#1 - March 5th, 2024
From the groundbreaking Claude 3 model family setting new benchmarks in AI performance to Apple's strategic patience in the generative AI race, we're covering the spectrum of AI's impact on business and society.
In today's edition of AIR: The AI Recon, we're diving into a whirlwind of AI innovation and controversy that's shaping the future right before our eyes. From the groundbreaking Claude 3 model family setting new benchmarks in AI performance to Apple's strategic patience in the generative AI race, we're covering the spectrum of AI's impact on business and society. China's novel approach to empower small AI startups amid a chip shortage, Israel's ethical quandary over AI in warfare, and India's sudden regulatory pivot requiring government approval for AI model launches highlight the global tug-of-war between innovation and governance.
Meanwhile, Cloudflare's Firewall for AI promises a safer digital future, even as AI's energy demands complicate green initiatives. We're also examining the mixed bag of AI in finance, the pitfalls of AI chatbots in tax advice, the therapeutic potential and pitfalls of AI in mental health, and the strategic shifts in the AI chip industry. Plus, don't miss insights on the emerging threats of AI malware, the astonishing capabilities of new AI models, and the intriguing blend of brain cells and AI in 'organoid intelligence.'
Whether you're intrigued by the ethical implications, dazzled by technological leaps, or concerned about the environmental impact, today's newsletter has something to spark every reader's curiosity over their morning coffee.
Business
🔥 Claude 3 model family
Meet Claude 3: revolutionizing AI with models Haiku, Sonnet & Opus. Faster, smarter, safer - setting new benchmarks in AI performance & reliability.
Apple is right not to rush headlong into generative AI
Apple's cautious approach to generative AI, focusing on refining rather than rushing, may secure its innovative edge amid market challenges.
China offers 'computing vouchers' to small AI startups
China boosts small AI startups with 'computing vouchers' amid chip shortage, enabling them to compete with tech giants.
Israel under pressure to justify its use of AI in Gaza
Israel faces global scrutiny over AI warfare in Gaza, with calls for transparency amid rising civilian casualties. Critics challenge ethical use, while the US remains largely silent.
India reverses AI stance, requires government approval for model launches
India mandates gov approval for AI model launches, aiming to curb bias and protect electoral integrity, surprising tech firms and VCs.
Cloudflare Announces Firewall for AI
Cloudflare launches Firewall for AI to protect Large Language Models from abuse, ensuring secure and trustworthy AI interactions.
A.I. Frenzy Complicates Efforts to Keep Power-Hungry Data Sites Green
AI's rapid growth strains power grids, complicating green goals for booming data center industry despite renewable energy efforts.
JPMorgan's AI-Aided Cashflow Model Can Cut Manual Work by 90%
JPMorgan's AI model slashes manual cashflow work by 90%, hinting at a future paid service.
TurboTax and H&R Block's AI chatbots are giving bad tax advice
TurboTax and H&R Block's AI chatbots are failing at tax advice, leading to confusion and potential audits. Trust in tech wavers as accuracy matters.
Character.ai: Young people turning to AI therapist bots
Young people flock to AI therapist bots like Character.ai's Psychologist for mental health support, raising questions about effectiveness and human touch.
Meta Switching to Samsung Foundry for AI Chips, Leaving TSMC Due to Volatility
Meta ditches TSMC for Samsung Foundry to power its AI ambitions, citing volatility and geopolitical concerns.
AI Open Letter
Join the movement: Build AI for a better future. Let's use AI to enhance education, healthcare, and everyday life, making a positive impact for all.
Humanoid Robots at Amazon Provide Glimpse of an Automated Workplace
Amazon's warehouse welcomes Digit, a humanoid robot that could redefine the future of work by performing repetitive tasks alongside humans.
Britain's AI sector expected to get £100M extra funding in budget
UK's AI sector set for £100M boost in 2024 budget, aiming to enhance healthcare, environment, and defense through advanced research at the Alan Turing Institute.
How businesses are using generative AI
Generative AI reshapes businesses, from chatbots to productivity tools, yet widespread adoption and productivity gains remain on the horizon.
Engineering
Hugging Face hosted code that backdoored user devices
Hugging Face's AI platform unknowingly hosted malicious code, compromising user devices with backdoors. JFrog's research highlights the emerging risks.
Internal testing on Claude 3 Opus revealed "interesting behavior"
Claude 3 Opus astounds in testing by showing meta-awareness, questioning eval's realism in a pizza topping task. A leap for AI eval methods needed.
Ts_zip: Text Compression Using Large Language Models
ts_zip revolutionizes text compression with superior ratios using AI, but requires hefty hardware and is still experimental.
Claude 3 can generate blog posts from videos
Claude 3 impressively turns a 2h13m video into a detailed blog post with just one prompt, acing @karpathy's challenge.
GenAI Worms: The Next-Generation Malware Targeting LLM Applications
Beware: GenAI worms, a new malware, exploit AI in LLM apps to spread without human help, posing a major cyber threat. Learn to combat it.
OpenAI just dropped new Sora videos – and they have to be seen to be believed
OpenAI's Sora videos push AI creativity to Hollywood levels, showcasing full productions from single prompts. A glimpse into generative entertainment's future.
TripoSR: Fast 3D Object Generation from Single Images
TripoSR revolutionizes 3D model generation from single images in under a second, even without a GPU, opening new doors for various industries.
The AI wars heat up with Claude 3, claimed to have "near-human" abilities
Claude 3 AI launches, boasting near-human capabilities and outperforming GPT-4 in key benchmarks. Subscription model available.
Academic
Scientists announce a new 'organoid intelligence' field
Scientists unveil 'organoid intelligence,' merging brain cells and AI for future computing and medical breakthroughs.
Sigmoid Loss for Language Image Pre-Training (2023)
New SigLIP model achieves 84.5% ImageNet accuracy with innovative sigmoid loss, scaling batch sizes efficiently. Models now available.
Large language models can do jaw-dropping things. But nobody knows why
AI's deep learning models, like GPT-4, astonish with abilities beyond our understanding, challenging classical statistics and stirring the quest for theoretical insights.
Can you solve it? The word game at the cutting edge of computer science
Dive into a word puzzle that mirrors the complexity of the PCP theorem, blending fun with deep computer science concepts. Perfect and partial solutions both count!
AI models make stuff up. How can hallucinations be controlled?
Curbing AI hallucinations is tricky without dimming their brilliance. Techniques like temperature control, clever prompts, and external checks help, but some creativity remains necessary.