🚀 Medial Secures Investment on Shark Tank India - Fueling the Future of Professional Social Networking. 🔥
✕
Login
Home
News
Messages
Startup Showcase
Trackers
Premium
Premium Content
Jobs
Notifications
Settings
Try our Valuation Calculator →
Log In
News on Medial
Large Language Models’ Emergent Abilities Are a Mirage
Wired
·
1y ago
Medial
Researchers at Stanford University have published a paper suggesting that the sudden emergence of new abilities in large language models (LLMs) is not as unpredictable as previously thought. They argue that the measurement of LLM performance plays a significant role in determining whether abilities appear to emerge abruptly or gradually. The researchers propose using different metrics that show a gradual improvement in performance as LLMs scale up, challenging the notion of emergent behavior. However, other scientists argue that the work does not fully dispel the idea of emergence and that further research is needed to understand the behavior of LLMs.
View Source
Related News
UAE's G42 launches Jais, an open-source Arabic language AI model
Economic Times
·
1y ago
Medial
A collaborative effort between engineers, researchers, and chip company Cerebras Systems has resulted in the release of an advanced Arabic language software called "Jais." This large language model, with 13 billion parameters, powers generative AI applications and is bilingual, combining Arabic and English data. Named after a peak in the UAE, Jais was trained using Cerebras' supercomputer and will be available through an open-source license. The inclusion of computer code within English data helped enhance the model's reasoning abilities. The project aimed to fill the gap for large bilingual language models.
View Source
Researchers made an IQ test for AI, found they're all pretty stupid
Gizmodo
·
1y ago
Medial
Recent research by Yann LeCun and other scientists challenges the notion of artificial general intelligence (AGI) being achieved anytime soon. The study compared AI's general-purpose reasoning with human capabilities using a series of conceptually simple but challenging questions. The results showed that large language models, including GPT-4, struggled to outperform humans in real-world problem-solving scenarios. LeCun argues that AI systems are far from reaching human-level intelligence, lacking a deep understanding of the physical world and planning abilities.
View Source
How BharatGPT’s Indic-First LLM Is Bridging Language Barriers, Empowering Enterprises
Inc42
·
1y ago
Medial
OpenAI's ChatGPT, powered by generative AI (GenAI), is a major breakthrough in natural language understanding and processing. The intelligent chatbot utilizes large language models to generate content in various formats and languages. McKinsey & Company estimates that GenAI and other technologies could automate the majority of work activities, leading to increased productivity. However, there are concerns about GenAI falling short of community expectations, as demonstrated by the limited accuracy in generating a portrait of Lord Rama. Access to information is a crucial aspect of GenAI's future, as there is a significant gap in availability of non-English content for the majority of the Indian population. OpenAI is continually working to improve the conversational abilities of its GenAI models, with claims that the upcoming GPT-4o will be faster and more natural in its responses.
View Source
Toyota's robots are learning to do housework—By copying humans
Wired
·
1y ago
Medial
Toyota Research Institute is developing robots that can learn household chores by combining machine learning techniques and language models. The robots are trained using a diffusion policy machine learning system, similar to ones used in AI image generators, to determine their actions based on various data sources. Toyota aims to utilize language models to enable robots to learn tasks by watching videos, potentially from platforms like YouTube. The goal is to have robots adapt, improvise, and flexibly handle tasks that require human-like abilities.
View Source
The man who made robots dance now wants them to think for themselves
Wired
·
1y ago
Medial
The Boston Dynamics AI Institute, founded by Marc Raibert, is focused on developing robots that can comprehend and handle complex situations without human assistance. While Boston Dynamics' robots have garnered attention for their physical capabilities, the institute aims to enhance their cognitive abilities. Raibert discusses the institute's mission and its efforts to incorporate language models and reinforcement learning into robotics. The goal is to advance AI beyond large language models and explore the integration of physical and cognitive intelligence. The institute is also working on projects such as bicycle repair robots and dexterous manipulation. Raibert addresses concerns about robots and AI becoming dangerous, emphasizing that most bad things in the world do not come from robots.
