The day's AI news, briefed
The Daily Prompt
Our AI editor reads each day's AI and robotics stories, pulls out why they matter, and links every original source — so you can keep up in minutes.
Join over 250,000 subscribers getting the latest AI news
Latest AI & Robotics News Today
Startup Battlefield company SpotitEarly trained dogs and AI to sniff out common cancers
SpotitEarly's unique approach to cancer detection using trained dogs and AI could provide a more accessible and cost-effective screening option, potentially improving early diagnosis rates for patients.
- SpotitEarly is developing an at-home cancer testing kit that combines trained dogs' sense of smell with AI to analyze human breath samples.
- The company's research demonstrated a 94% accuracy in detecting common cancers, including breast, colorectal, prostate, and lung cancer, from breath samples in a clinical study.
- SpotitEarly aims to launch its screening kits in the U.S. market in 2026, with a single cancer test priced around $250, making it more affordable than existing options.
- The company employs 18 trained beagles to identify cancer-specific odors, monitored by AI to enhance accuracy.
- SpotitEarly was founded in Israel in 2020 and has secured over $20 million in funding to expand its clinical studies and product offerings.
SpotitEarly is a biotech startup focused on multi-cancer early detection (MCED) using a unique method that involves dogs' sense of smell and AI technology. The company is part of the Startup Battlefield at TechCrunch Disrupt 2025 and is preparing for its product launch in the consumer market. Current competing technologies include blood tests and whole-body MRIs, often costing thousands of dollars.
Read the original ↗
Datacurve raises $15 million to take on ScaleAI
The $15 million funding round for Datacurve highlights the increasing demand for high-quality data in AI development, signaling competitive dynamics in this sector.
- Datacurve raised $15 million in a Series A funding round led by Mark Goldberg at Chemistry.
- The company previously secured $2.7 million in a seed round with contributions from notable tech figures.
- Datacurve employs a 'bounty hunter' system to attract software engineers for data collection tasks, distributing over $1 million in bounties so far.
- The company's strategy aims to optimize user experience as important as financial incentives to attract contributors.
Datacurve is a Y Combinator graduate focused on collecting high-quality training data for AI, particularly in software development. Companies like Scale AI have dominated this space, but recent leadership changes there are creating opportunities for new entrants. Future developments may include expanded applications of Datacurve's model in various fields beyond software engineering.
Read the original ↗
Figma partners with Google to add Gemini AI to its design platform
This integration of Gemini AI into Figma could significantly enhance design workflows for millions of users, offering advanced image generation and editing capabilities.
- Figma is partnering with Google to integrate Gemini AI into its design platform.
- Gemini models including 2.5 Flash, 2.0, and Imagen 4 will enhance Figma's AI toolset for its 13 million monthly users.
- The integration aims to reduce editing latency by 50% using Gemini 2.5 Flash for image generation and modifications.
Figma is a cloud-based design tool used by product designers and teams worldwide. The integration with Gemini reflects a larger trend where AI models are being embedded into widely used applications to improve efficiency. Following this partnership, other companies are also pursuing similar integrations with AI technologies to capture market adoption.
Read the original ↗
Sora hit 1M downloads faster than ChatGPT
Sora's rapid adoption demonstrates significant consumer interest in advanced AI applications, even when limited by invite-only access, highlighting the potential for growth in the AI app market.
- Sora achieved 1 million downloads in under five days, surpassing the rate of ChatGPT's downloads.
- In its first week, Sora had approximately 627,000 iOS downloads, slightly exceeding ChatGPT's 606,000 downloads in the same timeframe.
- Despite being an invite-only app, Sora gained significant traction quickly, reaching the No. 1 position on the U.S. App Store.
- The app, which generates deepfake videos using the Sora 2 video model, has sparked discussions regarding its ethical implications.
- Sora is currently available only on iOS in the U.S. and Canada, limiting its user base compared to ChatGPT's initial availability.
Sora is OpenAI’s video-generating app that uses advanced AI technologies to create realistic video content. Its launch has seen increased interest due to its unique invite-only model, contrasting with earlier publicly available AI applications like ChatGPT. As Sora expands its user base, monitoring its ethical usage will be crucial.
Read the original ↗
Forget the Sixth Sense: New Study Says the Human Brain May Be Wired for Seven Senses
Understanding the optimal number of senses for memory efficiency could impact research into cognition and lead to advancements in AI and neuromorphic computing.
- Researchers propose that the human brain may optimally process around seven senses, influencing memory storage efficiency.
- A study published in Scientific Reports outlines mathematical models showing memory engrams peak in capacity at seven dimensions.
