Welcome to the Wild World of AI! (or, What Happens When Robots Get Too Clever)
Hey there, lovely humans! It’s great to see you again. Welcome to the TechCrunch AI circus, where the clowns are actually very smart machines, and the tightrope walkers are just trying not to trip over their own algorithms. This week’s newsletter is coming to you live from the TC Disrupt 2024 conference in sunny San Francisco. And let me tell you, we’ve got speakers more varied than my sock drawer – a delightful mix from the AI industry, academia, and policy. Grab your popcorn; this is going to be a show!
Featured Acts on Stage
In one corner, we have Devin Coldewey interviewing the CEO of Perplexity, Aravind Srinivas, about the search engine that’s serving up 100 million queries a week! Now that’s impressive – but hold on! The crowd’s gasping in horror, as News Corp’s Dow Jones is accusing them of “content kleptocracy.” It’s like the nasty breakup in an Alice in Wonderland-themed rom-com where everyone sues everyone!
Next up, we have Kirsten Korosec chatting with Jesse Levinson, co-founder of Zoox. They’re about to dive into how autonomous cars are preparing for their debut on our streets! Will they get lost Googling directions? Will they respond sarcastically when we need them the most? Stay tuned for the rollercoaster of human-robot interaction!
And let’s not forget the panel tackling the elephant in the room – you know, the one about how AI is flooding the internet with disinformation. Featuring big names like Pamela San Martin and Imran Ahmed, they’ll untangle how generative AI is being creatively misused, proving once more that giving the wrong people access to technology is like giving a hyperactive toddler a box of matches.
All the News That’s Fit to Print… or Not?
Now, let’s hop into the high-speed train of AI news. Apple has launched Apple Intelligence, the AI sidekick that promises to make your iPhone actually useful (no pressure!). Meanwhile, Bret Taylor’s customer service AI startup Sierra just raised a jaw-dropping $175 million. I mean, at this rate, I should just start flipping pancakes for a few bucks—because apparently, people are throwing money at anything with AI in the name!
Watch out! Generative AI could lead us to a delightful future filled with e-waste, potentially discarding 10 billion iPhones a year by 2030. Great, just what we needed; a “trophy wife” style of electronics, constantly updated and discarded after the newest release!
The Potpourri of Research Papers
This week, Google has unveiled their magic trick – a model called InkSight, which can turn photos of your poorly written notes into digital ink. Hooray for innovation! But before you get too excited, let’s hope it doesn’t end up signing any checks or legal documents – because we sure don’t need an AI with a career in fraud!
And speaking of impressive feats, Cohere released the Aya Expanse. These new multilingual models are reportedly smarter than Meta’s Llama (yeah, you heard that right!). They can probably outsmart your average party Alex Reed, so be careful about what you say around them!
The Grab Bag of Fun
OpenAI’s Advanced Voice Mode has rolled out to more users, making it easier for it to breathe down your neck in various accents. So now, if you feel lonely, just boot up your app. Just don’t let it whisper sweet nothings; we don’t need you falling for a chatbot (trust me, I tried that; it never ends well!).
So there you have it, folks! AI is advancing while we’re still struggling to remember our passwords. Join me on this wild ride as we navigate through the world of technology – just remember, all jokes aside, it’s the smart machines that hold the power now. And who knows, maybe one day they’ll make us all better comedians. Or they might just plot to take over the world. Either way, stay tuned!
Sign Off
Until next time, keep laughing at the absurdity of it all. And remember, if your coffee machine starts giving you life advice, maybe it’s time to upgrade before it starts sending you motivational quotes!
This is a conversational, cheeky summary that captures the essence of the article while adding a touch of humor reminiscent of the comedians mentioned. Enjoy the ride into the quirky world of AI!
Hiya, folks, welcome to TechCrunch’s regular AI newsletter. If you want this in your inbox every Wednesday, sign up here.
This week, the TechCrunch team, including yours truly, is on the ground at TC’s annual Disrupt conference in the bustling city of San Francisco. With an impressive array of speakers from the realms of AI industry, academia, and policy, I’ve decided to forgo my typical op-ed to give you a sneak peek of the fascinating content lined up for you.
