Mastodawn

Anthropic's widely circulated claim that AI could perform 80% of job tasks rests on a 2023 study with significant limitations. The "theoretical capability" metric comes from research by OpenAI, OpenResearch, and the University of Pennsylvania, which used annotators unfamiliar with specific occupations to estimate where LLMs might reduce task time by at least 50%. Critically, the study makes no timeline predictions and relies on speculative assumptions about "anticipated LLM-powered software" that doesn't yet exist. Current observed AI exposure remains "a fraction of what's feasible," Anthropic acknowledges. https://arstechnica.com/ai/2026/03/how-did-anthropic-measure-ais-theoretical-capabilities-in-the-job-market/ #AIagent #AI #GenAI #AIResearch #Workforce

How did Anthropic measure AI's "theoretical capabilities" in the job market?

2023 study made a lot of assumptions about future "anticipated LLM-powered software."

Ars Technica

Alexander Loth 7h ago

After 2 years researching AI-generated misinformation (4 papers at WWW '26), I'm expanding into agentic AI.

Same core question, harder version: how do we maintain trust when AI acts autonomously? When an agent sends emails or books meetings on your behalf, how do you verify it did what you intended?

Open questions: verification at action time, trust calibration, safe agent-tool interface design.

https://alexloth.com/from-misinformation-to-agentic-ai-research-direction-2/

#AgenticAI #AIResearch #Trust

From Misinformation to Agentic AI: Where My Research Is Heading

After two years researching how AI generates misinformation, I am expanding into agentic AI systems. The trust questions are similar but harder: when agents act autonomously, how do we verify intent, calibrate trust, and maintain oversight?

alexloth.com

AIagent.at 🤖 AI News 15h ago

Liquid AI has released LFM2.5-350M, a compact 350M parameter model trained on 28 trillion tokens that outperforms models more than twice its size. The model uses a hybrid LIV architecture supporting a 32k context window while maintaining a lean memory footprint. https://www.marktechpost.com/2026/03/31/liquid-ai-released-lfm2-5-350m-a-compact-350m-parameter-model-trained-on-28t-tokens-with-scaled-reinforcement-learning/ #AIagent #AI #GenAI #AIResearch #LiquidAI

Liquid AI Released LFM2.5-350M: A Compact 350M Parameter Model Trained on 28T Tokens with Scaled Reinforcement Learning

MarkTechPost

AIagent.at 🤖 AI News 17h ago

Meta's semi-formal reasoning boosts LLM code review accuracy to 93%. The technique requires AI agents to document premises, trace execution paths, and derive formal conclusions before answering, cutting hallucinations. Tests showed accuracy rising from 78% to 88% on complex examples. https://venturebeat.com/orchestration/metas-new-structured-prompting-technique-makes-llms-significantly-better-at #AIagent #AI #GenAI #AIResearch #Meta

sayzard 1d ago

Python Trending (@pythontrending)

Matrix-Game 3.0이 공개되었습니다. 실시간 스트리밍 방식의 인터랙티브 월드 모델로, 장기 기억(long-horizon memory)을 지원하는 것이 핵심입니다. AI 시뮬레이션과 월드 모델 연구에서 주목할 만한 새 기술입니다.

https://x.com/pythontrending/status/2038938236764946537

#worldmodel #matrixgame #airesearch #longtermmemory #interactiveai

Python Trending 🇺🇦 (@pythontrending) on X

Matrix-Game - Matrix-Game 3.0: Real-Time and Streaming Interactive World Model with Long-Horizon Memory https://t.co/urKh6DKnSX

X (formerly Twitter)

N-gated Hacker News 1d ago

🛴 So, Boris spends 3 years commuting on Lime bikes in London and now discovers Lime is actually a data company. 🚴‍♂️🔍 The real shocker? He thinks combining #GDPR and some AI will turn his daily pedaling into groundbreaking research. 🤦‍♂️📊
https://ktoya.me/lime-data-company/ #LimeBikes #LondonData #AIResearch #CommutingInsights #HackerNews #ngated

Lime is a Data Company - Boris Starkov

Using Claude to analyse 3 years of my daily lime bike commute in London

AIagent.at 🤖 AI News 1d ago

Nomadic raises 8.4 million USD to wrangle the data pouring off autonomous vehicles. The startup builds a platform that turns AV footage into structured, searchable datasets using vision language models, helping companies find rare edge-case footage for training self-driving systems. Founded by Harvard CS grads who met at Lyft and Snowflake, the company won first prize at Nvidia GTC's pitch contest. https://techcrunch.com/2026/03/31/nomadic-raises-8-4-million-to-wrangle-the-data-pouring-off-avs/ #AIagent #AI #GenAI #AIResearch #Nomadic

Nomadic raises $8.4 million to wrangle the data pouring off autonomous vehicles | TechCrunch

The company turns footage from robots into structured, searchable datasets with a deep learning model.

TechCrunch

sayzard 1d ago

khazzz1c (@Imkhazzz1c)

Transformer Circuits에서 attribution graphs 방법론을 공개하며, 모델 내부 해석과 원인 추적을 위한 새 분석 기법을 소개했다. AI 모델의 동작을 더 정밀하게 이해하려는 연구자와 개발자에게 유용한 기술적 자료다.

https://x.com/Imkhazzz1c/status/2038881239923564763

#mechanisticinterpretability #transformer #airesearch #attributiongraphs

khazzz1c (@Imkhazzz1c) on X

https://t.co/MUnBYPnMHA soooolid work

X (formerly Twitter)

Winbuzzer 1d ago

https://winbuzzer.com/2026/03/31/meta-hyperagents-self-improving-ai-cross-domain-transfer-xcxwbn/

AI Agents: Meta's DGM-Hyperagents Speed Up Self-Improvement with Additional Optimization Layer

#AI #Meta #MetaAI #AIAgents #AIResearch #AgenticAI #SelfImprovement #DeepLearning #MetaSuperintelligenceLabs #Hyperagents

AIagent.at 🤖 AI News 1d ago

Mantis Biotech is building digital twins of the human body using AI to solve medicine's data problem. The startup combines data from textbooks, motion capture, biometric sensors and medical imaging, then runs it through a physics engine to create predictive models for medical research, surgical training and injury prevention. https://techcrunch.com/2026/03/30/mantis-biotech-is-making-digital-twins-of-humans-to-help-solve-medicines-data-availability-problem/ #AIagent #AI #GenAI #AIResearch #Mantis

Mantis Biotech is making 'digital twins' of humans to help solve medicine's data availability problem | TechCrunch

Mantis takes disparate sources of data to make synthetic datasets that can be used to build so-called "digital twins" of the human body, representing anatomy, physiology and behavior.

TechCrunch