Why machine translation still struggles with the world's most popular languages; what's next for generative video; Adobe bets on ethical AI; inside Stability AI's implosion
Chatbot hype hits reality; the fight for AI talent; Databricks' DBRX is the most powerful open source LLM; Ray Ban smart glasses get an AI upgrade; Sakana AI introduces evolutionary AI algorithm;
In an article for The Atlantic this week, Louise Matsakis (check out her recently launched newsletter You May Also Like!) discusses how advances in generative AI are leading to the proliferation of automatic translation tools that could make learning new languages seem less necessary and therefore cause a decline in enrollment for foreign language courses in the United States and other countries.
Louise describes how she used an AI app to create a deepfake video of herself speaking fluent Chinese, despite only having basic skills in the language, and interviews experts who argue that something profound may be lost when we rely solely on machine translation: the ability to truly understand other cultures and ways of thinking that comes from learning a new language.
While AI excels at technical translation, it has historically lacked the cultural context and nuance that humans provide. Traditional machine translation tools have struggled with metaphors, cultural references, and choosing the most accurate phrasing. But newer AI models based on transformer architectures are improving machine translation by going beyond simple word swaps. They learn the intricate dance of language – syntax, context, and cultural references – resulting in translations that are more natural and accurate. This paves the way for real-time conversation translation tools that can bridge the gap between languages instantaneously.
The benefits are undeniable: global collaboration can flourish, fostering innovation and cultural exchange. Imagine attending a virtual conference where every participant can understand presentations regardless of their native tongue. Language-based discrimination could become a thing of the past, as job markets open up and cultural understanding deepens.
However, a world without language barriers is not without its complexities. There's a beauty in the poetry and humor specific to each language. Will constant translation homogenize these linguistic treasures? Furthermore, cultural identity is often interwoven with language. Will the ability to seamlessly switch languages dilute this connection?
But perhaps the most interesting hurdle for machine translation lies in low-resource languages – those with limited written materials and digital presence. Historically, the term low-resource languages was used to describe relatively obscure tongues spoken by small populations. With too little example data to learn from, large language models systems struggle with word ordering, idioms, context, and nuanced meaning so the resulting translations for low resource languages are often wrong or simply incomprehensible. For example, in October 2023, Instagram started adding the word “terrorist” to the biographies of some of its users because it incorrectly translated a word from Arabic.
Tiktok failed to load.Enable 3rd party cookies or use another browser
For years, there has been an assumption in the field of machine translation that some of the world's most widely spoken languages like French, Spanish, Russian and Arabic have ample data for AI systems to learn from. However, AI researchers I’ve spoken to have shared a worrying trend: even languages considered mainstream and high-resource are at risk of becoming low-resource.
Major demographic, cultural, and technological shifts are putting some of the world's linguistic giants in jeopardy. The rise of regional dialects and slangs, accelerated by social media and digital subcultures, has fractured languages like Arabic into numerous divergent varieties. While a mother tongue may have abundant learning data overall, its vernacular offshoots are becoming low-resource languages unto themselves.
Furthermore, script styles and new character sets are evolving faster than ever. Regional writing systems for the Arabic script proliferate new glyphs and ligatures adapted for speed and new vocabularies. These don't match established encodings, creating low-resource problems within high-resource languages.
Most concerning is the decline of linguistic diversity worldwide as globalizing forces lead to language extinction at unprecedented rates. Therefore, as native speakers dwindle, these major world languages are now at risk. For example, in cities such as Riyadh, Dubai or Abu Dhabi, English has become the go-to language for young people, creating a situation where many people from the Middle East struggle to write or even speak in Arabic.
The low-resource crisis extends beyond machine translation. As languages lose speakers and stop evolving new vocabulary and usage examples, training data for all language AI like chatbots, video captioning, and text summarization will grow stale and inadequate. Already, we've seen preliminary effects with technical jargon and academic domains where English training data is relatively low-resource compared to general-use English corpora. Enterprise language models struggle with esoteric fields from medicine to finance despite English's high-resource reputation.
