AIR#116 - June 28, 2024
Good morning, AI aficionados! As you sip your morning coffee, get ready to dive into today's edition of AIR: The AI Recon. Leading the headlines is CriticGPT, an exciting development where GPT-4 helps find its own mistakes, boosting accuracy by a remarkable 60%. This self-reflective AI promises to refine feedback loops and enhance the reliability of AI interactions, making it a must-read for developers and tech enthusiasts alike.
But that's not all! Claude 3.5 Sonnet is taking the AI world by storm, offering faster, cheaper, and more effective interactions with new coding and collaborative features. If you're keen on the latest advancements in language models, this story is a game-changer you won't want to miss. Meanwhile, Cloudflare's new AI scraper blocking option is stirring up conversations about the balance between innovation and honesty in the tech space. This feature's effectiveness hinges on AI companies' transparency, making it an intriguing read for those invested in online security and ethics.
And for a splash of creativity, check out Open-Sora, which allows you to generate stunning videos on consumer GPUs with ease. This tool, accessible via Jupyter, is revolutionizing video production for developers and digital artists. Whether you're here for groundbreaking tech updates, ethical debates, or innovative business moves, today's edition is packed with stories that will both intrigue and challenge you. So, sit back, sip your coffee, and let's delve into the dynamic world of artificial intelligence together!
Business
Hypercar Maker Rimac Reveals Fully Autonomous Verne Robotaxi
Rimac unveils Verne robotaxi: a fully autonomous, 2-seater pod launching in Zagreb in 2026, expanding across Europe and the Middle East.
YouTube Negotiates AI Music Licensing Deal with Record Labels
YouTube negotiating AI music licensing with record labels to expand its AI song generator, aiming to balance innovation and artist rights.
OpenAI Partners with TIME for 101-Year Archive Access
OpenAI partners with TIME for 101 years of archives, enhancing AI responses with trusted journalism and supporting accurate information access.
AI-Driven Resume Coach and Interview Trainer
AI-powered ResumeFromSpace: Build resumes, scan them, and train for interviews with tailored questions for any job.
Amazon Investigates Perplexity AI for Scraping Abuse
Amazon investigates Perplexity AI for potentially violating AWS rules by scraping content from websites that block bots.
Anthropic CEO on Underdog Status, AI Safety, and Economic Inequality
Anthropic CEO discusses AI safety, economic inequality, and competition with OpenAI, launching Claude 3.5 to set new industry standards.
German AI Defence Start-up Helsing to Triple Valuation to $4.5B
German AI defence start-up Helsing to triple valuation to $4.5B with $500M funding from Accel and Lightspeed.
NBC to use AI version of Al Michaels' voice for Olympics recaps
NBC to use AI-generated voice of Al Michaels for Paris Olympics recaps on Peacock, offering personalized daily highlights.
AI Al Michaels to Recap 2024 Olympics on Peacock
NBC to use AI-generated Al Michaels for daily 2024 Olympics recaps on Peacock, sparking debate on tech replacing human commentators.
Engineering
[Github] Open-Sora: Impressive Video Generation on Consumer GPUs
Open-Sora: Generate stunning videos on consumer GPUs with ease. Access via Jupyter for seamless integration.
🔥 CriticGPT: GPT-4 Finds Its Own Mistakes
CriticGPT, based on GPT-4, helps trainers spot ChatGPT errors, improving accuracy by 60%. This aids in refining AI feedback loops.
🔥 Claude 3.5 Sonnet: The New Best LLM
Claude 3.5 Sonnet is the new top LLM, offering faster, cheaper, and more effective AI interactions, now with enhanced coding and collaborative features.
Cloudflare Adds AI Scraper Blocking Option
Cloudflare introduces AI scraper blocking, but effectiveness depends on AI companies' honesty in identifying themselves.
🔥 Creating a Dataset for LLM Fine-Tuning Evaluation: A Deep Dive by Alex Strick van Linschoten
Alex Strick van Linschoten dives into creating a dataset for evaluating fine-tuned LLMs, focusing on accuracy, edge cases, and complex scenarios.
Meta LLM Compiler: Revolutionizing Compiler Optimization with Code Llama
Meta's LLM Compiler leverages Code Llama to optimize code, enhancing compiler efficiency and reducing resource costs.
Gemini 1.5 Pro 2M Context Window and Code Execution Now Live
Gemini 1.5 Pro now offers a 2M context window, code execution, and Gemma 2 in Google AI Studio, enhancing developer capabilities.
[Guide] Training a 70B Model from Bare Metal: Infrastructure & Scripts
Guide: From bare metal to a 70B model—detailed infrastructure setup, scripts, and troubleshooting insights shared for seamless AI training.
Wider vs. Deeper: Optimal Transformer Configurations Explored
Optimal Transformer: Balanced models (4 layers, 1024 embd_dim) outperform deeper or wider ones, offering better efficiency and performance.
Baseten Chains – Multi-Model AI Framework and SDK Launched
Baseten launches Chains, a framework & SDK for efficient multi-model AI workflows, halving processing times and improving GPU use 6x. 🚀
Milk Infrastructure: AI-Driven, No-DevOps Kubernetes Deployment
Milk Infrastructure uses AI to deploy and manage Kubernetes on any cloud, eliminating the need for DevOps and cutting costs.
Meta LLM Compiler: Neural Optimizer & Disassembler Unveiled
Meta unveils LLM Compiler, optimizing and disassembling code using AI. Available for research and commercial use.
[GitHub] Agentpanel: Rust-Powered Universal LLM API and Observability Server
Agentpanel: Rust-based AI gateway & observability server for optimizing multi-agent workflows, supporting 100+ LLMs across 20+ platforms. 🚀
Academic
AI Revolutionized Protein Science, but Didn’t End It
Google's AlphaFold revolutionized protein science by solving the protein folding problem, but it hasn’t replaced the need for biological experiments.
Large Language Models Are Not Search Engines
LLMs aren't search engines. They're prone to "hallucinations," predicting likely words, not facts. Companies must address this misuse.