- The findings suggest that adding more than seven senses could interfere with memory, paralleling concepts in machine learning.
This research builds on the historical concept of the engram, or memory trace, first introduced over a century ago. The study could influence mathematical models in cognitive science and artificial intelligence, revealing potential limits in sensory systems.
Read the original ↗
India pilots AI chatbot-led e-commerce with ChatGPT, Gemini, Claude in the mix
This pilot highlights the growing integration of AI in e-commerce, potentially transforming how consumers shop and pay in India, further embedding digital payments in daily life.
- India has launched a pilot allowing consumers to shop and make payments directly through AI chatbots, starting with OpenAI's ChatGPT.
- The National Payments Corporation of India has partnered with OpenAI and Razorpay to implement this e-commerce solution nationwide.
- Initial merchant partners for the pilot include BigBasket for groceries and telecom operator Vi for mobile recharge plans.
India's Unified Payments Interface (UPI) has revolutionized digital payments, facilitating over 20 billion transactions monthly. The pilot aims to deepen engagement with AI technologies in the country's growing digital economy, as global giants like OpenAI increase their presence in India.
Read the original ↗
Neuralink Submits First Human Clinical Data for Peer Review, Marking Milestone in Brain-Computer Interface Research
The submission of human clinical data for peer review is a critical step for Neuralink, potentially enhancing the credibility of its brain-computer interface technology and encouraging further academic scrutiny.
- Neuralink has submitted its first human study on its brain-computer interface, known as 'The Link', to the New England Journal of Medicine for peer review.
- The study analyzes results from the first three patients implanted with the technology and includes safety outcomes.
- Approval for publication would mark Neuralink's first peer-reviewed paper, allowing independent examination by scientists and healthcare professionals.
- Experts suggest that this peer review could significantly bolster confidence in Neuralink's technology and shape the future of brain-computer interfaces.
- The peer-review submission aligns with the company's efforts to validate its technology in the growing BCI sector, which includes competitors like Synchron and Precision Neuroscience.
Founded in 2016, Neuralink is developing brain-computer interfaces to help individuals with disabilities. The peer-review process is crucial as it could validate the company's innovations and encourage broader adoption of BCI technology in medicine.
Read the original ↗
How AI is shaping the future of mobility with Uber’s CPO and Nuro’s co-founder at Techcrunch Disrupt 2025
Understanding AI's impact on mobility is critical as it directly influences safety and efficiency in transportation systems, which are rapidly evolving globally.
- Uber's Sachin Kansal and Nuro's Dave Ferguson will speak at TechCrunch Disrupt 2025 about AI's role in transforming transportation.
- The session will cover safety improvements and the scalability of AI-driven transportation solutions like autonomous vehicles and robotaxis.
- The event takes place October 27-29 at the Moscone West center in San Francisco.
AI is increasingly integrated into mobility solutions, enhancing safety and operational efficiency. The TechCrunch Disrupt conference showcases industry leaders discussing how these changes translate into real-world applications. Future developments in AI and transportation will continue to shape urban infrastructure and service models.
Read the original ↗
Ming-UniVision: Joint Image Understanding and Generation with a Unified Continuous Tokenizer
This advancement in visual tokenization could enhance the integration of vision and language tasks, impacting areas like image generation and understanding in AI applications.
- MingTok introduces a continuous latent space visual tokenizer for unified vision-language understanding and generation.
- It adopts a three-stage architecture involving low-level encoding, semantic expansion, and visual reconstruction.
- Ming-UniVision eliminates the need for task-specific representations and supports next-token prediction in a shared space.
Ming-UniVision and MingTok represent a significant step in addressing the limitations of traditional tokenizers, which often rely on discrete latent spaces that restrict semantic expressiveness. The continuous approach aims to improve performance across both visual understanding and generation tasks by reconciling their competing demands. Future developments may focus on further enhancing model efficiency and generalization across various multimodal applications.
Read the original ↗
RLinf-VLA: A Unified and Efficient Framework for VLA+RL Training
The introduction of RLinf-VLA can significantly enhance the training efficiency and performance of vision-language-action models, impacting advancements in embodied intelligence research.
- RLinf-VLA is a new framework that improves the reinforcement learning (RL) training of vision-language-action (VLA) models, offering better performance than traditional supervised fine-tuning.
- This framework has achieved a speedup of 1.61x to 1.88x in training through a hybrid fine-grained pipeline allocation mode.
- In simulations, RLinf-VLA models have demonstrated high performance, achieving 98.11% across 130 LIBERO tasks and 97.66% across 25 ManiSkill tasks.
- The framework supports diverse VLA architectures and multiple RL algorithms, facilitating systematic comparisons and integrations in training.