Solutions are urgently required to preserve linguistic resources and dynamically adapt machine learning to evolving low-resource languages and dialects. Otherwise, the machine translation systems we take for granted today will become increasingly error-prone and useless across more areas of modern communication.
Despite these complexities, the potential of AI-powered real-time translation is undeniable. The road ahead involves creating technology that preserves the richness of languages while fostering global understanding.
And now, here are this week’s news:
❤️Computer loves
Our top news picks for the week - your essential reading from the world of AI
MIT Technology Review: What’s next for generative video
MIT Technology Review: How Adobe’s bet on non-exploitative AI is paying off
The Atlantic: The End of Foreign-Language Education
WSJ: Want to Know if AI Will Take Your Job? I Tried Using It to Replace Myself
The Information: This New Orleans museum uses AI to allow visitors to speak with WWII veterans
Washington Post: AI hustlers stole women’s faces to put in ads. The law can’t help them.
New York Times: Meta’s Smart Glasses Are Becoming Artificially Intelligent. We Took Them for a Spin.
FT: How Silicon Valley’s ‘Oppenheimer’ found lucrative trade in AI weapons
New Yorker: The Lifelike Illusions of A.I.
WSJ: The Fight for AI Talent: Pay Million-Dollar Packages and Buy Whole Teams
VentureBeat: Sakana AI’s evolutionary algorithm discovers new architectures for generative models
MIT Technology Review: The tech industry can’t agree on what open-source AI means. That’s a problem.
Wired: Inside the Creation of the World’s Most Powerful Open Source AI Model
⚙️Computer does
AI in the wild: how artificial intelligence is used across industry, from the internet, social media, and retail to transportation, healthcare, banking, and more
MIT Technology Review: How three filmmakers created Sora’s latest stunning videos
MIT Technology Review: AI could make better beer. Here’s how.
Business Insider: AI is helping fragrance companies unlock the sensational possibilities of smell
9to5Google: Gemini updated to automatically start Google Maps navigation
FT: Media groups look to AI tools to cut costs and complement storytelling
The Economist: Artificial intelligence is taking over drug development
FT: AI is accelerating the energy transition, say industry leaders
Business Insider: A Microsoft-powered medical AI spotted cancer in 11 women where doctors didn't
Bloomberg: Visa Adds New AI Tools to Help Fight Digital Fraud on Payments
TechCrunch: Vibrant Planet uses AI for land mapping and improving climate resiliency
The Verge: Google adds ratings to go with its new AI tools for shoppers
The Guardian: AI to track hedgehog populations in pioneering UK project
Reuters: Romanian state agency turns to AI to help farmers tap EU funds
VentureBeat: OpenAI shows off first examples of third-party creators using Sora
🧑🎓Computer learns
Interesting trends and developments from various AI fields, companies and people
VentureBeat: AI21 Labs juices up gen AI transformers with Jamba
The Verge: How a Windows shake-up could position Microsoft to capitalize on AI PCs
VentureBeat: Google DeepMind unveils 'superhuman' AI system that excels in fact-checking, saving costs and improving accuracy
Reuters: Musk's Grok-1.5 AI chatbot to be available next week
Fast Company: Researchers tap AI and patternless collections to help predict catastrophic failure
VentureBeat: SambaNova announces new AI Samba-CoE v0.2 that already beats Databricks DBRX
Business Insider: Inside Big Tech's nasty battle for coveted AI talent
VentureBeat: Microsoft launches new Azure AI tools to cut out LLM safety and reliability risks
VentureBeat: Lightning AI launches next-gen AI compiler ‘Thunder’ to accelerate model training
Reuters: French company Valeo to use more Google Cloud AI tools
Business Insider: Some of Silicon Valley's biggest names are wading into the battle for AI expertise
Bloomberg: Amazon Bets $150 Billion on Data Centers Required for AI Boom
Business Insider: Microsoft customers complain Copilot doesn't work as well as ChatGPT. Microsoft says they're not using it right.
