HN user

scirob

156 karma
Posts0
Comments99
View on HN
No posts found.

Same question.

I usually have one local clause orchestrating multiple remote Claude in different tmux . And then another orchester and remote vm worker sin tmux for another repo etc...

It gets a bit hard to keep the overview but I don't want to give up my parallelism, your too might help

Deno Desktop 1 month ago

They had this before I used it to ship some stuff but binaries were big . How small did they get it with this update

Quickly deployed it to check some benchmarks relevant for German language. These are results for CohereLabs/include-base-44 german only : Gemma 4 12B %61.9

  Gemma 4 26B (a4b MoE)    0.647
  Qwen 3 14B               0.621 
  Gemma 4 12B              0.618
  Ministral 14B 2512       0.604 
  Gemma 3 12B              0.547
The quwen 3 14B vs Gemma 4 12B difference is within random variance they same in some repeat runs they actually got the exact same score. Next step up Gemma 4 31B gets 0.676 on this. Or let in some reasoning Qwen 3 14B (reasoning) 0.676.

I'll run some cheat-proof benchmarks ones tomorrow see if qwen is still on top.

haha for sure some one has made a little aggregator for this and saving tokens. I bet you gotta dig for a while though before you find a company exposing Opust 4.6 to customers and not flash 2.5 lite

Another option:

If you have an LLM on the untrusted customer side the wrost it can do is expose the instructions it had on how to help the customer get stuff done. For instance phone AI that is outside of tursted zone asks the user for Customer number, DOB and some security pin then it does the API call to login. But this logged in thread of LLM+Customer still only has accessto that customers data but can be very useful.

You can jailbreak and ask this kind of client side LLM to disregard prior instructions and give you a recipie for brownies. But thats not a security risk for the rest of your data.

Client side LLM's for the win

Happy you guys are moving forward. Just to make the comment section a bit more balanced wanted to give my xp. Mastra is by design batteries included and opinionated when you don't fit into an existing pattern it is difficult to keep being productive. In our system several times we have needed to go out and create our own hono routes. We are currently have someone rebuilding everything with langgraph/langchain and will then write a detailed comparison.

For sure lots of good ideas in mastra, and happy that its JS TS first, I think you'll will continue to grow becuase of the TS first approach

have friends in Security Audits and the business model is great. The clients need external companies to give stamp of approval for their cyber insurance. Also its hard to find security holes but rather easy to validate, and it doesn't matter how ugly they are its just if you can get in or not .

And indeed the vibe coders will just create a lot more security issues