•
October 17, 2024
•
5 minutes
Hello Niuralogists!
Welcome to this week’s edition, where we dive into the latest advancements in artificial intelligence. We’ll explore how these innovations are shaping various aspects of our lives—from the workplace and business to policies and personal experiences. This issue highlights some fascinating updates, including Amazon's content creation tools, Adobe's audio creation prototype, Anthropic's AI safety measures, US AI chip export restrictions, and Google's AI-driven shopping platform, and more.
For more in-depth coverage, keep reading…
The New York Times has issued a cease-and-desist notice to Perplexity, a startup using generative AI, demanding it stop using the newspaper’s content without authorization. This move underscores ongoing concerns from publishers about AI systems exploiting their content for training and generating summaries. Despite assurances from Perplexity to halt "crawling" practices, NYT claims its articles are still being used, prompting legal action. Perplexity stated that it does not scrape data for model training but indexes web pages to provide factual citations. The startup plans to respond to NYT's demands by October 30. This dispute highlights broader tensions between AI firms and media outlets over content rights, especially as publishers seek to protect intellectual property from unauthorized use by generative AI.
Anthropic's updated Responsible Scaling Policy outlines its commitment to safely train and deploy AI models by implementing proportional safeguards. The policy introduces AI Safety Level (ASL) Standards, setting technical and operational measures to mitigate risks associated with advanced models. To address emerging threats, new Capability Thresholds (e.g., for chemical, biological, or autonomous AI R&D) will trigger stricter safeguards. Regular capability assessments ensure models remain below risk thresholds, with comprehensive testing to identify potential threats. The update emphasizes flexible, outcome-focused safeguards and enhanced internal and external governance. This policy matters as it highlights the need for adaptable, transparent governance in the fast-evolving AI landscape, aiming to minimize risks while fostering technological advancement.
Adobe unveiled Project Super Sonic, an experimental tool that simplifies creating audio effects for videos. While text-to-audio generation already exists commercially, Super Sonic introduces two innovative features: object recognition, allowing users to click on a video frame to generate matching sounds, and vocal control, where users mimic sounds, and the tool generates precise audio effects. This approach offers creators more control, making it easier to add tailored audio to videos. Though currently a demo, Super Sonic reflects Adobe’s push to integrate advanced AI features into its Creative Suite, following similar work on the audio extension of their Firefly model. Whether it will become a permanent feature remains to be seen.
Amazon Ads introduced two new generative AI tools, AI Creative Studio and Audio Generator, aimed at simplifying ad creation and enhancing customer engagement. The AI Creative Studio allows advertisers to generate image, video, and audio content from a single product shot, enabling easy content refreshes and customization across formats. Advertisers can also explore an AI gallery for inspiration, and the tool provides unlimited storage for ad assets. The Audio Generator produces 30-second audio ads quickly and at no cost, lowering the barrier to audio advertising. Additionally, Amazon is testing features like Video Generator and Live Image, which create dynamic video and animated ads, responding to advertiser needs for more engaging content. These tools are currently in beta for U.S. advertisers.
Google has revamped its Shopping platform using AI, integrating 45 billion product listings with Gemini models to enhance user experience. The new AI-driven features offer personalized shopping recommendations, dynamic filters, and virtual try-ons, making it easier to find relevant products. Users can receive AI-generated briefs summarizing key considerations for their searches, like finding a winter jacket for specific climates. The updated platform also includes a personalized feed, allowing users to pick up where they left off, and a dedicated deals page for price comparisons and tracking. Rolling out in the U.S., the new Google Shopping aims to simplify research, provide more accurate results, and offer a seamless shopping journey, while still in an experimental phase with user feedback encouraged.
📬 Receive our amazing posts straight to your inbox. Get the latest news, company insights, and Niural updates.
A recent Meta study introduces Thought Preference Optimization (TPO) to enhance LLMs by teaching them to "think" before responding, without extra human data. Traditional LLMs respond directly, often struggling with complex tasks. TPO allows models to generate internal thoughts and optimize these via an iterative scoring process, using a judge model that evaluates response quality. This method improves LLM performance across various tasks, including marketing, health, and general knowledge, not just reasoning and problem-solving. By refining internal thoughts, TPO helps models plan and structure better responses, enhancing their ability to handle diverse, multi-step instructions effectively.
The Biden administration is considering further restrictions on the sale of advanced AI chips from American companies like Nvidia to certain countries, citing national security concerns. Reports suggest that nations in the Persian Gulf, including the UAE and Saudi Arabia, may be affected. This follows an earlier ban on exporting Nvidia and AMD chips to some Middle Eastern countries, amid concerns that the UAE could be a transshipment point for Russia to bypass sanctions and due to the UAE’s growing ties with China. Similar restrictions impacted countries in Africa and Asia starting in August 2023. While the White House declined to comment directly, it emphasized the need for safeguards in AI development despite its potential benefits.
🔎 Perplexity for Mac helps with search and discovery on Mac
⚙️ Gradio 5.0 builds and shares delightful machine-learning apps
📊 Strella is an end-to-end customer research platform with AI-moderated interviews and real-time synthesis
👏 Claap 3.0 is an AI sales coach that handles meeting notes, CRM enrichment, deal reviews, and more
👨💻 Code2ai turns your ideas into code so you can build with your own assistant
