This Week’s Covers
This week’s cover combines several top themes into one image: Elon Musk’s AI tool, Grok, supports largely uncensored image creation within Twitter/X; Google has launched an assistant called ‘Gemini Live’; AI agents are set to change how we interact with customer service; and several bills up for legislation are shaping politics. The image depicts Donald Trump and Joe Biden sitting in an abandoned Google call center, reviewing a piece of legislation. It was created using Grok’s image tool, Flux.
The category covers continue the theme of mobile AI agents, showing a hotel concierge holding an iPhone displaying the category name and some context in the background. The cover images were all single-shot, first-try results using Grok 2 on Twitter/X. As usual, my habits from MidJourney initially stymied the results. Flux handles complex themes and text way better. The eventual rubric was something like: “A hotel concierge holds an iPhone toward the camera. The phone screen displays the words ‘Category Name.’ On the concierge’s desk is [describe category]. On the wall behind the desk is [additional detail].” The goal was not to showcase my ability to generate great images but rather to test “lazy prompting” single-try results to see how well AI performs for casual users. Here are a few of the better results:
Executive Summaries
Special Section – The Agents are Arriving – I’ve been shouting this for 18 months
AI Agents Set to Transform the Internet and Industries
AI agents are poised to revolutionize the internet and how businesses operate, according to Jeremiah Owyang. As these agents take on the task of gathering information for users, the traditional structure of websites and online strategies will need to adapt. This shift is expected to fundamentally reshape how we interact with the web. On a more specific level, industries like e-commerce could be disrupted as AI agents handle shopping for users, workplaces may see AI completing more tasks than humans, and governments will likely interact with AI agents on behalf of citizens. The impact will be broad and far-reaching.
Jowyang
Apple Unveils Agentic LLM Tool Use Benchmark
Apple has introduced a new benchmark that sets a high standard for evaluating how AI agents use tools, emphasizing real-world, state-dependent tasks. Unlike traditional benchmarks that focus on single-turn, stateless tool usage (like calling APIs or using an app once), Apple’s approach centers on agents operating within ongoing dynamic environments, such as phone settings or multiple apps, where maintaining and accessing the state is crucial. This marks a major shift, as it addresses how AI should handle more complicated tasks that involve remembering and adjusting to changes. With Siri, Apple is well-positioned to lead in this area. Their benchmark, ToolSandbox, measures the challenges even top AI models face, particularly in handling state dependencies and complex tasks.
RajaswaPatil | arxiv.org | github.com
Google Gemini Transforms Mobile Devices into Powerful AI Assistants
Google has launched Gemini Live, an AI-powered mobile assistant now available to advanced subscribers on Android. Gemini offers a natural, conversational experience, allowing users to engage in real-time discussions and receive assistance for both simple and complex tasks. Users can even continue talking with the Gemini app in the background or when their phone is locked. It integrates seamlessly with Google apps like Keep, Gmail, Calendar, and YouTube Music, providing personalized help without switching between services.
Reddit | blog.google | GoogleDeepMind | theverge
Cohere’s Guide to Building Agentic AI for Enterprises
Cohere outlines the key resources and strategies businesses need to build agentic AI—autonomous AI systems capable of handling complex, multistep workflows. Leveraging retrieval-augmented generation (RAG) as a foundation, agentic AI connects large language models (LLMs) with tools like databases and APIs to perform tasks, not just generate responses.
cohere.com
Salesforce Introduces Agentforce for Custom AI Development
Salesforce has launched Agentforce, a platform that allows users to build, test, and scale custom AI agents. Built on the low-code Einstein1 platform, Agentforce enables seamless integration with enterprise data, automates actions, and connects effortlessly with APIs. Users can test agents in sandboxes before deployment and use them across apps, web, social media, and Slack, empowering companies to boost customer success through AI-driven solutions.
Benioff
Salesforce Launches Multi-Agent Tool for Better Software Problem Solving
Salesforce has released DEI (Diversity Empowered Intelligence), a tool that helps groups of AI software agents work together more effectively. DEI manages teams of these AI agents, allowing them to solve problems better than any single agent could on its own. In tests on SWE-Bench Lite, a group of open-source AI agents improved their problem-solving success rate from 27.3% to 34.3% with the help of DEI, which is a 25% boost. This makes DEI a strong competitor even against many closed-off, private AI tools.
