Robots Evolve Rapidly

PLUS: China Matches OpenAI, Twitter's AI News Summaries and more.

Author

Wes Roth
May 06, 2024

Today:

  • Robots Evolve Rapidly

  • iOS 18 AI Upgrades Leak 

  • YouTube’s Jump Ahead

  • China Matches OpenAI 

  • Twitter's AI News Summaries

  • Fei-Fei Li's Spatial Startup

  • Confluent's AI Streamlining Tool

SHOCKING Robots EVOLVE in the SIMULATION plus OpenAI Leadership Just... LEAVES?

OpenAI-backed robots are advancing rapidly. Figure 01, featured on 60 Minutes, uses visual reasoning via OpenAI's pre-trained models to identify and handle objects based on common-sense prompts. Dr. Eureka, a new LLM agent, automates training robot skills by bridging simulation-to-reality gaps and developing new robot behaviors. 

OpenAI has faced internal shakeups with the departure of senior executives. Meanwhile, Devon 2.0, an AI agent for software development, is gaining traction despite mixed reviews. Lastly, Brilliant Labs introduced Frame, open-source AI glasses for developers, emphasizing a new era in augmented reality.

More details of the AI upgrades heading to iOS 18 have leaked

Apple's iOS 18, set to debut at WWDC 2024, will focus heavily on artificial intelligence enhancements. Key features include AI-powered text summarization across Safari and Siri, using the Ajax Large Language Model (LLM). Siri will offer fast, private, on-device responses, with cloud-based AI like Google Gemini or ChatGPT for complex prompts. 

Enhanced integration with Contacts and Calendar apps will improve productivity, while Spotlight on macOS will deliver smarter search results. Apple emphasizes privacy and reliability in these upgrades, promising significant boosts in efficiency and user experience.

YouTube Premium members can now test AI-powered ‘Jump ahead’

YouTube

YouTube Premium users can now test the new AI-powered "Jump ahead" feature, which quickly skips to where viewers typically move forward. The feature is available on the YouTube Android app for English videos in the US and appears after double-tapping to skip ahead. Accessible through youtube.com/new experiments, it uses AI and viewing data to identify optimal skip points. 

Currently, it's exclusive to Premium members, with a potential expansion beyond the June 1 end date. The feature joins the AI-generated "Ask" questions option, enhancing YouTube's goal of providing smarter viewing experiences.

Four start-ups lead China’s race to match OpenAI’s ChatGPT

Four Chinese generative AI start-ups, including Zhipu AI, Moonshot AI, MiniMax, and 01.ai, have been valued at between $1.2bn and $2.5bn in the past three months. China's generative AI start-ups have raised Rmb14.3bn ($2bn) in the first four months of the year, defying the chill in investment seen elsewhere in consumer technology.

Chinese AI start-ups are leveraging their engineering talent and open-source models to develop competitive AI products and overcome US restrictions on advanced chip exports.

Elon Musk wants to use AI to summarize the news on Twitter and ... it's not a terrible idea?

Elon Musk aims to integrate AI into Grok, X/Twitter's chatbot, to create a news summarizer. The idea, detailed by Peter Kafka of Business Insider, involves using Grok to condense the latest headlines and trending topics. Despite not replacing traditional news outlets, the feature could offer value by providing quick summaries of major stories. 

Grok's summary tool is part of Musk's broader vision for enhancing the platform with AI capabilities, aligning with his ambition to reshape social media. While the potential is promising, it remains unclear how Grok will differentiate from other news summarization tools.

Stanford AI leader Fei-Fei Li building 'spatial intelligence' startup

Fei-Fei Li, a leading AI expert and Stanford professor, is developing a startup focused on spatial intelligence to enhance AI's visual processing and reasoning capabilities. Supported by investors like Andreessen Horowitz and Radical Ventures, her company aims to replicate human-like 3D perception and reasoning in machines, potentially leading to advances in artificial general intelligence (AGI). 

The technology will help AI understand visual information and act based on predictions, like preventing a glass from falling off a table. This venture complements Li's pioneering work at Stanford and her efforts to bridge AI research gaps between academia and industry.