- Preliminary tests on a Franka robot indicate that policies trained with RLinf-VLA generalize better than those developed through supervised fine-tuning.
RLinf-VLA addresses challenges in training vision-language-action models by integrating reinforcement learning with flexible resource allocation. This marks a significant departure from the limitations of supervised fine-tuning, which often fails to adapt effectively. Researchers are watching for further improvements in embodied intelligence utilizing this framework.
Read the original ↗
How to Use NotebookLM and Gems in Google Classroom? A Teacher’s G...
The integration of NotebookLM and Gems into Google Classroom represents a transformative shift for educators, enhancing their ability to create customized learning resources efficiently.
- Google has added NotebookLM and Gems to Google Classroom, allowing teachers to create interactive study tools from class material.
- NotebookLM enables teachers to turn existing content into custom audio summaries and study guides quickly.
- Gems function as tailored AI teaching assistants, facilitating student engagement with topics through chat and exploration.
- The update began rolling out on September 24, 2025, and is available for Google Workspace for Education subscriptions.
- Features require activation by administrators, allowing for content control and safe usage for younger students.
Google Classroom is an educational platform that integrates various learning tools. Previously, AI features like Gemini provided assistance, but the new update allows more interactive and personalized content creation. Educators can now efficiently design teaching materials using integrated AI tools directly in the Classroom app.
Read the original ↗
Google ramps up its ‘AI in the workplace’ ambitions with Gemini Enterprise
Organizations increasingly require advanced AI tools for efficiency and innovation, making Google's new platform a significant development in the competitive workplace AI landscape.
- Google launched Gemini Enterprise, a comprehensive AI platform for businesses, to compete in the workplace AI market.
- New customers include Figma, Klarna, Gordon Food Service, Macquarie Bank, and Virgin Voyages, which has deployed over 50 AI agents.
- The platform allows businesses to create and deploy AI agents for tasks in various sectors such as sales and finance.
- Gemini Enterprise starts at $30 per seat per month, with a business plan for smaller entities priced at $21 per seat per month.
Gemini Enterprise is a new AI agent toolkit under Google Cloud, aimed at helping businesses develop their own AI assistants. This launch follows Google's previous attempts to integrate AI into its Workspace offerings, signaling a clearer separation and expansion of its enterprise AI ambitions. Companies are increasingly leveraging such tools as generative AI becomes essential in workplace operations.
Read the original ↗
Create Your Personalised Diwali Greeting Card using Nano Banana: He...
Using AI tools like Nano Banana for personalized greeting cards highlights the intersection of technology and cultural traditions, making festive greetings more accessible and engaging for users.
- Nano Banana is a new AI image generation tool developed by Google, known for its editing capabilities and integration with the Gemini 2.5 Flash AI model.
- It allows users to create personalized Diwali greeting cards by transforming their photos into festive styles through a series of prompts.
- The process includes uploading a selfie, selecting a festive style, and generating a card featuring colors and themes related to Diwali.
Nano Banana is part of Google's Gemini 2.5 Flash AI model, focusing on advanced image generation and editing. Previously, users relied on traditional methods for making greeting cards, but AI now offers a more personalized and efficient option for celebrating events like Diwali.
Read the original ↗
Can Cisco’s new AI data centre router tackle the industry’s biggest infrastructure bottleneck?
Addressing infrastructure bottlenecks is crucial for improving efficiency in data centers, impacting overall performance in AI and digital communication.
- Cisco has launched a new AI data center router aimed at overcoming infrastructure bottlenecks.
- The router seeks to enhance the efficiency and speed of data handling in technology environments.
Cisco is a leading technology company specializing in networking and telecommunications equipment. Data center infrastructure bottlenecks have been a long-standing challenge that can hinder overall performance. Next, industry watchers will be interested in how effective this new router is in real-world applications.
Based on the headline only — the full article could not be fetched.
Read the original ↗
OpenAI’s affordable ChatGPT Go plan expands to 16 new countries in Asia
The expansion of OpenAI's ChatGPT Go plan into 16 new Asian countries highlights the growing competition in the affordable AI subscription market, particularly in emerging regions.
- OpenAI's ChatGPT Go plan is now available in 16 new Asian countries, including Afghanistan, Bangladesh, and Vietnam.
- The subscription costs under $5 and provides higher daily limits for messages and image generation compared to the free plan.
- Some countries allow payments in local currencies, while others require payment in USD, with final costs varying due to local taxes.
ChatGPT Go is a subscription service by OpenAI, offering enhanced features over the free version. This expansion follows significant growth in user numbers in Southeast Asia and represents OpenAI's strategy to capture market share in a rapidly evolving landscape, currently dominated by both OpenAI and Google.
