Good morning, it’s Tuesday! OpenAI's hiring an AI detective (to watch its own employees?!), Elon's Grok-2 is giving ChatGPT a run for its money, and Salesforce is rolling out new AI sales agents. Oh, and did I mention Boardwalk Robotics' new humanoid torso robot named Alex?
There’s a lot to cover today in the world of AI, let’s get started.
Your Daily Roundup:
The open-source AI model might be crucial for democratizing technology and promoting economic and scientific advancement.
Meta Reality Labs introduces Sapiens, a set of vision models for various tasks, showing superior performance and adaptability.
xAI's Grok-2 models receive a speed and accuracy boost after code optimization, competing with top AI models.
Enterprises are hesitant to integrate Microsoft Copilots due to security and governance concerns, delaying AI adoption.
Cursor, an AI-assisted programming tool, gains popularity and investment, demonstrating the potential of AI in developer workflows.
OpenAI seeks a Technical Insider Risk Investigator, highlighting concerns about internal security threats and the need for AI safety compliance.
👉️ Top Stories
Statement of Support for Meta’s Open Approach to Today’s AI
Open-source AI, as championed by Meta and Spotify, is posited as the key to democratizing technological advancements, bolstering economic growth, and fostering scientific progress. The current fragmented and risk-averse regulatory landscape in Europe, exemplified by inconsistent GDPR application, is criticized for stifling innovation and potentially causing Europe to lag behind in the global tech race, underscoring a call for clearer, more harmonized policies to leverage Europe's talent and competitive potential. Continue reading here.
Abstract Meta Reality Labs introduces Sapiens
Abstract Meta Reality Labs introduces Sapiens, a series of advanced vision models designed for tasks such as 2D pose estimation, body-part segmentation, depth estimation, and surface normal prediction, capable of high-resolution processing and adaptable through fine-tuning. The models, pre-trained on a diverse dataset of over 300 million images, display superior generalization capabilities, with scalable design allowing for enhanced performance with up to 2 billion parameters, outperforming current standards on multiple benchmarks. Continue reading here.
Grok-2 gets a speed bump after developers rewrite code in three days
Elon Musk's AI company xAI has recently improved its Grok-2 large language models, Grok-2 and Grok-2 mini, both in terms of speed and accuracy, following a complete rewrite of their inference code stack by developers using the open-source system SGLang. Grok-2 has achieved a #2 ranking on the LMSYS Chatbot Arena leaderboard, demonstrating its competitiveness amongst the world's most powerful AI models, while also excelling specifically in mathematical tasks. Continue reading here.
AI copilots are getting sidelined over data governance
Enterprises are facing challenges integrating Microsoft Copilots into their systems due to security and governance issues, as the AI tool can inappropriately summarize sensitive data and exacerbate access rights complexities. Jack Berkowitz from Securiti highlights that while some AI initiatives show good returns, there is significant caution regarding Copilot deployment, with about half of the surveyed companies grounding or restricting its use, pointing to a need for improved data and security management to safely implement such AI services. Continue reading here.
Investing in Cursor | Andreessen Horowitz
Large Language Models (LLMs) have shown proficiency in basic programming tasks and are advancing towards more complex problem-solving through AI-human collaboration, highlighting the importance of interfaces that integrate AI into developer workflows. Cursor, a customized version of VS Code for AI-assisted programming that supports existing LLMs and offers features like next action prediction and natural language edits, has attracted a significant user base and investment, positioning it as a key tool for developers in harnessing the potential of AI. Continue reading here.
OpenAI is hiring someone to investigate its own employees
OpenAI is recruiting a Technical Insider Risk Investigator, highlighting concerns about internal security threats amidst broader debates on AI safety and interventions to stop malicious uses of their AI models by state actors and hackers, including a 2023 breach of their messaging system. The role involves activity analysis and investigation coordination, part of OpenAI's compliance with AI safety commitments, while the company also faces internal calls for transparency on AI risk from its own employees. Continue reading here.
