Good morning, it’s Friday! Today, we’re talking about Elon Musk's uncensored AI art, OpenAI's stealthy ChatGPT upgrade, and SoftBank's ambitious AI chip dreams. We'll also jump into the complexities of a GenAI-driven future, grapple with the ethics of AI-written resumes, and explore the fallout from the Dutch crackdown on AI datasets.

Take a deep breath… Okay, let’s jump right in!

Get the best of multiple LLMs in one place! ChatHub enables you to compare and combine the responses from six different AI bots simultaneously, enhancing your use of LLMs and transforming your interactions with AI. Try ChatHub now: https://chathub.gg/?utm_source=berman

👉️ Top Stories

Musk’s new Grok upgrade allows X users to create largely uncensored AI images

xAI, Elon Musk's AI company, introduced Grok-2 and Grok-2 mini language models, which integrate with the controversial Flux image synthesis model to generate uncensored, photorealistic images—raising questions about misuse and responsibility as they lack safeguards against creating sensitive content. Grok-2, although it claims improvements in language processing and visual tasks, faces criticism for potentially incorporating unreliable tweet information and is currently offered exclusively to X platform subscribers, with future plans for an enterprise API. Continue reading here.

SoftBank discussed AI chips tie-up with Intel to rival Nvidia

The Financial Times offers various subscription options providing complete digital access to its journalism across devices, with benefits like expert analysis, curated newsletters, video content, and the ability to gift articles. Subscribers can opt for monthly or yearly payments, with the latter offering a 20% discount, and the most premium package includes digital access as well as the print edition of the newspaper delivered throughout the week. Continue reading here.

OpenAI updates ChatGPT to new model that exhibits multi-step reasoning

VentureBeat reports on OpenAI's quiet release of the latest GPT-4o model within ChatGPT, which users have observed to display improved multi-step reasoning and more detailed natural language explanations. Additionally, the updated model integrates advanced native image generation capabilities, surpassing the performance of its predecessor, which relied on DALL-E 3—sparking user debate on the extent and transparency of the changes made. Continue reading here.

Introducing SWE-bench Verified

OpenAI has released SWE-bench Verified, a human-validated subset of the SWE-bench benchmark aimed at more accurately evaluating AI models' software engineering capabilities. SWE-bench Verified addresses issues in the original dataset, such as poorly specified problem statements and irrelevant or overly specific unit tests, which led to underestimations of model performance. By collaborating with software developers, OpenAI filtered out problematic samples and introduced a new evaluation harness to enhance reliability. The updated dataset improves model performance assessments, especially for tasks that were previously difficult or impossible to evaluate fairly. Continue reading here.

How to Prepare for a GenAI Future You Can’t Predict

The article discusses the rapid advancement of generative AI technology and its potential impact on the workforce, highlighting that executives are considering its implementation but may be underestimating the complexities of integrating AI with human employees. It presents a real-world example of a banking CEO contemplating the use of generative AI to address skill gaps and improve services, while also facing the challenge of adapting workforce expectations in a post-Covid environment. Continue reading here.

Jobhunters flood recruiters with AI-generated CVs

Many job seekers, up to 50%, are relying on generative AI tools like ChatGPT to write or enhance their CVs and cover letters, resulting in a significant increase in low-quality, easily identifiable applications due to unedited, generic language. This surge in AI-generated submissions has complicated the hiring process for recruiters, particularly as labor markets tighten, while major employers like the Big Four accounting firms actively discourage AI use in applications, and AI-driven advantages in recruitment assessments are raising concerns about fairness and access disparities. Continue reading here.

Copyright Group Takes Down Dutch Language AI Dataset

Dutch copyright enforcement group BREIN removed a large language dataset containing unauthorized content from books, news, and media used for AI training. The move comes amid growing global scrutiny, with EU regulations and legal actions pushing for transparency on datasets used by AI companies. Continue reading here.

