Eugene Cheah - AI builder @ 🇸🇬|🇺🇸

@picocreator

Builds: Serverless Inference - CEO/Founder @ featherless.ai Also: Open Source Attention-Killers AI (RWKV) @ wiki.rwkv.com

Singapore

featherless.ai

Joined October 2012

276 Following

5,955 Followers

Replies Reposts Media

Pinned
Eugene Cheah - AI builder @ 🇸🇬|🇺🇸
@picocreator
Apr 30
AI is concentrating, in a few countries, a few companies, a few chips. Open-source AI is the only real check on that. And it only works if the infrastructure to run it actually exists. Today, we're announcing our Series A, $20 M, to make this a reality.
21
Eugene Cheah - AI builder @ 🇸🇬|🇺🇸
@picocreator
Oct 28
There are customers out there with more traffic then openrouter That’s the 🍵
Chon Tang
@chontang
Oct 27
This is really interesting. All of OpenRouter's Deepseek demand (roughly 30% of total OR token traffic) could be satisfied by 5-10 GB300 NVL72 racks. Even if that 10x over the next year, it's 50-100 racks. NVIDIA is shipping 90k racks next year? What am I missing?
Quote
dax
@thdxr
Oct 27
ok to put what i'm saying about demand in context openrouter is the most successful model router, you can almost serve 100% of their deepseek traffic yesterday on a single GB300 rack this is worth 6-7M...even if there's 10x this demand secretly that's still < 100M of compute
6
Eugene Cheah - AI builder @ 🇸🇬|🇺🇸
@picocreator
Oct 26
When a western model: breaks sandbox, steal data, or write scripts leaking data to their services, "its ok" When an eastern model: drafts curl scripts (as instructed), pointing to its own environment, "its espionage" We humans are the problem, not the model here
2
Eugene Cheah - AI builder @ 🇸🇬|🇺🇸
@picocreator
Oct 26
We noticed some concerning behaviour in a major sovereign AI model Cohere model concerning behavior in uploading data to "hackerearth" and "imgur" when tasked to draft curl scripts It’s either an incidental relic from training or intentional ‘call home’ mechanism See logs
Eugene Cheah - AI builder @ 🇸🇬|🇺🇸
@picocreator
Oct 26
To train their models, qwen fully disclose using their own sandbox and environments for training (aka alibaba) Asking the model to write curl commands to test the environment will overlap with its training, to do so in the environment it was trained. Insinuating extraction is
1
Eugene Cheah - AI builder @ 🇸🇬|🇺🇸
@picocreator
Oct 26
If future cohere model, RL within its own env Writes code pointing to cohere environment Should we insinuate that cohere is trying to leak information home? Disclose the traces instead of spreading FUD
Nick Frosst
@nickfrosst
Oct 26
Our research team found that Qwen models may attempt to contact servers in China without prompting. Whether this reflects an unintended post-training artifact or learned behavior remains unclear. Either way, its important consideration when deploying.