lucas g
10.5K posts
lucas g
@DaLucasGonzalez
teaching models to use computers at anthropic. prev: google deepmind, google brain. opinions are my own.
เข้าร่วม Temmuz 2012
556 กำลังติดตาม1.7K ผู้ติดตาม
I joined Anthropic as a member of the technical staff.
Excited to work on frontier modeling at a place with unwavering values and a generational mission.
English
We just hacked macOS to build Moonshot, the first AI agent with its own computer on your Mac!
Moonshot works as a separate macOS user alongside you. 2 users, 1 mac. Join the waitlist for early access at moonshot.computer
comment "MOON" for priority access
English
@Conor_D_Dart @trq212 you can use quick mode with haiku as part of subscription or enable overages to use opus 4.6 fast on quick mode
English
@trq212 Are we going to pay the fast api prices in our Claude code subscription or is this just a fast mode?
English
Claude in Chrome is now significantly faster with our Quick Mode experiment, and you can also pair with Opus 4.6 fast to get blazing speeds
lucas g@DaLucasGonzalez
Watch Claude race itself in our just launched "Quick Mode" on the Anthropic Claude in Chrome Extension.. we made it 3x faster than before!
English
@ibekidkirsch agreed! and for these demos, we went screenshots cause it’s fun to watch. but the model also has a js tool, it’ll show up in the sidepanel UI when it uses it
English
@DaLucasGonzalez Wow! I don’t think people realize what a big unlock faster and reliable browsing is going to be. Can you share anything that you have learned getting this faster. Is it all screenshot or is there dom parsing too?
English
@DaLucasGonzalez How much time this tool saves me…. Love using it.
English
@spamlorde try it with Opus 4.6 in fast mode:
lucas g@DaLucasGonzalez
Head to head race between our new quick mode vs existing latency. Filling a sample permit form:
English
@DaLucasGonzalez More impressive is that you are actually using Haiku
English
@MonstaDomains the main wins are coming from harness optimizations
English
@DaLucasGonzalez that's sick. does quick mode actually help with the latency or is it more just cutting down on the thinking time?
English
Oftentimes, the model will catch itself when it makes a mistake and will correct it. You can see in this example where the model fixes the search filters it set. However this is not always the case and we are rapidly working on improvements. Please share feedback and examples so we can continue to iterate.
English
lucas g รีทวีตแล้ว
Less than a year and a half ago computer use was barely even a thing and now we're near human-level capability.
Another reminder that things are improving very fast.
Claude@claudeai
Sonnet 4.6 also shows a major improvement in computer use skills. Early users are seeing human-level capability on tasks like complex spreadsheets and multi-step web forms.
English
We're at an interesting stage of the evolution where the models are 1000x faster than I am at coding and 10x slower at using a web browser.
English