_akhaliq | huggingface
The Rest of the Executive Summaries
California AI Regulation Bill Sparks Debate in Silicon Valley
A proposed California bill aimed at regulating artificial intelligence is raising concerns in Silicon Valley, where tech giants like Google, Meta, and OpenAI fear it may stifle innovation. Authored by State Senator Scott Wiener, SB 1047 would mandate safety testing for powerful AI technologies before their release and hold companies accountable for harm caused by their systems.The bill targets developers of AI systems costing over $100 million to train, like those from tech giants Google, Microsoft, and OpenAI, mandating safety protocols, third-party audits, and emergency shutdown mechanisms (impossible via open source?). Supporters argue the bill is necessary to prevent potential disasters, while critics warn it could push AI development out of the state and restrict open-source innovation. As the bill advances through the legislature, it has become a flashpoint in the broader debate over AI safety and regulation.
nytimes.com | techcrunch.com
Musk’s Grok AI Sparks Debate with Uncensored Image Generation on X
Elon Musk’s AI company, xAI, has introduced two new language models, Grok-2 and Grok-2 mini, available to X Premium subscribers. Notably, Grok integrates with the Flux model to allow users to generate largely uncensored, photorealistic images, including controversial content like deepfakes and depictions of real people in compromising scenarios. While Grok claims to limit explicit content, early reports suggest the safeguards are inconsistent. The move raises concerns about AI misuse and legal liability, with the proposed NO FAKES Act potentially holding platforms like X accountable for deepfake creations. Grok-2 is also being compared to other GPT-4-class models, but it has not surpassed OpenAI’s GPT-4 in capabilities. While Grok’s image generation dominates headlines, its close connection to Twitter data continues to frustrate some users, who find the model’s responses irrelevant at times.
arstechnica.com | bilawalsidhu | bilawalsidhu
Microsoft’s Chief Scientist Predicts AI Creativity Surge
Eric Horvitz, Microsoft’s Chief Scientific Officer, believes AI systems are already demonstrating remarkable creativity, but he predicts a major leap in the next 18 months. According to Horvitz, we are on an exponential growth curve, and soon there will be no question about AI’s ability to rival human creativity.
Tsarnick | youtube
AI Visuals and Charts
“WrestleMania of Power Made with: MidJourney, LumaLabsAI, and Uudiomusic – Featuring the world’s most powerful individuals at WrestleMania.
“Some Chinese dudes imitating AI videos lol this is next level”
“People aren’t ready for the X hive mind to unleash the power of state-of-the-art, uncensored AI image generation upon the world. Today the day AI image generation went mainstream and said fuck it, we ball. 10 of the most wild and cursed generations I’ve come across:”
Top 18 Links of The Week Sorted by Category
Audio
A.L.S. Stole His Voice. A.I. Retrieved It. – The New York Times
Sonova’s AI hearing aids offer crystal-clear speech in noisy places
Augmented and Virtual Reality
“FruitNeRF A Unified Neural Radiance Field based Fruit Counting Framework discuss:
AI captioning glasses for the hearing impaired
“This is what AI is for. Just this use case alone is worth every dollar invested in it.
Business and Enterprise
How Walmart is using AI to improve its business, and save money – MarketWatch
Chips, Hardware, and Infrastructure
Cisco cuts thousands of jobs as it shifts focus to AI, cybersecurity | AP News
Ethics/Legal/Security
ChatGPT unexpectedly began speaking in a user’s cloned voice during testing | Ars Technica
SAG-AFTRA strike: How AI is changing video game development : NPR
Exclusive | Huawei Readies New Chip to Challenge Nvidia, Surmounting U.S. Sanctions – WSJ
Huawei readies new AI chip to challenge Nvidia in China, WSJ reports | Reuters
Mobile
14 new AI features in Google’s Pixel 9 series devices
Multimodality
“Language Model Can Listen While Speaking Overview: This paper introduces the listening-while-speaking language model (LSLM) for real-time, interactive speech dialogue. LSLM supports simultaneous listening and speaking, allowing interruptions during responses. It combines a TTS
Publishing
Prediction marketplace Polymarket partners with Perplexity to show news summaries | TechCrunch
“We’re thrilled to announce our partnership with @Polymarket. Now, when you search for events on Perplexity, you’ll see news summaries paired with real-time probability predictions, such as election outcomes, market trends, and beyond.
Google’s AI Search Gives Sites Dire Choice: Share Data or Die – Bloomberg
“Designing a website will never be the same again. Take a look at this quick video I recorded. I generated a complete landing page in Figma in less than 60 seconds. This is game-changer for people like me, who don’t know much about design but want to create something decent with
Robotics and Embodiment
Boston Dynamics
“Boston Dynamics revealed Atlas’ dexterity with its ability to do pushups and burpees during a presentation at RSS 2024 This is the companies fully-electric robot that they announced in April
Twitter/X/Grok
“BREAKING: Here’s an early look at Grok 2.0 features and abilities! It’s better at coding, writing, and generating news! It’ll also generate images using the FLUX.1 model!
Leave a Reply