Sentient's mission is to ensure that Artificial General Intelligence is open-source and not controlled by any single entity.
External link for Sentient Labs
San Francisco, US
Sentient Labs reposted this
Meta open-sources a 30B model small enough for a single consumer GPU, IBM commits $240M to open-source inference at data center scale, and Audrey Tang makes the case for AI that serves the loop of humanity. From your desk to the data center, open source AI is scaling in both directions ↓
Catch Sentient researcher Darshan Tank at ACM SIGKDD & Annual KDD Conference 2026 to learn why LLM agents struggle with multi-tool, long-form analysis. Whether you're a researcher, builder, or working on LLM agents, we hope to see you in Jeju, South Korea 🇰🇷
It feels incredibly good when your work gets recognized. I’m excited to share that our paper “CryptoAnalystBench: Failures in Multi-Tool Long-Form LLM Analysis” has been accepted as a poster at ACM SIGKDD & Annual KDD Conference 2026 — one of the leading conferences for AI, ML, and data mining. 🔗 Paper: https://lnkd.in/gzhxzWi3 The work from Sentient Labs explores how LLM agents struggle with multi-tool, long-form analysis in crypto research, and introduces a benchmark designed to uncover these failure modes. A huge thanks to amazing co-authors for making this possible: Anushri Eswaran, Oleg Golev,Sidhant R., and Himanshu Tyagi. 🙌 📍 Poster #393 📌 ICC 2 🕡 6:45 PM – 8:30 PM If you're attending #KDD2026, come say hi! Would love to connect with researchers, builders, and anyone working on LLM agents, evaluation, benchmarks and agent self-improvement. Looking forward to some great conversations in Jeju! 🇰🇷 #KDD2026 #SIGKDD #LLMAgents #AI #MachineLearning #LLMEvaluation #Benchmarking #Research #AgenticAI #CryptoAI
On the latest episode Open Commons ( ), Taiwan's 1st Minister of Digital Affairs Audrey Tang makes the case that AI should serve the loop of humanity, not the other way around ↓
Audrey Tang's father was a journalist for three decades. After a health scare he started talking to ChatGPT. It got harder and harder to put down. Not his health. Not his sleep. Not his family. Just the subscription. So Audrey and her mother built him his own AI, on a laptop, with one instruction: give him peace of mind, so he spends less time on the screen and more with the people in front of him. That's the whole argument in one story. Audrey was Taiwan's first Digital Minister. She thinks the AI closest to you should answer to you. Full episode of Open Commons ( ): link in comments.
Sentient Labs reposted this
[YAI x OpenAI Hackathon AGENT:24] 24 hours, one rule: build an agent. On August 1–2, YAI (Yonsei Artificial Intelligence) hosted AGENT:24, a 24-hour overnight AI agent hackathon in Seoul, with OpenAI as our Technology Partner. ~83 students. 36 teams. One night to build an autonomous agent from scratch — and then present it with almost no slides. Product explanation had to happen through a live demo, including a surprise task the judges assigned on the spot. No pre-recorded videos, no safety net. Teams competed across three tracks: Creative Agent, Social Impact Agent, and the Liner API Agent track. Congratulations to our winners 🏆 - Overall Winner — 이찬회 - Best Creative Agent — Team J (문성형, 감준서, 김환욱, 백승민) - Best Social Impact Agent — Team Z (오재형, 노예찬) - Liner API Award — 이상원 This event was led end-to-end by Hyunah Ko, who took AGENT:24 from a concept memo in May to about 90 people building agents overnight in August. Huge thanks to Tyler Ryu for the opening keynote and for serving as head judge, to Deeps D. and the OpenAI team for backing this from the concept stage, and to Professor 이시윤 and our judging panel. Thank you as well to INFOCZ(인포시즈) , Liner , and Sentient Labs for supporting the event, and to the Yonsei University 연세대학교 Project for making it possible. And to the 83 people who stayed up all night debugging tool calls at 4am — thank you. Watching 36 agents run live, in front of judges, with nowhere to hide, was the best part of the whole thing. Organized by YAI Yonsei (Yonsei Artificial Intelligence) · Event Lead Hyunah Ko #AGENT24 #YAI #OpenAI #AIAgents #Hackathon #Yonsei
Sentient Labs reposted this
It feels incredibly good when your work gets recognized. I’m excited to share that our paper “CryptoAnalystBench: Failures in Multi-Tool Long-Form LLM Analysis” has been accepted as a poster at ACM SIGKDD & Annual KDD Conference 2026 — one of the leading conferences for AI, ML, and data mining. 🔗 Paper: https://lnkd.in/gzhxzWi3 The work from Sentient Labs explores how LLM agents struggle with multi-tool, long-form analysis in crypto research, and introduces a benchmark designed to uncover these failure modes. A huge thanks to amazing co-authors for making this possible: Anushri Eswaran, Oleg Golev,Sidhant R., and Himanshu Tyagi. 🙌 📍 Poster #393 📌 ICC 2 🕡 6:45 PM – 8:30 PM If you're attending #KDD2026, come say hi! Would love to connect with researchers, builders, and anyone working on LLM agents, evaluation, benchmarks and agent self-improvement. Looking forward to some great conversations in Jeju! 🇰🇷 #KDD2026 #SIGKDD #LLMAgents #AI #MachineLearning #LLMEvaluation #Benchmarking #Research #AgenticAI #CryptoAI
Most AI agents don't just get the math wrong. They get the number from the wrong place: outdated table, mismatched row, wrong year. That's why Minjae Lee built an agent that won't calculate until it knows where the number came from ↓
Hard questions stump every agent, no matter the harness. Sentient researchers Naman Vats and Deep Halder analyzed 13K+ agent runs from the OfficeQA Public Challenge and found harnesses agreed 88-93% of the time on which tasks pass and which fail. TLDR: Difficulty lives in the task, not the harness.
Next stop: Jeju, South Korea 🇰🇷 6 days left to RSVP ↓
6 days left to save your spot! Sentient Foundation invites AI builders and researchers to discuss Sentient Labs' ACM SIGKDD & Annual KDD Conference 2026-accepted paper, "CryptoAnalystBench: Failures in Multi-Tool Long-Form LLM Analysis." 📍The Cliff, Seogwipo-si, Jeju-do 🗓 Thursday, August 13th, 11AM-2PM KST RSVP: https://luma.com/frwp2t3i
One of the best decisions Viktor Solyan made during the OfficeQA Public Challenge was to delete a single line: "You are a financial analyst." Here's how he placed 2nd in the Arena using goose and MiniMax M2.7 ↓
More skills should mean smarter agents right? Sentient researchers Darshan Tank and Baran Nama found the opposite happens: adding new skills can break tasks the agent has already solved. They call this the "Regression Tax", a hidden cost that erased 59% of the gains skills delivered across nearly 6,000 agent runs ↓