wow. just saw The Economic Times newspaper published an article about me 😃 definitely feels so unreal that Sundar Pichai and Jeff Bezos follows me here. @X is truly a miracle. Forever thankful to all of my followers 🙏🫡

New @AIatMeta paper shows how a vision language model can train its own judge using only synthetic data it generates itself. The vision language model answers image based questions, and the judge model compares two answers for the same question and picks the better one.

🇨🇳 Beijing is making a big bet that embodied AI (AI coupled to physical machines that see, reason, and act) robots in factories and battlefields will decide who leads the next phase of AI power. Chinese leaders frame this as a way to push generative AI out of chat apps and deep

Stuart Russell says an AI with IQ 150 could upgrade itself to 170, then 250, very soon leaving humans way behind. A recent Meta paper also warns self-improving AIis appealing but risky since removing humans from the loop can amplify misalignment. https://t.co/DJbCLP40cq

A solar-powered, autonomous farming robot now plants, weeds, and monitors crops without fuel, wires, or human help. Using AI, GPS, and sensors, it navigates fields, checks plant health, and removes weeds, all off-grid. https://t.co/Z6M3iZqTLx

The paper shows that retrieval helps small vision language models but can actually hurt big ones. It does this using M4-RAG, a benchmark with 80,000 image question pairs across 42 languages and 56 dialects. The core problem is that vision language models are trained on frozen

This paper builds a simple map of all the main ways large language models can cause harm. Using about 200 research papers and incident reports, it groups harms into 5 stages of a model's life. Before the model is released, it highlights issues like scraped personal data,

👨‍💻 Jensen Huang says AI adoption and reshaping of work will be slow. He argues AI hits tasks, so narrow repetitive roles like chopping vegetables vanish first, while mixed roles like radiology keep value through diagnosis and human contact. As physical AI assistants spread, he

💼 Jamie Dimon says AI will cut some jobs but mainly raise productivity, so people could eventually work less. He compares AI to tractors, fertilizers, and vaccines, expects big health and efficiency gains, and wants strict guardrails from regulators. He argues that today’s

This paper connects the computing used to train advanced AI models with the actual physical metals inside the GPUs that run them. Up to 7 tons of toxic metals are tied to training one big GPT4 model run on these chips. The authors ground up an Nvidia A100 AI chip, tested it in

The paper says the best way to manage AI context is to treat everything like a file system. Today, a model's knowledge sits in separate prompts, databases, tools, and logs, so context engineering pulls this into a coherent system. The paper proposes an agentic file system where

This paper shows how to bolt a thinking layer on top of existing image editors so they fix their own mistakes. With that layer, several popular editors suddenly follow user instructions much more accurately on 4 tough real world editing benchmarks. Instruction based image

📡 Fluidstack is lining up about 700 million dollars at a 7 billion dollar valuation, built on Google backed leases for new AI data centers. Situational Awareness, a firm started by ex-OpenAI researcher Leopold Aschenbrenner, is in talks to lead the financing. Fluidstack is a

🦾 Great milestone for AgiBot. They just announced that they have hit 5,000 humanoid units in mass production today. Its 1,742 A series, 1,846 X series, and 1,412 G series units together cover reception, exhibition, entertainment, service, and heavy industrial tasks. All of

Apple is going through its biggest leadership shake-up since Steve Jobs passed away, with more than half a dozen major executives getting ready to leave. There's one glaring problem Apple needs to tackle: its failure to field an AI competitor that matches Google's or

The paper shows Big Tech funded AI papers get more citations, cluster around each other, and focus more on recent work. Only about 10% of papers are Big Tech funded, yet around 12% of those become highly cited. They study about 50K top conference papers and use acknowledgments

📎 Anthropic is wiring Claude Code into Slack When someone tags @Claude on a bug or feature thread, the Slack app sends the request plus conversation and linked repos to Claude Code, then posts investigation steps and candidate patches back. Early users report big time

📈 IBM is reportedly nearing a $11B acquisition of Confluent to strengthen its cloud and data stack, with a deal possibly announced as soon as today. Confluent builds real-time data streaming on top of Apache Kafka so events like payments, app clicks, and sensor logs flow as

Inspirational data point. "Science and medicine doubled life expectancy over the last century."

https://t.co/aG0bYMx3T2