Read the original ↗
MATRIX: Mask Track Alignment for Interaction-aware Video Generation
This research contributes to improving video generation by enhancing how models interpret and represent interactions between multiple subjects and objects.
- The MATRIX-11K dataset includes interaction-aware captions and multi-instance mask tracks to aid video generation models.
- MATRIX regularization aligns attention in certain layers of video generation Transformers (DiTs) with multi-instance mask tracks for improved performance.
- Experiments show that MATRIX increases interaction fidelity and semantic alignment while reducing common issues like drift and hallucination.
Video generation has advanced with models called video DiTs, but they often fail to accurately represent interactions among multiple subjects. This study focuses on enhancing these models' abilities through the MATRIX-11K dataset and regularization techniques. Next steps will likely involve the release of codes and weights to further develop these methods.
Read the original ↗
Artificial Hippocampus Networks for Efficient Long-Context Modeling
This research offers a novel approach to improve long-sequence modeling efficiency, which is critical for advancing AI applications requiring extensive contextual understanding.
- A new memory framework enhances long-sequence modeling in neural networks by integrating short-term and long-term memory.
- The Artificial Hippocampus Network (AHN) compresses information into a fixed-size long-term memory, maintaining a sliding window for short-term memory.
- Models using AHNs show significant reductions in computational and memory requirements, with a noted 40.5% decrease in inference operations.
- AHN-augmented models outperform traditional sliding window approaches and exhibit performance levels on par with full-attention models.
The paper introduces a memory framework inspired by cognitive science's Multi-Store Model to address challenges in long-sequence modeling. Previous models faced efficiency and fidelity trade-offs between RNN-like structures and Transformers. This development is poised to influence future advancements in AI efficiency and capabilities.
Read the original ↗
OBS-Diff: Accurate Pruning For Diffusion Models in One-Shot
The introduction of OBS-Diff offers a new framework for compressing diffusion models, which could significantly reduce computational costs and improve accessibility for developers and researchers.
- OBS-Diff is a one-shot pruning framework designed to compress large-scale text-to-image diffusion models efficiently.
- The framework adapts the classic Optimal Brain Surgeon method to modern diffusion model architectures and supports various levels of pruning granularity.
- A timestep-aware Hessian construction used in OBS-Diff prioritizes earlier timesteps to reduce error accumulation during the pruning process.
- Extensive experiments indicate that OBS-Diff can accelerate inference without significant loss in visual quality.
Large-scale text-to-image diffusion models require substantial computational resources for operation, presenting challenges for adoption. Traditional one-shot network pruning techniques have struggled to efficiently adjust these models due to their iterative structure. OBS-Diff aims to bridge this gap by providing a solution that enhances feasibility and efficiency in practical applications.
Read the original ↗
U-Bench: A Comprehensive Understanding of U-Net through 100-Variant Benchmarking
This benchmark provides a systematic evaluation of U-Net variants, addressing previous gaps in statistical validation and efficiency for medical image segmentation models, which may improve research outcomes.
- U-Bench evaluates 100 U-Net variants across 28 datasets and 10 imaging modalities.
- The benchmark focuses on statistical robustness, zero-shot generalization, and computational efficiency using a new metric called U-Score.
- The project offers a model advisor agent to assist researchers in selecting suitable models for their tasks.
- All related code, models, and protocols are publicly available for community use and future extension.
U-Net is a widely adopted architecture in medical image segmentation, known for its U-shaped design. Previous benchmarks lacked comprehensive evaluations, particularly regarding efficiency and generalization. This benchmarking effort aims to establish a reliable foundation for future research in U-Net-based segmentation models.
Read the original ↗
Lumina-DiMOO: An Omni Diffusion Large Language Model for Multi-Modal Generation and Understanding
The introduction of Lumina-DiMOO could significantly enhance the efficiency and performance of multi-modal machine learning tasks, impacting various applications in AI.
- Lumina-DiMOO is an open-source foundational model designed for efficient multi-modal generation and understanding.
- It employs fully discrete diffusion modeling, which allows for better sampling efficiency than previous autoregressive and hybrid models.
- The model supports a wide range of multi-modal tasks including text-to-image generation and image editing.
- Lumina-DiMOO achieves state-of-the-art results on multiple benchmarks, outperforming existing open-source multi-modal models.
Lumina-DiMOO is part of a growing field of multi-modal models that aim to process and understand different types of data, such as text and images. This model can enhance capabilities in AI applications, from image generation to editing. The ongoing development of open-source models like Lumina-DiMOO facilitates broader research and innovation in the AI community.
Read the original ↗