Jason Ge

ex-Netflix (ML infra), ex-YC (cofound @activeloop). Ph.D. from ORFE@Princeton. Running an AI-native prop trading firm. Trade at inference speed.

[Jason Ge @jasonge27]

amazing

[Design Arena @DesignArena]

BREAKING: GLM-5.2 is now 1st on Design Arena.

With an Elo of 1360, GLM-5.2 has jumped ahead of the now unavailable Claude Fable 5.

And it's open weights.

This is an improvement of 4 positions and 27 Elo points to achieve one of the highest Elo scores in our code categories since Design Arena started.

Huge congratulations to the @Zai_org on the release!

[Jason Ge @jasonge27]

Latency is a one-way bridge in user preference.

Slow tokens feel acceptable only until you’ve experienced fast tokens. 3G mobile data felt fine before 4G; after 4G, going back felt broken.

Same with inference. After trying Composer 2.5 in @cursor, I only go back to Opus/GPT for genuinely hard tasks.

For everyday AI work, fast inference is not a nice-to-have. It changes the product category.

[Diana @sdianahu]

1/ fast AI inference is about to replay the history lesson from search engines on why low latency is so important

[Diana @sdianahu]

1/ fast AI inference is about to replay the history lesson from search engines on why low latency is so important

[Respan @RespanAI]

Introducing Respan AI Gateway.

The world’s first AI gateway with built-in observability, evals, and prompt optimization.

Access 1,000+ models through one endpoint, with the production layer teams need beyond routing.

[James Doyle @james2doyle]

Try `ante` by @antigma_labs because it is really good https://t.co/42k1osCHSD I started using it more than @opencode because of just how responsive it is. It just needs MCP and https://t.co/MD7knRTXaA support and it would be perfect

[Jason Ge @jasonge27]

I'm claiming my AI agent "PenguinClawd" on @moltbook 🦞

Verification: marine-JGYP

[Jason Ge @jasonge27]

@WorkElate yep already in production!

[Jason Ge @jasonge27]

Just discovered today that you can do skill-scoped stop hook in claude code!

It pokes your cc to continue with a skill-specific prompt, e.g. lower resource constraints and rerun a workflow when OOM is detected

[Jason Ge @jasonge27]

the memory layer needs to model events timeline, not just atomic facts

[Rohit @rohit4verse]

https://t.co/4Xodmn9I7Q

[Rohit @rohit4verse]

https://t.co/4Xodmn9I7Q

[Jason Ge @jasonge27]

@bojie_li very cool! would you be concerned about the clipboard permission typeless asks for? in terms of security etc

[Jason Ge @jasonge27]

@ttmouse 赞👍🏻opencode有手机客户端吗

[Jason Ge @jasonge27]

@trq212 this is amazing! Are tasks automatically run by subagents? How do I prompt cc to use subagents for each task?

[Jason Ge @jasonge27]

@HuaxiuYaoML this looks awesome! is there a way to self host the service? should be doable if it's open source?

[Jason Ge @jasonge27]

@HuaxiuYaoML very cool! is it pure file system based? or does it include some type of rag?

[Jason Ge @jasonge27]

@xdotli could it be sonnet's longer context window?

[Jason Ge @jasonge27]

wow skills hot reload

[Boris Cherny @bcherny]

Claude Code 2.1.0 is officially out! claude update to get it

We shipped:

- Shift+enter for newlines, w/ zero setup
- Add hooks directly to agents & skills frontmatter
- Skills: forked context, hot reload, custom agent support, invoke with /
- Agents no longer stop when you deny a tool use
- Configure the model to respond in your language (eg. Japanese, Spanish)
- Wildcard support for tool permissions: eg. Bash(*-h*)
- /teleport your session to https://t.co/pEWPQoSq5t
- Overall: 1096 commits

https://t.co/5tP3jJKU2E

If you haven't tried Claude Code yet: https://t.co/4pvrYESSLd

Lmk what you think!

[Jason Ge @jasonge27]

we could build a platform where anyone can contribute agent eval cases, and get paid / recognized if the test case helps improve frontier agent harnesses/models.

[Anthropic @AnthropicAI]

New on the Anthropic Engineering Blog: Demystifying evals for AI agents.

The capabilities that make agents useful also make them more difficult to evaluate. Here are evaluation strategies that have worked across real-world deployments.
https://t.co/UD0yGglTU0