#2 - March 6th, 2024
Grab your favorite mug and settle in because today's edition of AIR: The AI Recon is brimming with the frothiest AI concoctions from around the globe.
From Sergey Brin's candid admission on Google's Gemini gaffe to the dark art of AI deepfakes meddling in democracy, we're stirring the pot on all things AI. The gambling world's risky bet on AI has ethicists raising their eyebrows, while Google vows to declutter our search results from AI-generated spam, promising a cleaner digital future.
As the Gemini debacle unfurls its cultural insensitivity worldwide, OpenAI and the NYT lock horns in a legal showdown that could redefine content rights. Meanwhile, Nvidia asserts its dominance in the server realm, hinting at a seismic shift in AI hardware demand. But that's not all - from the tantalizing promise of generative AI in advertising to the geopolitical chess game over AI chip supremacy, we've got your curiosity covered.
And for our developer friends, innovations like BASED, StableDiffusion 3 and optical computing promise to revolutionize coding, image generation and AI efficiency. So, sip slowly and savor the rich blend of stories we've brewed for you today – it's a full-bodied blend of innovation, controversy, and transformative potential that's sure to kickstart your day.
Business
Sergey Brin says in rare appearance that company 'messed up' Gemini image launch
Sergey Brin admits Google "definitely messed up" the Gemini image launch, highlighting the challenges and potential of AI's evolving landscape.
AI deepfakes come of age as billions prepare to vote in a big year of elections
AI deepfakes challenge democracy as billions vote in 2024, sparking global debate on digital integrity and election security.
Concern as the gambling industry embraces AI
AI's integration in gambling sparks debate: Enhances user experience but raises addiction and ethical concerns. Critics call for responsible use.
Google is starting to squash more spam and AI in search results
Google fights web spam by updating search algorithms, aiming to reduce unhelpful AI and SEO content by 40%. Big changes ahead for cleaner searches.
Google's Gemini problem will be even worse outside the U.S.
Google's AI, Gemini, faces backlash for racial gaffes; issue may worsen globally as it struggles with cultural sensitivity and stereotypes.
Fighting words: a closer look at OpenAI's 35 page motion to dismiss
OpenAI counters NYT lawsuit with bold claims of evidence manipulation, igniting a fiery legal battle over AI-generated content rights.
Beware of the Newsroom-Destroying "Answer Engine"
AI "answer engines" like Perplexity threaten newsrooms by satisfying readers' queries directly, cutting off vital traffic and revenue streams for journalism.
Nvidia now plays kingmaker in the server court, says Omdia
Nvidia dominates server market as AI demand spikes, making GPUs crucial for growth. Server sales up, shipments down as firms invest in AI hardware.
The AI Wave Starts Lifting Dell and HPE
Dell and HPE ride the AI wave with improved GPU allocations, signaling a rebound in server spending and a stronger position in the AI server market.
AMD Hits US Roadblock in Selling AI Chip Tailored for China
AMD's AI chip for China hits a US export barrier, caught in tech crackdown despite being tailored for lower performance.
Has the A.I. Hype Driven 'Magnificent Seven' Stocks Beyond the 90s Tech Bubble?
AI hype boosts 'Magnificent Seven' stocks, stirs 90s tech bubble fears, but transformative potential remains. Experts caution, markets watch.
Generative AI: Transforming Advertising Beyond Imagination
Generative AI redefines advertising, enabling rapid, imaginative ad creation across platforms, transforming storytelling and marketing efficiency.
Demand for chips fuelled by AI could reshape global politics and security
AI-driven chip demand could shift global politics and security, with US and Taiwan at the core and China fast catching up. Global investment and tech bottlenecks highlight the strategic value of semiconductors.
Engineering
🔥 Based: Simple linear attention language models
BASED boosts AI language model efficiency with a simple design, achieving 24x throughput over Transformers & excelling in recall tasks.
Generative AI and the widening software developer knowledge gap
Generative AI risks eroding software engineering's foundation, creating a knowledge gap as automation takes over basic tasks. #AI #SoftwareDevelopment
Prompt injection and jailbreaking are not the same thing
Confusing prompt injection with jailbreaking? They're not the same! One attacks applications, the other targets LLM safety filters. Know the difference to protect your AI.
Claude 3 translates a low-resource language from a few thousand examples
Claude 3 astounds by mastering Circassian translation with minimal data, outperforming specialized models and marking a breakthrough for low-resource languages.
LoRAX and Outlines: Better JSON Extraction with Structured Generation and LoRA
LoRAX v0.8 integrates Outlines for precise JSON extraction, blending structured generation with LoRA fine-tuning for high-quality, schema-adherent outputs. Perfect for AI-driven data structuring tasks.
Local code completion with Ollama and Cody
Cody and Ollama revolutionize coding with local AI assistance, making code completion accessible offline. Perfect for air-gapped devs!
Optical computing can perform matrix computations at the speed of light
Revolutionary photonic chip accelerates AI training, promising computations at light speed and enhanced security against hacks.
Like grep but for natural language questions. Based on Mixtral 8x7B
Discover fltr: grep for natural language queries using Mixtral 8x7B, transforming text search with AI.
Microsoft: Copyright law didn't stop the VCR and shouldn't stop the LLM
Microsoft defends AI training with NYT content, likens copyright clash to VCR's past legal battles, arguing for technological progress.
You can even teach it to play Mahjong: Claude 3 leads GPT4 across the board
Claude 3 outperforms GPT-4, showcasing superior reasoning in complex tasks like Mahjong, hinting at a competitive AI landscape ahead.
Academic
🔥 Stable Diffusion 3: Research Paper
Stable Diffusion 3 sets new benchmarks in text-to-image AI, outshining DALL·E 3 & others in typography & prompt accuracy with its novel MMDiT architecture.
🔥 Elliptic curve 'murmurations' found with AI
AI unveils unexpected 'murmurations' in elliptic curves, mirroring starling flocks, revolutionizing math and cryptography understanding.
Stop Treating AI Models Like People (2023)
AI isn't sentient: don't confuse chatbot responses with human emotions or intelligence. They're just algorithms, not friends or therapists.
Can AI Solve Science?
AI enhances science by leveraging computational reducibility, offering new tools for discovery and practical applications, though its fundamental exploratory power is outmatched by irreducible computations.
Wikipedia Wisely Downgrades CNET Reliability Score After Lazy AI Screw Ups
Wikipedia downgrades CNET's reliability due to AI-generated content mishaps, highlighting journalism's AI integration challenges.
Cognition Is All You Need – The Next Layer of AI Above Large Language Models
Cognitive AI: The new framework set to revolutionize AI by enabling complex reasoning, beyond large language models' capabilities.
AI passes 100 IQ (Claude 3)
Claude-3 AI surpasses human average IQ, hitting over 100 in a landmark achievement. A new era of intelligence begins.
GraphRAG: Improving RAG performance using LLM-generated knowledge graphs
Microsoft Research's GraphRAG leverages LLM-generated knowledge graphs for superior Q&A on unseen data, outperforming traditional methods in complex analyses.
Geoffrey Hinton: Will digital intelligence replace biological intelligence? [video]
Geoffrey Hinton explores if AI will surpass human intelligence in latest Romanes Lecture on YouTube.