Gemini makes your mobile device a powerful AI assistant

Gemini is an AI-powered mobile assistant that offers advanced conversational capabilities, allowing users to engage in natural, free-flowing dialogue and execute complex tasks hands-free across both Android and iOS devices. It integrates seamlessly with Google apps and is expanding its functionality with new extensions and voice options, promising a better user experience with context-aware assistance and continuous improvements, such as the upcoming deeper integrations with Google Home, Phone, and Messages. Continue reading here.

✍️ Editor Picks

Research

ControlNeXt: Powerful and Efficient Control for Image and Video Generation

ControlNeXt is introduced as an efficient architecture for controllable image and video generation, simplifying existing models by significantly reducing additional computational costs and seamlessly integrating with LoRA weights for style changes without retraining. The paper also presents Cross Normalization (CN) as a novel approach for quick and stable training convergence, showcasing the method's robustness across numerous experiments with varied base models. Continue reading.

Robotics

Body Transformer: Leveraging Robot Embodiment for Policy Learning

The Body Transformer (BoT) is a novel architecture designed to improve machine learning for robot learning by integrating the structure of robots, using a graph representation of sensors and actuators and masked attention, to surpass the performance of both vanilla transformers and classical multilayer perceptrons in task completion, scalability, and efficiency. Continue reading.

Models

Grok-2 Beta Release

Grok-2 and its smaller version, Grok-2 mini, are the latest AI language model releases on the 𝕏 platform, boasting superior performance in reasoning, comprehension, and coding compared to their precursor Grok-1.5 and other contemporaneous models, as evidenced by benchmarks like GPQA, MMLU, and DocVQA. The Grok models are in beta on 𝕏 with an updated interface and new features for premium users, while also set to become available to developers through an enterprise API with enhanced security and multi-region support later in the month. Continue reading.

🚀 AI Launches: Ignition Sequence Start!

• Opera One Browser Launches on iOS: Opera has launched its AI-powered Opera One browser on iOS, featuring a clutter-free, minimalist design and an intelligent interface.

• In-browser Postgres with an AI interface: Create and interact with multiple Postgres databases directly in the browser.

• Snowflake launches Cortex Analyst: An agentic AI system for accurate data analytics.

• Sakana AI: In collaboration with academic partners, Sakana AI has introduced The AI Scientist, a system designed for autonomous scientific research.

🧰 AI Toolbox

Prompt caching with Claude - The Anthropic API has introduced prompt caching for its Claude models, allowing developers to significantly cut costs by up to 90% and reduce latency by up to 85% in API calls by caching frequent context. The new feature supports various applications like conversational agents and large document processing, and is priced based on input token numbers with specific rates for writing to and reading from the cache.

Secure code more than three times faster with Copilot Autofix - GitHub has announced the general availability of AI-powered Copilot Autofix within GitHub Advanced Security, designed to enhance software development security by analyzing vulnerabilities, explaining their implications, and providing code suggestions to correct them efficiently. During its public beta, Copilot Autofix users were fixing vulnerabilities more than three times faster than manual methods, indicating a significant increase in the speed of secure software development processes.

📽️ New Videos:

🛰️ Houston, we have more headlines!

• Introducing Sarvam-2B: a 2-billion parameter language model focused on supporting 10 Indic languages

• Google’s Gemini Live AI Sounds So Human, I Almost Forgot It Was a Bot: The assistant offers near-human conversation abilities, quick responses, and dynamic interaction, setting it apart from competitors like Siri and Alexa.

• Salesforce Research DEI Agents: DEIBASE is an innovative implementation of the DEI framework designed for evaluating software engineering (SWE) solutions, particularly utilizing a Large Language Model (LLM) to review and score code patches

• Unleashing 10,000+ Word Generation: The LongWriter study introduces AgentWrite and LongWriter-6k dataset to enable LLMs to produce coherent outputs surpassing 20,000 words.

🏡 House Keeping

1. Add [email protected] to your contacts.

2. And add us to your Primary inbox OR Reply to this email with “spaceman”

3. Share Forward Future with your AI buddies