“80% of wealth held by the top 10% (highest since 1939)” – yet no major growth in GFP says a new projection on impacts of AI from @PTetlock and others.
RESEARCH
-
NVIDIA Blackwell Dominates MLPerf Inference v6.0 Token Factory Performance
By
–
AI data centers are token factories.
— NVIDIA AI (@NVIDIAAI) 1 avril 2026
See how NVIDIA extreme co-design maximizes token output to profitably scale AI revenue. https://t.co/aotG4jvuFkAI data centers are token factories. See how NVIDIA extreme co-design maximizes token output to profitably scale AI revenue. NVIDIA Data Center (@NVIDIADC) 📣 MLPerf Inference v6.0 results are in. Learn how systems powered by NVIDIA Blackwell set the pace on inference, delivering the highest AI factory throughput. 🔗 nvda.ws/4sFFA0k — https://nitter.net/NVIDIADC/status/2039359226712097227#m
-
DeepPresenter: AI Framework Masters Dynamic Human-Like Presentations
By
–
Can AI truly master the art of dynamic, human-like presentation creation? Researchers from the Chinese Information Processing Lab and the University of Chinese Academy of Sciences unveil DeepPresenter. This innovative AI framework introduces "environment-grounded reflection,"
-
New scalable method for evaluating 15 LLM AI models published
By
–
The current science of evaluating AI models, such as primarily relying on benchmarks, is far from optimal. @Nature today a new scalable way used to assess 15 LLMs with absolute demand scales, enhancing predictor power and expandability nature.com/articles/s41586-0…
→ View original post on X — @erictopol, 2026-04-01 15:13 UTC
-
DeepReinforce Wins Top Spot in AI Contest
By
–
DeepReinforce took first place in a competitive programming contest, with its AI Agent, outperforming all human contenders. Earlier, the previous best result was achieved by Gemini 3.1, hitting 8th place in February 2026. "GrandCode is a multi-agent AI system powered by
-
NVIDIA Blackwell Dominates MLPerf Inference v6.0 with Record Throughput
By
–
📣 MLPerf Inference v6.0 results are in.
— NVIDIA Data Center (@NVIDIADC) 1 avril 2026
Learn how systems powered by NVIDIA Blackwell set the pace on inference, delivering the highest AI factory throughput.
🔗 https://t.co/Abid9w6wx3 pic.twitter.com/yemqfVgS60📣 MLPerf Inference v6.0 results are in. Learn how systems powered by NVIDIA Blackwell set the pace on inference, delivering the highest AI factory throughput. 🔗 nvda.ws/4sFFA0k [Translated from EN to English]
-
Human DDoS: New Cybersecurity Risk from Autonomous AI Systems
By
–
AI agents are changing cyber defense, but they may also create a new denial-of-service layer: humans 🤯 New blog 👉The Few Humans Left: Risk of Denial Service Attacks on Humans trent.ai/blog/humans-ddos-ai… As autonomous systems scale, the few humans left in the loop inherit only: ☑️ambiguity ☑️edge cases ☑️final judgment calls That means the real bottleneck shifts from infrastructure to cognition. The system doesn’t need to be breached to fail. It just needs to overwhelm human decision-making. Eno Thereska, CEO & Co-Founder of Trent AI, calls this “human DDoS.” A key idea from our RSAC 2026 Cyber Startup Expo panel on machine-speed cyber battle. #AISecurity #CyberSecurity #AgenticAI #RSAC
-
Large Language Models Need Embodiment: New Neuron Perspective
By
–
The need for embodiment for large language models @NeuroCellPress an open-access perspectiive cell.com/neuron/fulltext/S08…
→ View original post on X — @erictopol, 2026-04-01 14:41 UTC
-
Self-Organizing LLM Agents Outperform Predefined Role Hierarchies
By
–
NEW papers on self-organizing LLM Agents. Assign an agent a role, and it'll follow instructions. Let agents figure out roles themselves, and they'll outperform your design. New research tested this across 25,000 tasks with up to 256 agents. The work shows that self-organizing LLM agents spontaneously develop specialized roles without any predefined hierarchy. A sequential coordination protocol outperformed centralized approaches by 14%, agents generated over 5,000 unique roles organically, and open-source models reached 95% of closed-source quality at significantly lower cost. Most multi-agent frameworks today start by defining roles: planner, coder, reviewer, critic. This paper provides large-scale evidence that the opposite approach works better. Give agents a mission, a protocol, and a capable model. The agents will figure out the rest. Paper: arxiv.org/abs/2603.28990 Learn to build effective AI agents in our academy: academy.dair.ai/
-
Achieving 78.9% on OS-world, Outperforming GPT-5.4 at Lower Cost
By
–
Ngl, reaching 78,9% on OS-world and outperforming even GPT-5.4 at 1/10 cost is a big deal