View Source
Meta Research Introduces Revolutionary Self-Rewarding Language Models Capable of GPT-4 Level Performance
Medial
·
1y ago
Medial
Language models can achieve superhuman capabilities with the new paradigm of Self-Rewarding Language Models (SRLMs). Meta has developed an innovative system where the model generates its own rewards, leading to continual improvement in instruction following and reward modeling abilities. The iterative Direct Preference Optimization (DPO) framework is used to train SRLMs, allowing the model to push itself to superhuman levels. After just three iterations, the SRLM outperformed state-of-the-art systems, demonstrating its potential to achieve GPT-4 level performance. SRLMs break free from fixed reward models, offering continuous improvement possibilities for language models.
View Source
Google’s new robot AI can fold delicate origami, close zipper bags without damage
Arstechnica
·
4m ago
Medial
Google DeepMind has announced Gemini Robotics and Gemini Robotics-ER, AI models designed to control robots more effectively and delicately. These models build upon Google's Gemini 2.0 and feature "vision-language-action" abilities, processing visual data, understanding language, and executing physical tasks. Apptronik's Apollo robot will utilize Gemini. Google's effort aims to solve the challenge of translating robot knowledge into precise actions. A "Robot Constitution" framework, inspired by Asimov's robot laws, accompanies a new ASIMOV dataset for safety evaluation.
View Source
The fine art of human prompt engineering: How to talk to a person like ChatGPT
Arstechnica
·
1y ago
Medial
This article discusses the concept of human prompt engineering, which involves effectively instructing humans to generate meaningful responses, similar to how AI language models work. By crafting specific prompts, humans can provide context and draw upon their training and knowledge to generate accurate and creative outputs. The article explores how human language models process information, their black-box nature, and their abilities in few-shot and zero-shot learning, as well as common sense reasoning and open-ended text generation tasks.
View Source
‘Haters will say this is AI’: Musk shares video of himself and Trump dancing
Livemint
·
11m ago
Medial
Tech mogul Elon Musk shared an AI-generated video on social media featuring himself and former President Donald Trump dancing to the Bee Gees' "Staying Alive." Musk playfully captioned the video with "Haters will say this is AI." The post gained popularity, showcasing Musk's ability to blend technology and humor. Meanwhile, Musk's AI company, xAI, released its latest models, Grok-2 and Grok-2 Mini, with improved language processing and reasoning abilities. The models are available to xAI Premium users, offering advanced features and faster responses.
View Source
Alibaba launches Qwen3 AI models with hybrid reasoning power
YourStory
·
3m ago
Medial
Alibaba has introduced Qwen3, a suite of large language models designed with hybrid reasoning for AI tasks. Qwen3 models feature a dual-mode approach, enabling both quick responses and in-depth analysis. The series includes eight diverse models, including Mixture-of-Experts models for efficient performance. Qwen3 supports 119 languages and has shown strong benchmark results across various tasks. Open-sourced under Alibaba's initiative, these models are accessible on platforms like Hugging Face and GitHub.
View Source
Trackers
Active Indian VC’s
OG Capital
Email
With a hands-on approach, OG Capital aims to invest in over 20 promising...
Accel Partners
Email
Early and growth-stage investments in disruptive technology companies with...
Blume
Email
Early-stage venture capital firm investing in technology startups in India. Focus on...
Access All Trackers
Startup Showcase Winners
June 2025
Buddy
Helping your parents when you are miles away
BiteStop
The Pit Stop Your Cravings Deserve
Bloomer
The next generation E-commerce platform
Enter Ongoing Startup Showcase
Top Users
Trending News on Medial
Download the medial app to read full posts, comements and news.
Go to Medial App
Not Now
Know everything that’s happening in the startup ecosystem, first.
Enable Notifications?
No, thanks
Count me in