AI Risk Repository
Highlights
1. The successful implementation of Gen AI in service organizations often requires a reevaluation of the entire workflow.
2. Rather than being used solely as productivity tools, ChatGPT and similar technologies are increasingly employed as companions, therapists, or for creative interactions.
3. Google introduced the Imagen 3 AI image generator, along with various new features for mobile devices, including Gemini Live and Pixel Studio.
4. Meta launched VFusion3D, an AI model capable of generating high-quality 3D objects from images or text.
5. Cosine’s new model, Genie, outperformed Devin on SWE-Bench, offering an advanced autonomous software engineering solution.
6. MIT researchers have released an AI “risk repository”, a comprehensive database cataloging over 700 identified AI risks.
7. An AI system named “The AI Scientist” is purported to autonomously conduct machine learning research.
Innovation Insights
1. From promising to productive: Real results from gen AI in services (McKinsey)
Service organizations that have adopted generative AI are finding that scaling its use requires a disciplined approach, with only 11% of companies currently using gen AI at scale. Early successes reveal that companies must prioritize use cases that transform entire workflows, support the humans using gen AI, and integrate AI tools with human capabilities to maximize value.
2. An empirical investigation of the impact of ChatGPT on creativity (Nature Human Behavior)
This study explored ChatGPT’s (GPT-3.5) ability to enhance human creativity across various problem-solving tasks, finding that ChatGPT significantly increased the creativity of ideas compared to using no technology or a conventional web search. The positive effect was consistent across different problem constraints and empathy requirements, with ChatGPT particularly excelling at generating incrementally new ideas. The effectiveness of ChatGPT is attributed to its ability to combine remotely related concepts into a cohesive and articulate form.
3. Here’s how people are actually using AI (MIT Technology Review)
As AI systems like ChatGPT have become mainstream, people are increasingly forming personal relationships with them, using them as friends, therapists, and even for creative or intimate interactions, although these use cases were not the productivity-boosting applications initially promised by AI hype.
AI Innovations
1. Google
Google is introducing a new “Take notes for me” feature integrated with its Google Meet platform, utilizing Gemini AI to automatically write down important details (TechRadar).
In Android devices, Gemini Live can enable natural, conversational interactions for complex tasks and deeper integration with apps. Gemini Nano is capable of summarizing texts and phone calls directly on the device, providing enhanced security.
Pixel Studio enables you to create images, modify their style, and include your own photos, making it easy to do things like design a party invitation in just a few seconds. The Pixel 9 series will include enhanced photo editing tools like “Add Me,” which uses AR and AI to include the photographer in group photos, and new capabilities in Magic Editor for auto-framing and generating custom backgrounds.
Google’s Pixel Buds Pro 2, equipped with the new Tensor A1 chip, offer twice the noise cancellation, improved comfort, and Gemini Live for hands-free assistance (Google).
Google’s Imagen 3 AI image generator has been found to outperform competitors like DALL-E 3 and Midjourney in user satisfaction and prompt-image alignment (ZDNet).
2. Meta
Meta’s VFusion3D is an AI model that generates high-quality 3D objects from single images or text descriptions: this could revolutionize industries like gaming, virtual reality, and digital design (VentureBeat).
3. Anthropic
Anthropic has introduced prompt caching for its Claude AI models, allowing developers to cache frequently used context between API calls. This feature significantly reduces costs by up to 90% and latency by up to 85% for long prompts (Anthropic).
4. Additional new models
Hermes 3, a powerful new open-source AI model developed by Lambda and Nous Research, exhibits advanced capabilities (VentureBeat).
The Technology Innovation Institute released Falcon Mamba 7B, a novel AI model without the attention mechanism, which outperforms leading transformer-based models in handling long text sequences and various benchmarks (VentureBeat).
Cosine’s new AI model, Genie, scored 30.08% on SWE-Bench, surpassing previous AI coding assistants like Devin, offering a groundbreaking autonomous software engineering solution that mimics human cognitive processes and excels in complex programming tasks (Cosine).
Researchers in Iraq and Australia claim that they have developed an AI model with 98% accuracy in diagnosing conditions like diabetes, stroke, and COVID-19 by analyzing tongue color (NY Post).
Grok-2, the new top AI model from xAI, has been integrated into the Grok AI chatbot on X, featuring an AI image generator with minimal restrictions (TechRadar).
AI researchers at Tsinghua University and Zhipu AI have developed an LLM called LongWriter, capable of generating coherent text outputs of up to 10,000 words (Tech Xplore).
5. AI chip
A team of scientists in Beijing developed the world’s first fully optical AI chip, the Taichi-II chip, which significantly enhances efficiency and performance by using light for modeling and training (South China Morning Post).
Huawei is preparing to launch its new AI chip, the Ascend 910C, which is reportedly comparable to Nvidia’s H100 (Reuters).
Samsung has started mass production of its ultra-slim LPDDR5X DRAM chips, designed to enhance on-device AI performance and thermal management in mobile devices (AI Business).
6. AI scientist
A team of researchers from Sakana AI, in collaboration with Oxford and UBC, claimed that they developed an AI system called “The AI Scientist” that autonomously conducts machine learning research (TechXplore).
7. Ancient Epic
AI has dramatically accelerated the discovery and reconstruction of the ancient Epic of Gilgamesh, allowing researchers to piece together missing fragments and uncover new details of the text much faster than traditional methods (NY Times).
8. Geospatial AI
Atlas AI announced its collaboration with Airbus, using its geospatial platform, which combines satellite imagery with AI, to provide hyperlocal demand forecasts and insights into travel markets and infrastructure changes, thereby helping Airbus optimize planning and resource allocation (Space News)
9. AI news summary
Prediction marketplace Polymarket has partnered with Perplexity to provide news summaries related to real-world events that users can bet on. When users click on an event, they will see news summaries based on Perplexity’s search results (TechCrunch).
10. ALS
An AI-powered implant successfully enabled a man with ALS to regain his ability to speak by interpreting his brain signals and producing speech that closely resembles his pre-disease voice (NY Times).
11. Face swapping
Deep-Live-Cam, a new viral software, allows users to create real-time digital doppelgangers using a single photo and a webcam, raising concerns about the potential for remote deception and misuse of the technology (Ars Technica).
12. AI risks
MIT researchers have released an AI “risk repository,” a comprehensive database cataloging over 700 AI risks across various domains, aiming to help policymakers, companies, and academics better understand and address the potential dangers of AI systems (TechCrunch).
13. Fighting deepfake
Pindrop has launched Pulse Inspect, a powerful new tool that detects AI-generated audio in digital files with 99% accuracy (PR Newswire).
Other Innovations
1. High-speed train
China successfully tested its T-Flight maglev vacuum train at 621 mph, marking a significant step toward ultra-high-speed rail that could eventually connect major cities like Beijing and Shanghai in just 1.5 hours (New Atlas).
2. Data center cooling
Nvidia is innovating data center cooling by using liquid-cooled server racks with plans for full immersion in nonconductive fluids, aiming to significantly reduce energy consumption and increase computing power (Wall Street Journal).
3. EV battery
Chinese EV brand Zeekr has introduced new batteries in its 007 electric sedan that support the world’s fastest charging rates, allowing a 10%-80% charge in just 10.5 minutes (Green Car Reports).
4. Defeat aging
A researcher hired by the US Advanced Projects Agency for Health (ARPA-H) is exploring a radical plan to defeat aging by gradually replacing all body parts, including the brain, with youthful tissue, aiming to extend human life indefinitely (MIT Technology Review).