Confluent launches plug-and-play offering to accelerate realtime streaming AI

Confluent has launched a new plug-and-play feature that simplifies real-time AI streaming. Introduced at the first Kafka Summit in Asia, this tool enables seamless AI model inference within Apache Flink, helping developers efficiently train and infer AI models. Users can call AI engines like OpenAI, AWS SageMaker, GCP Vertex, and Microsoft Azure through simple SQL statements without complex coding. 

The platform's new Freight Clusters offer cost-effective data replication with up to 90% savings, and an upcoming AI assistant will streamline coding tasks. Confluent aims to expand its presence in APAC while focusing on data governance and cataloging improvements.

🧠RESEARCH

Prometheus 2, an open-source language model, is designed to evaluate other language models. It aligns closely with human and GPT-4 judgments, supporting both direct assessment and pairwise ranking. Prometheus 2 excels in flexibility, transparency, and affordability, outperforming other open-source evaluators. 

LoRA Land, a technical report, introduces 310 fine-tuned LLMs using Low Rank Adaptation (LoRA) that outperform GPT-4 by 10 points on average. The LoRAX server powers 25 specialized LLMs on a single GPU, highlighting the quality, affordability, and efficiency of using multiple fine-tuned models over general-purpose LLMs.

StoryDiffusion introduces Consistent Self-Attention for consistent content in generated images and videos. It uses a Semantic Motion Predictor for smooth video transitions, ensuring stable subjects across long-range sequences. This framework enhances text-to-image diffusion models and supports seamless visual storytelling. 

WildChat presents a dataset of 1 million user-ChatGPT conversations, comprising over 2.5 million interaction turns. It includes diverse user prompts, various languages, and demographic data for geographic and temporal analysis. This comprehensive dataset aids researchers in studying chatbot usage and fine-tuning instruction-following models.

NeMo-Aligner is a scalable toolkit designed to align Large Language Models (LLMs) with human values and preferences. It supports major alignment methods like RLHF, DPO, SteerLM, and SPIN while efficiently utilizing hundreds of GPUs.

🛠️TOP TOOLS

Layla - Smart, always-awake, baby sleep coach

The Fastest AI - Reliable measurements for the performance of popular models

Sonnetai - Your End-To-End Meeting Assistant

Low Difficulty - Getting you on the first page of Google fast

Muddy - AI-first browser for working with others

📲SOCIAL MEDIA

🗞️MORE NEWS

Google bans advertisers from promoting deepfake porn services

Google will ban ads promoting services that create deepfake porn or altered nude images. This change, effective May 30th, aims to explicitly prohibit synthetic sexual content. Violating ads will be removed. The move follows rising concerns over nonconsensual deepfake pornography and emerging legislation on digital forgery. THE VERGE

Indie artist Washed Out's new music video was fully AI-generated

Indie artist Washed Out released a fully AI-generated music video for his song "The Hardest Part." Directed by Paul Trillo using OpenAI's Sora text-to-video model, the video explores a couple's life journey with surreal, dreamlike visuals. Trillo sees AI as a tool for expanding creative possibilities in music videos. NBC

AI-Voiced Audiobooks Top 40,000 Titles on Audible

Audible now hosts over 40,000 AI-voiced audiobooks, offering authors a new revenue stream. However, some listeners are dissatisfied with the influx of AI-generated content. This development reflects the evolving landscape of audiobook production and consumer preferences. BLOOMBERG

JPMorgan Unveils IndexGPT in Next Wall Street Bid to Tap AI Boom

JPMorgan Chase has launched IndexGPT, a tool that uses OpenAI's GPT-4 to create thematic investment baskets. The AI analyzes keywords to identify relevant companies by scanning news articles. This product showcases JPMorgan's strategic move into the burgeoning AI-powered finance sector. BLOOMBERG

Randy Travis Sings New Song Through AI, Years After Stroke Took His Voice

Country music legend Randy Travis released "Where That Came From," his first song since a stroke in 2013. The AI-generated track uses past recordings to recreate his voice, showcasing the potential of AI in creative expression. Fans and fellow artists celebrated this technological aid that brought Travis' voice back to life. FORBES

Generative AI will be designing new drugs all on its own in the near future

Generative AI is revolutionizing drug discovery. Eli Lilly and Nvidia use AI to design novel molecules and proteins, accelerating research and improving drug success rates. Google's DeepMind AlphaFold paved the way in 2021, demonstrating AI's creative potential in biology. Experts predict AI-designed drugs will become the norm soon. CNBC

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Join the conversation

or to participate.