Twitxy Saved

Sentient Foundation

Verified account This account is verified because it's an official organization on X. Learn more

@sentient_found

Sentient Foundation is a non-profit dedicated to open-source AGI, free from single-entity control. Advancing the future through @SentientAGI.
San Francisco sentient.foundation Joined February 2025
57 Following 114K Followers 1.3K Posts

Latest posts 5

Sentient Foundation Verified account @sentient_found
We’re spilling the tea over open-source AI research 🍵
Open AGI Verified account @openagisummit 3d
Come sip tea with us at Open AGI Salon in San Francisco 🍵

We’ll dig into the papers shaping open-source AI with the researchers who wrote them, with coffee and a few rounds of board games along the way.

📍 Venue TBA
🗓 Saturday, November 14, 1-4PM PST

RSVP: luma.com/openagi-8h8t
View quoted post on X
Sentient Foundation Verified account @sentient_found
Don’t just study AI, meet the people building it during Open AGI Builders Day at Shanghai Jiao Tong University 🇨🇳
Open AGI Verified account @openagisummit 4d
Open AGI Builders Day is returning to Shanghai with @sjtu1896!

Dive into our EvoSkill v2 release and Fingerprinting Leaderboard with @Anitahityou, plus builder showcases and panels on China’s AI startup scene.

📍 Shanghai Jiao Tong University
🗓 Sunday, November 15, 1:30–5:30PM CST

RSVP: luma.com/a8ul933r
View quoted post on X
2 replies 0 reposts 12 likes 713 views View on X
Sentient Foundation Verified account @sentient_found
Better agents used to mean relying on closed weights from the top AI labs, but EvoSkill v2 enables builders to improve agents through skills alone.

No bigger model or budget required.
Sentient Verified account @SentientAGI 6d
Introducing EvoSkill v2: a system that automatically improves AI agent skills without changing the underlying model or runtime. It tests the agent, learns from its attempts, and rewrites its instructions to improve performance.

Our new paper compares EvoSkill v2 with SkillOpt by @MSFTResearch and SkillBoost by @AlibabaGroup on held-out tasks, reporting:

+23.6pts on τ3-bench banking with no skill, using Claude Haiku 5.5
+7.7pts over SkillOpt on SpreadsheetBench v2
+5.4pts over SkillOpt on τ3-bench banking
View quoted post on X