Vividh

math @uwaterloo

[Vividh @lasdvv]

I just beat every provider on @nottecore 's BrowserArena benchmark.

Introducing BaseLayer, browser infrastructure for AI agents. 224ms p50 sequential browser-session lifecycle latency, 100+ ms faster than the next-fastest provider.

Firecracker microVMs + Chromium snapshot/restore, self-hosted on baremetal. CDP-ready browser restored per session, no startup on the hot path.

read more: https://t.co/zAVUEzYOlH

[Vividh @lasdvv]

@PanduSpams food has never had so much aura

[Vividh @lasdvv]

@theo Baremetal on lakebed👀??

[Vividh @lasdvv]

@kenwuuuu @perplexity_ai 12 languages!?!

[Ken Wu @kenwuuuu]

I noticed @perplexity_ai Comet only route your query to google or perplexity if your query is in English and default to google for languages like Chinese. As a Chinese speaker I developed my own router that routes my queries for me and it works in 12 languages.

Introducing JieHuo (解惑): a browser-runnable Google vs Perplexity query router across 12 languages.

0.883 macro F1 on a balanced multilingual gold set (240 queries, equal labels per language). 0.962 macro F1 when auto-routing only high-confidence queries, 57% coverage, ambiguous queries stay with the user.

GPT-4.1-mini teacher labels → fine-tuned multilingual-e5-small → temperature scaling → INT8 ONNX → transformers.js Web Worker. No server inference on the classification hot path.

try it: https://t.co/BntQzMgZo2

[Vividh @lasdvv]

On concurrency: my latest c10 run was ~555ms p50, on one 48-vCPU m5zn.metal host, not a fleet routing each request to the least-loaded node.

The project was centered around “how far can a single node go,” with experimental density profiles and fluid-compute experiments for active/idle CPU policy.

[Vividh @lasdvv]

@lucas_gdno for sure, will open a PR

[Vividh @lasdvv]

@nottecore github: https://t.co/QAXqheWU7U

[Vividh @lasdvv]

I just beat every provider on @nottecore 's BrowserArena benchmark.

Introducing BaseLayer, browser infrastructure for AI agents. 224ms p50 sequential browser-session lifecycle latency, 100+ ms faster than the next-fastest provider.

Firecracker microVMs + Chromium snapshot/restore, self-hosted on baremetal. CDP-ready browser restored per session, no startup on the hot path.

read more: https://t.co/zAVUEzYOlH

[Vividh @lasdvv]

@samatnotte appreciate the kind words, BrowserArena made this whole project possible. one neutral benchmark changes everything

[Vividh @lasdvv]

@theo @nottecore hello 🙃

[Vividh @lasdvv]

I just beat every provider on @nottecore 's BrowserArena benchmark.

Introducing BaseLayer, browser infrastructure for AI agents. 224ms p50 sequential browser-session lifecycle latency, 100+ ms faster than the next-fastest provider.

Firecracker microVMs + Chromium snapshot/restore, self-hosted on baremetal. CDP-ready browser restored per session, no startup on the hot path.

read more: https://t.co/zAVUEzYOlH

[Vividh @lasdvv]

@jxnlco I just play along to music and vibe while waiting

[Vividh @lasdvv]

@ryanvogel Lets go

[Vividh @lasdvv]

@lucashjin Wtf

[Vividh @lasdvv]

@GaddipatiHarsha Liked dash more tbh

[Vividh @lasdvv]

@thdxr I was so sad I didnt get into this event😭

[Vividh @lasdvv]

@BenjDicken theyd hire you and make you take chess classes lol

[Vividh @lasdvv]

@kenwuuuu @zmwang lmaoo

[Vividh @lasdvv]

@theo I find the desktop app better for parallel work and cli for megathreads. I also like that same chat can be used in either interface with codex