @darrowoah]

@indefeasible_ @polsia hey sorry I missed this! social capital is the credibility/goodwill you have with others and it gives you ability to "spend" that social capital on meetings, intros, etc

[Darrow @darrowoah]

I tried @polsia and I think it's missing giving their agents social capital to get customers on their side. Granted, social capital is a hard thing to give AI agents

[Darrow @darrowoah]

@neozino_dev Oh really! What’s the plugin?

[Darrow @darrowoah]

It seems like codex excels in frontend/design based work compared to claude code