Business Insider: Mark Zuckerberg says Nvidia CEO Jensen Huang is basically the Taylor Swift of tech
Business Insider: AI is making gadgets weird again — and it could radically redesign your smartphone
Business Insider: Palmer Luckey says Anduril is working on AI weapons that 'give us the ability to swiftly win any war'
Business Insider: Using AI in classrooms could help or hurt students — that's why educators should teach tech literacy, scholars say
The Telegraph: BBC defends replacing actress with AI for voiceover
MIT Technology Review: Four things you need to know about China’s AI talent pool
WSJ: Companies Are Seeking Real-World Supply-Chain Gains in New AI Tools
Reuters: New AI benchmark tests speed of responses to user queries
Bloomberg: Salesforce Paid $20 Million for the Face of Its AI Strategy
TechCrunch: Adobe’s Firefly Services makes over 20 new generative and creative APIs available to developers
Wired: Is AI the Future of NPCs?
VentureBeat: Nvidia triples and Intel doubles generative AI inference performance on new MLPerf benchmark
The Verge: OpenAI is experimenting with sharing revenue with builders in its GPT Store
TechCrunch: Adobe’s GenStudio brings brand-safe generative AI to marketers
CNBC: CEO of Chinese AI company 4Paradigm discusses business outlook under U.S. sanctions
Reuters: Musk's xAI to enable chatbot Grok for all premium subscribers of X
FT: Financial services counting on AI for a productivity boost
Business Insider: Big tech's desperate scramble for AI talent
FT: Academics express confidence that they and AI can work together
Business Insider: Stability AI founder's jokes about Satya Nadella's influence have a dark truth to them
Business Insider: Sergey Brin personally called a Google employee to convince them to turn down a job at OpenAI: report
Business Insider: Watch the crazy AI short films and videos created by artists with early access to OpenAI's Sora tool
Business Insider: Why some creators are limiting or stopping their use of AI tools
The Economist: The AI doctor will see you…eventually
The Economist: Medical AIs with human faces are their way
The Economist: Can artificial intelligence make health care more efficient?
MIT Technology Review: Apple researchers explore dropping “Siri” phrase & listening with AI instead
WSJ: CFOs Tackle Thorny Calculus on Gen AI: What’s the Return on Investment?
Business Insider: Managers are worrying that their salaries will get cut because of AI, a survey found
Axios: The future of AI: Personalized systems tuned to your needs, Amazon exec says
The Verge: Financial Times tests an AI chatbot trained on decades of its own articles
TechCrunch: Large language models can help home robots recover from errors without human help
The A.I. Boom Makes Millions for an Unlikely Industry Player: Anguilla
The Information: A Booming Nvidia Supplier Says AI Costs Need to Drop
Business Insider: Sam Altman may have Siri and Alexa in his sights after OpenAI filed a 'digital voice assistant' trademark application
TechCrunch: Can you hear me now? AI-coustics to fight noisy audio with generative AI
Reuters: Behind the plot to break Nvidia’s grip on AI by targeting software
Reuters: Can artificial intelligence extend healthcare to all?
Reuters: Tokyo, Tokyo, make me a match! Metropolis hopes AI app will spur marriages
The Verge: What our shopping haul taught us about the promise of AI
The Information: Meta Pursues AI Talent With Quick Offers, Emails From Zuckerberg
Wired: Large Language Models’ Emergent Abilities Are a Mirage
Fast Company: One year in, Khan Academy’s AI has 65,000 students, and is still learning new skills
The Verge: AI-generated blues misses a human touch — and a metronome
VentureBeat: Pure Storage, Nvidia partner to democratize AI with new infrastructure solutions
Bloomberg: The Economist Who Believes AI Will Be Great for the Middle Class
Keep reading with a 7-day free trial
Subscribe to Computerspeak by Alexandru Voica to keep reading this post and get 7 days of free access to the full post archives.