AIR#5 - March 9th, 2024
As the sun peeks over the digital horizon, today's edition of AIR: The AI Recon brews a potent blend of stories that will jolt your morning routine with a shot of AI espresso.
We're navigating through the tempest of public opinion with a startling revelation that over half the globe is giving AI the side-eye, questioning the trust and trajectory of our silicon companions. Meanwhile, the AI gold rush is on, with ChatGPT leading the charge into uncharted territories, promising riches for those who can stake their claim in the tech frontier. Amidst governance dramas, OpenAI's captains, Altman and Brockman, secure their helm with an expanded crew, while Google's Pixel 8 finds itself adrift, unable to harness the full power of its AI sails.
On the developer deck, the winds of change are blowing with new AI models challenging the reign of GPT-4, and innovations like AIConfig and SkyServe promising to streamline our digital endeavors. For the scholarly navigators among us, the seas are rich with discoveries on AI's reasoning capabilities, the ethical whirlpools of closed-source LLMs, and the mirage of AI hallucinations.
As we set sail into today's AI odyssey, let's ponder the regulatory compass guiding us through these turbulent waters, ensuring our voyage remains true to the stars of privacy, governance, and innovation. So, grab your cup and let's embark on this journey together, exploring the vast and vibrant landscape of AI that lies ahead.
Business
The Public Is Rapidly Turning Against AI, Polling Shows
Global trust in AI plummets; 53% now skeptical, highlighting a need for responsible innovation and management.
The AI Gold Rush
AI's current boom mirrors a gold rush, with ChatGPT leading the frenzy and hinting at both immediate hype and long-term shifts in tech and markets.
Review completed and Altman, Brockman to continue to lead OpenAI
Altman & Brockman to stay at OpenAI's helm after review; board expands with 3 new members to bolster governance & mission focus.
Google says the AI-focused Pixel 8 can't run its latest smartphone AI models
Despite Google's hype, the Pixel 8 can't run its latest AI model, Gemini Nano, due to hardware limits, unlike the Pixel 8 Pro and Samsung S24.
The Fear That Inspired the Creation of OpenAI
Musk's fear of AI dominance by giants like Google led to OpenAI's creation, aiming for open, beneficial AI development, now facing internal disputes.
Employers are willing to pay up to 44% more for AI-skilled workers in IT
AI skills boost salaries up to 44% in IT, with Asia-Pacific markets leading in AI adoption and salary increases. Demand for AI talent outpaces supply.
Amazon's generative AI bot Rufus makes online shopping easier
Amazon's AI bot Rufus revolutionizes online shopping, making it quicker and easier to find and compare products.
Engineering
AIConfig Extension for VS Code
Transform VS Code into a generative AI hub with AIConfig Editor: Run any AI model, manage prompts, and integrate seamlessly into your code.
Four new models that are benchmarking near or above GPT-4
GPT-4's reign ends as four new AI models from Google, Mistral, Claude, and Inflection benchmark near or above, shaking up the AI landscape.
Claude Sonnet and Pi AI give exact same response to a prompt
InflectionAI's Pi AI mirrors Claude-3-Sonnet's response, sparking debate on originality and AI capabilities.
I Reduced Our LLM Costs by over 85%
Pierson Marks slashed Jellypod's AI costs by 85% by fine-tuning open-source models, reimagining AI use, and optimizing architecture.
SkyServe: A cost-efficient, multi-region/cloud library for serving GenAI models
SkyServe slashes AI model serving costs by 50%, offering multi-region, multi-cloud deployment for optimal GPU use and high availability.
Torvalds Speaks: Impact of Artificial Intelligence on Programming [video]
Linus Torvalds discusses AI's transformative role in programming in a new video.
Writer.com's Graph-Based RAG Alternative to Vector Retrieval
Writer.com innovates with a graph-based RAG, ditching vector retrieval for semantic graphing, aiming to enhance AI's data context and accuracy.
Matrix multiplication breakthrough could lead to faster,more efficient AI models
New breakthrough in matrix multiplication promises to speed up AI models by addressing inefficiencies, marking the biggest leap in over a decade.
Academic
Machine learning and information theory concepts towards an AI Mathematician
Bengio explores bridging AI's gap in math reasoning, focusing on discovery over proof, through machine learning and information theory.
Can Large Language Models Reason and Plan?
Exploring if AI can self-correct like humans, Kambhampati's study questions LLMs' reasoning and planning abilities.
Data Contamination and Evaluation Malpractices in Closed-Source LLMs
New study exposes data leaks and evaluation flaws in top closed-source LLMs, urging for transparent practices.
Hallucination in Visual AI Models
Oras Al-Kubaisi warns: Visual AI models like Opus & GPT-4 can "hallucinate" errors in data, even in plausible-looking results. Verify carefully!
Exposing Systemic Vulnerabilities of LLMs Through a Prompt Hacking Competition
Global prompt hacking contest reveals critical LLM vulnerabilities, challenging AI security with 600K+ adversarial prompts.
What U.S. Members Think About Regulating AI
IEEE survey reveals U.S. tech professionals demand stronger AI regulation, prioritizing data privacy and governance on par with health and environment.