✍️ Editor Picks
Research
Automating Thought of Search: A Journey Towards Soundness and Completeness
AutoToS advances the use of large language models in automated planning, by guiding them to generate sound and complete search components without human intervention, successfully solving planning problems with 100% accuracy on all evaluated domains, using feedback from unit tests. This method builds upon the Thought of Search approach but automates the process, showing promise for LLMs in complex reasoning tasks and code generation. Continue reading.
Robotics
Meet Boardwalk Robotics' Addition to Humanoid Robots
“With Alex, we’re focusing on the manipulation side first, getting that well established. And then picking the mobility to match the task.”
Boardwalk Robotics has introduced a new commercial humanoid upper torso robot named Alex, designed to enhance work in sectors such as manufacturing, logistics, and maintenance, with a focus on manipulation tasks and the potential for future legged mobility. Although derived from a collaboration with the renowned IHMC, which has a robust history in bipedal robotics, Alex is an in-house developed robot that is currently legless for practicality and safety, aiming to be a functional and cost-effective solution for various industries, with pilot programs underway. Continue reading.
Models
Introducing Pharia-1-LLM: transparent and compliant
The newly launched Pharia-1-LLM-7B models, available for non-commercial research under the Open Aleph License, offer concise, length-controlled responses optimized for German, French, and Spanish. These models excel in domain-specific applications, especially in automotive and engineering, and are designed in compliance with EU data regulations. The models underwent extensive fine-tuning for instruction following, with the control-aligned variant featuring additional safety and alignment measures. Built using the GPT architecture, Pharia-1-LLM-7B emphasizes efficiency, robustness, and multilingual capabilities, making it a strong alternative in the open-source LLM landscape. Continue reading.
🚀 AI Launches
Google debuts free ‘Prompt Gallery’ in AI Studio, supercharging developer tools and enhancing the tools available for developers using the Gemini API, and showcasing a variety of pre-built prompts for diverse applications ranging from code testing to creative projects.
Microsoft will release controversial Windows Recall AI search feature to testers in October - Microsoft announced that its AI-powered search feature Recall, which takes screenshots of user activity, will be available for testing by Windows users starting October, but will remain off by default due to privacy concerns raised by security researchers.
Meet Einstein SDR and Einstein Sales Coach: Two New Autonomous AI Sales Agents to Scale Your Sales Team - Salesforce introduces two AI sales agents: Einstein SDR Agent and Einstein Sales Coach Agent, with the former autonomously engaging and nurturing inbound leads 24/7, and the latter providing role-play coaching to sales reps.
Meet Paperguide: Discover, Read, Write, and Manage Research with Ease Using AI - Paperguide is a cutting-edge platform designed to enhance the academic research experience by leveraging AI tools tailored for students, researchers, and academics.
🧰 Toolbox
Poe Introduces Playground v3 with Enhanced Image Generation and Expanded Features
Boost your AI with Azure's new Phi model, streamlined RAG, and custom generative AI models
LMSYS Introduces 'Overview' Feature for Holistic Model Evaluation on Leaderboards
LinkedIn Releases Liger Kernel: A Revolutionary Tool Boosting LLM Training Efficiency by Over 20%
CiteME is a benchmark designed to test the abilities of language models in finding papers that are cited in scientific texts
Nous Research Introduces DisTrO: A Decentralized Approach to LLM Training
📽️ New Videos
🛰️ Houston, we have more headlines!
Elon Musk takes tour through Cortex, the new AI training supercluster being built at Tesla HQ
Boston Dynamics Atlas as gymnast
Chinese Entities Use Amazon Cloud and Rivals to Access Restricted US Chips and AI
Google Makes Former Character.AI CEO Shazeer a Co-Leader of Gemini AI
Warnings AI tools used by government on UK public are ‘racist and biased’
Apple Explores Robotics in Search of Life Beyond the iPhone
The AI Guys Are Driving Themselves Mad
Strategist: Learning Strategic Skills by LLMs via Bi-Level Tree Search
Penny a Minute Voice Bots Help India Startups Vie With OpenAI
🏡 House Keeping
Add forwardfuture@mail.beehiiv.com to your contacts.
And add us to your Primary inbox OR Reply to this email with “spaceman”
Share Forward Future with your AI buddies
