i've had enough of all the over-engineering happening during adversarial review loops, causing no-mistakes runs to take a long time and sometimes create scope creep, usually caused by cough... sol.. cough..
so i just spent a whole day tracking down the traces and improved agent instructions in no-mistakes to specifically combat over-engineering
mostly did two things:
1. scope expansion now gets escalated to human review more often
2. auto fixes now attempts solving problems through simplification over adding machinery
in my own private eval set this worked really well!
if you've been using no-mistakes, update to 1.60.2+ now. if not, here it is - https://t.co/6ldO9b8Cxx
A dial is all you need. https://t.co/Isj57Gcggm
The best way to SSH control your computer, terminal and agents from your phone requires no keyboard.
Clean terminal, tappable URLs, built-in voice input, image upload, mosh, color themes, and the signature command dial.
Mistakes != Malice
Thomas Watson was the CEO of IBM from 1914 to 1956. At his death, he was called the world's greatest salesman. From his legacy comes a wonderful example of how to handle mistakes within an organization.
Sometime in the 1940's an IBM…https://t.co/JYZl2d9w7P
Avoid:
1. eating candy bars
2. watching the news
3. deploying code with downtime
Before:
1. running a marathon
2. learning to meditate
3. migrating to kubernetes
Avoid stupidity before seeking brilliance.
What are some things you should be avoiding?…https://t.co/QoW3rcaWF6