Twice a week, Data Points brings you the latest AI news, tools, models, and research in brief. In today’s edition, you’ll find:
- ByteDance’s China-only video app
- U.K. watchdog probes Anthropic’s deals
- New Writer models for medicine and finance
- SambaNova’s speedy inference platform
But first:
OpenAI develops (but has not released) a powerful ChatGPT watermarking tool
OpenAI created a method to detect generated text by altering the token selection process in ChatGPT, leaving an imperceptible pattern called a watermark. The technique is 99.9 percent effective when sufficient new text is generated, providing a score indicating the likelihood that ChatGPT wrote part or all of a document. However, concerns exist about potential workarounds, such as using translation services or manual editing to erase the watermarks. The tool’s effectiveness and potential impact have sparked a two-year internal debate at OpenAI, highlighting the complexities of deploying such technology in educational and commercial contexts. (The Wall Street Journal)
Google DeepMind’s table tennis robot achieves amateur-level play
Google DeepMind trained a robotic arm to play table tennis at an amateur-competitive level, winning 13 out of 29 games against human opponents of varying abilities. The system uses a two-part approach, combining computer simulations for skill mastery and real-world data for continuous improvement, allowing it to adjust tactics and behavior during matches. This achievement represents progress toward creating robots that can perform useful tasks skillfully and safely in real environments, with potential applications beyond sports in areas such as homes and warehouses. (MIT Technology Review)
TikTok owner ByteDance unveils new video generation app
ByteDance launched Jimeng AI, a new text-to-video generation tool, on Android and Apple’s App Store for Chinese users. The software, developed by ByteDance-owned Faceu Technology, joins similar offerings from Chinese companies like Kuaishou, Zhipu AI, and Shengshu, which have recently introduced their own text-to-video models. Jimeng AI offers subscription plans starting at 69 yuan ($9.65) monthly, with options for single-month or annual subscriptions, allowing users to create about 2,050 images or 168 AI videos per month. This surge in AI video generation tools from Chinese tech firms highlights their rapid advancement in the field, as they compete with OpenAI’s unreleased Sora model. (Reuters)
U.K. government investigates Amazon-Anthropic partnership
The U.K.’s Competition and Markets Authority (CMA) launched an investigation into Amazon’s partnership with AI startup Anthropic, following a similar probe into Alphabet’s collaboration with the same company. The CMA will decide by October 4 whether to begin a deeper investigation or clear the partnershup of competition concerns. This move reflects growing concern among global antitrust regulators about deals between big tech companies and AI startups, as authorities work to ensure fair competition in the rapidly evolving AI industry. (Reuters)
Writer releases specialized models for medical and financial sectors
Writer introduced two new domain-specific large language models, Palmyra-Med and Palmyra-Fin, designed for medical and financial applications. Palmyra-Med outperformed other models in medical benchmarks, achieving an average of 85.9% accuracy across various tests, while Palmyra-Fin passed the CFA Level III exam with a 73% score on the multiple-choice section. These specialized models aim to provide AI developers with more accurate and compliant tools for building applications in highly regulated industries. (Writer)
SambaNova sets speed record for Llama 3.1 inference
SambaNova achieved 114 tokens per second on Meta’s Llama 3.1 405B model, setting a performance record independently verified by Artificial Analysis. The company’s platform, powered by its fourth-generation RDU chip, enables enterprises to deploy private language models with real-time capabilities for use cases like intelligent document processing, AI copilots, explainable AI, and agentic AI automation. This breakthrough in speed and efficiency allows businesses to leverage large language models more effectively for improving customer satisfaction and employee experience, which Gartner identified as top AI priorities for CEOs. (SambaNova)
Still want to know more about what matters in AI right now?
Read last week’s issue of The Batch for in-depth analysis of news and research.
Last week, Andrew Ng introduced his new sequence of courses, AI Python for Beginners, aimed at teaching anyone to code with the help of AI:
“If you know someone who is curious about coding (or if you yourself are), please encourage them to learn to code! The case is stronger than ever that pretty much everyone can benefit from learning at least a little coding. Please help me spread the word, and encourage everyone who isn’t already a coder to check out AI Python for Beginners.”
Read Andrew’s full letter here.
Other top AI news and research stories we covered in depth: Google gets Character.AI co-founders, how employers and prospective employees are embracing automated hiring tools, Ukraine's aquatic drones, and ArtPrompt, a technique to test the impact of text rendered as ASCII art on LLM performance.