AIR#82 - May 25, 2024

Good morning, AI enthusiasts! As you sip your morning brew, prepare for a whirlwind of AI news that's sure to stir your curiosity. Leading today's edition is a fascinating study from Chicago Booth researchers who found that GPT-4 outperforms human analysts in financial statement analysis. This breakthrough not only showcases AI's potential in the financial sector but also raises questions about the future role of human expertise in data-driven industries.

But it's not all about academic triumphs. The Washington Post is making headlines with its bold pivot to AI amid financial struggles, aiming for "AI everywhere in our newsroom." This move underscores the growing reliance on AI to drive innovation and efficiency in traditional media, even as it grapples with the challenges of maintaining journalistic integrity. Meanwhile, Google finds itself in hot water as its AI search engine delivers bizarre suggestions like using glue to keep cheese on pizza, sparking a debate on the reliability of AI-generated content.

As we explore these stories, let's not forget the broader implications they bring to the table. From the alarming statistic that 52% of ChatGPT's programming answers are wrong, yet still preferred by many for its politeness, to the intriguing launch of Perplexica, an open-source AI search engine, today's edition is packed with insights that challenge our understanding of AI's role in society. So, sit back, sip your coffee, and let's dive into the dynamic world of artificial intelligence together.

Business

The Washington Post Pivots to AI Amid Financial Struggles
The Washington Post pivots to AI to recover from $77M loss, aiming for "AI everywhere in our newsroom," says CEO Will Lewis.

Google AI Suggested Using Glue to Keep Cheese on Pizza
Google's AI suggested using glue to keep cheese on pizza. Users question AI reliability after bizarre advice. 🍕🤖

Ex-DeepMind Scientists Raise $220M for Paris-Based AI Startup H
Ex-DeepMind scientists raise $220M to launch Paris-based AI startup H, aiming to develop advanced agentic AI models for complex tasks.

Self-Hosting: The Ultimate Shield for Your Data Against AI Models
Self-hosting your software is the best way to keep your private data safe from being used to train AI models.

🔥 Google Search's AI Hallucinations Exposed
Google's AI in Search is hallucinating, giving bizarre answers like glue-topped pizza and false info about Obama. Trust is at stake.

🔥 Google's AI Search Keeps Mistaking Satire for Real News
Google's AI search mistakes satire for real news, citing The Onion as a credible source. Rough start for AI-generated results.

OpenAI Releases Former Employees from Controversial Exit Agreements
OpenAI retracts controversial NDAs, allowing former employees to keep vested equity and lifting non-disparagement clauses.

Sam Altman Under Fire as OpenAI Faces Safety Controversies
Sam Altman faces mounting pressure as OpenAI grapples with safety controversies and criticism from former executives and AI experts.

I Uploaded 11 Years of My Journal to GPT-4o: A Life-Changing Experience
Darryl Kelly uploaded 11 years of his journal to GPT-4, creating a profound family experience and deeper connections with his children.

$6M Fine for AI-Cloned Biden Voice in Robocalls
FCC fines scammer $6M for using AI to clone Biden's voice in illegal robocalls during NH primary, marking a crackdown on high-tech scams.

Engineering

52% of ChatGPT's Programming Answers Are Wrong, Study Finds
Study: 52% of ChatGPT's programming answers are wrong, yet many programmers still prefer its polite, comprehensive responses.

🔥 [GitHub] Perplexica: Open-Source AI Search Engine Alternative to Perplexity AI
Perplexica is an open-source AI search engine alternative to Perplexity AI, offering enhanced privacy and advanced query modes.

[GitHub] Eddie: Retro AI-Powered CLI Assistant by Mirascope
Eddie: Mirascope's retro AI-powered CLI assistant offers educational resources and open-source tools for building AI apps. Try it now!

Binarizing CLIP for Multimodal Retrieval and Ranking
Binarizing CLIP boosts multimodal retrieval efficiency but hurts performance. Training with pseudo-quantization mitigates this, preserving 87-93% accuracy.

Google's "AI Overview" Delivers Dangerous Misinformation
Google's AI Overviews are spreading false and dangerous info, often mistaking jokes for facts, raising concerns about AI reliability.

Introducing RAG 2.0: Revolutionizing Enterprise AI
Introducing RAG 2.0: Enterprise AI redefined with end-to-end optimization, outperforming previous models in benchmarks and real-world tasks. Join the waitlist!

GitHub Accelerator 2024: 11 Open-Source AI Projects to Watch
GitHub Accelerator 2024 welcomes 11 innovative open-source AI projects, driving global AI advancements with $400K in support.

Searchable ChatGPT: Instantly Search Your ChatGPT History Locally
Searchable ChatGPT extension: Instantly search your ChatGPT history locally with no ads, tracking, or external data usage. 🔍⌨️

[Medium] Implementing Long-Context RAG with RAPTOR by InfiniFlow
InfiniFlow's RAGFlow v0.6.0 integrates RAPTOR for improved long-context retrieval, enhancing data quality and answer accuracy in AI.

Open-source SDK for AI Code Interpreters by E2B
E2B launches open-source SDK for secure AI code interpreting in Python & JavaScript, enhancing AI app development.

Academic

🔥 [Paper] Financial Statement Analysis with GPT-4 by Chicago Booth Researchers
Chicago Booth researchers find GPT-4 excels at financial statement analysis, outperforming human analysts and state-of-the-art ML models.

Can AI Save Indonesia's 700 Languages?
Indonesia uses AI to preserve its 700 languages, digitizing texts to train multilingual LLMs and prevent extinction.

Who Will Open Source AlphaFold3? Scientists Race to Crack AI Model
Scientists race to open-source AlphaFold3 after DeepMind delays code release, aiming to boost drug discovery and maintain open science.

Journal of Computational Linguistics | MIT Press
MIT Press's Computational Linguistics journal: latest NLP research, high impact factor, exclusive insights for academia and industry.

OpenAI’s Voice Assistant: Why Stick with Scarlett Johansson?
OpenAI's new voice assistant mimics Scarlett Johansson's "Her" character, sparking debate over its lack of innovation despite tech marvels.