Banger paper from Google.
If you maintain a skill library for your agents, you might want to check this out.
(bookmark it)
This work separates three things that skill-evolution systems usually collapse into one. Raw execution traces, a persistent wiki of accumulated knowledge, and the executable skills themselves.
Experience gets consolidated into the wiki, and every later skill update builds on that wiki instead of on a scattered optimization history.
Ablations confirm the wiki is what carries a lot of the gain. Two results stand out in particular. Smaller models with evolved skills beat substantially larger models without them. And skills evolved by one model transfer across families, where skills evolved elsewhere sometimes beat self-evolved ones.
Paper: https://t.co/6qftGirTpE
Chat with Paper: https://t.co/rrVzkkR1ij
Tasks AI is good at:
* Generating JavaScript and Python code
* Funny/somewhat interesting pictures
* Summaries / transcripts
* Handling stupid busywork
* Overcoming inertia and getting you to start
Tasks AI is bad at:
* Exercising judgement
* Providing a unique perspective
Upon paid evaluation, GPT4 is still the GOAT for my job. Thanks for all the sarcastic GH PR descriptions, stupid image macros, and interview candidate feedback summaries.
Comparisons:
* Gemini is just not as good
* Claude makes up stuff
* Meta doesn't know how to write
A new way for companies to quantify the business impact of their process improvement programs.
This is the Celonis Transformation Hub.
Learn more: https://t.co/IJRia52iHF
Does anyone else enjoy working & playing from power-limited computers? I spend a lot of time on my tiny Chromebook vs. my powerful desktop. It's always exciting to be pushing the limits and seeing what this thing can do.
having a baby has taught me that big head mode in NBA Jam is completely unrealistic; in reality the players would just fall over on the court, unable to hold up their giant heads
@hurrycurrytokyo I'm so sorry. Google records me having visited your fine establishment at least 42 times since 2008 (probably many more). The world be worse without Hurry Curry in it.
These photos via @stltoday are quite something: Voters infected with Covid-19 cast their ballots with the help of election staffers covered in PPE.
Full gallery: https://t.co/UTN8sIQZb1
App developers should take into account the negative ill will generated by notifications that interrupt a user for trivial matters who might be waiting for important news (e.g. job, school, health issues).