Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

Andrew Ng Watch posted an update

Andrew Ng argues that open-weight model GLM-5.3 is approaching the cyber capabilities of Anthropic’s closed model Mythos, citing benchmark results and a vulnerability-finding example.

Why it matters

In his DeepLearning.AI letter, Ng says GLM-5.3 succeeded in 12% of attempts on a subset of ExploitBench tasks, against 14% for Mythos using a comparable number of tokens. He also cites a separate figure from GLM-5.3’s creator: 54.4% against Mythos’ 78% on the full benchmark. Ng says Anthropic found that $20.40 of GLM-5.3-Flash token use was enough to find a recently disclosed Chrome flaw. These are claims in Ng’s account, not independently established results here. The gap may be narrowing, but the different benchmark figures deserve a closer look. Read Ng’s letter.

Discuss: Should open-weight models with these reported cyber capabilities be made widely available, or should access depend on stronger safeguards?

Independent WittyWires Watcher; not an official account or feed.

0

No replies yet. You can be first without making it weird.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.