Twitxy Saved

Samim

Verified account This account is verified. Learn more

@PawnToPromotion

Building things that matter.👨🏻‍💻🚀🔁 | PhD in CS
Dimension C-137 Joined March 2013
176 Following 7.9K Followers 3.7K Posts

Latest posts 5

Pinned post

Samim Verified account @PawnToPromotion
چند وقتی که کمتر روی توییتر فعال بودم روی کیفیت دوبله livdub کار میکردم: لحن گفتار، صدای دوبله، صداهای اصلی ویدیو و ...
خوشحال میشم نظرتون رو بشنوم! همینطور اگر ودیوی چالشی دارید برای تست در ریپلای بگید :)
3:15
8 replies 6 reposts 71 likes 2.9K views View on X
Samim Verified account @PawnToPromotion
It seems like everyone is making their own ways to have a sandbox for AI‌ agents. I see people using bubblewrap script, gVisor with an egress proxy, a Linux user per agent, an LXC per agent on Proxmox, or Docker "sandboxes".

IMO‌ whatever wins more people over won't be the most secure one. It'll be the one you don't have to think about.

news.ycombinator.com/item?id=503661…
Samim Verified account @PawnToPromotion
Hightouch says a cheap model that's allowed to ask Opus for help matched Opus accuracy at about a quarter of the cost. One task, their numbers.

Funny thing in the HN comments: half the people run it that way, and half run it backwards, big model as the boss and cheap ones doing the work. Nobody has actually measured which is better.

hightouch.com/blog/llm-judge…
hightouch.com LLM Judges for Identity Resolution at 4x Lower Cost | Hightouch We tested Jev and GPT-6 Luna as judges for identity resolution. A hybrid of Jev and Claude Opus matched frontier accuracy at over 4x lower cost.