My colleague Devin Coldewey is set to engage Perplexity CEO Aravind Srinivas in an onstage discussion. The cutting-edge AI-powered search engine has recently experienced a surge in popularity, reaching an impressive milestone of 100 million search queries weekly. However, it’s not all smooth sailing; the company faces a lawsuit from News Corp’s Dow Jones, which accuses it of operating a “content kleptocracy.”
In a highly anticipated fireside chat, Kirsten Korosec, TC’s transportation editor, will converse with Jesse Levinson, co-founder and CTO of Zoox. With a decade of experience in autonomous car technology, Levinson is gearing up the Amazon-owned robotaxi company for a new phase, and rest assured, we’ll be covering all the details of this exciting journey.
This week, a critical panel will address the escalating issue of AI-driven disinformation flooding the internet. Leading voices such as Meta Oversight Board member Pamela San Martin, Center for Countering Digital Hate CEO Imran Ahmed, and UC Berkeley CITRIS Policy Lab founder Brandie Nonnecke will delve into the dire implications of generative AI tools, which are being exploited by various actors, including state entities, to produce deepfakes and disseminate misleading information.
Furthermore, we anticipate insightful perspectives from Cara CEO Jingna Zhang, AI Now Institute co-executive director Sarah Myers West, and ElevenLabs’ Aleksandra Pedraszewska as they navigate the complex legal and ethical landscape surrounding AI. The rapid ascent of AI technologies is not without its ethical challenges, manifesting in a surge of lawsuits that jeopardize both burgeoning and established AI firms, alongside the creators and laborers feeding these advanced models. Expect this panel to bring clarity to these pressing issues and more.
That’s merely a taste of the enlightening discussions set to unfold this week. Expect appearances from an array of AI experts, including U.S. AI Safety Institute director Elizabeth Kelly, California senator Scott Wiener, Berkeley AI policy hub co-director Jessica Newman, Luma AI CEO Amit Jain, Suno CEO Mikey Shulman, and Splice CEO Kakul Srivastava.
News
Apple Intelligence launches: Through a recent software update, iPhone, iPad, and Mac users are now able to explore the innovative features of Apple’s AI-powered Apple Intelligence capabilities.
Bret Taylor’s startup raises new money: Sierra, co-founded by OpenAI chairman Bret Taylor, has successfully secured $175 million in a funding round, solidifying the startup’s valuation at $4.5 billion.
Google expands AI Overviews: Google Search’s AI Overviews, which provide a concise summary at the top of results pages, are now being rolled out to more than 100 countries and territories globally.
Generative AI and e-waste: A troubling projection from researchers suggests that the ever-growing computational demands of advanced AI models could lead to the generation of e-waste equivalent to over 10 billion discarded iPhones annually by the year 2030.
Open source, now defined: This week, the long-standing Open Source Initiative unveiled version 1.0 of its comprehensive definition for open source AI, aiming to clarify standards in this evolving landscape.
Meta releases its own podcast generator: Meta has rolled out an “open” version of the innovative feature that enables users to generate podcasts, originally popularized in Google’s NotebookLM.
Hallucinated transcriptions: Researchers have found that OpenAI’s Whisper transcription tool is facing hallucination problems, with instances of introducing inaccuracies ranging from racial commentary to fabricated medical treatments in its transcripts.
Research paper of the week
Google says it successfully trained a model to convert handwritten notes into “digital ink.”
The model, known as InkSight, was specifically designed to identify written words and produce strokes that closely mimic natural handwriting. Google researchers indicated that the aim was to “capture the stroke-level trajectory details of handwriting,” enabling users to seamlessly store the resulting strokes in their preferred note-taking applications.
InkSight isn’t without flaws; Google acknowledges it makes errors in transcription. Nevertheless, the company claims that the model can perform effectively across a diverse set of conditions, including challenging lighting scenarios.
Model of the week
Cohere for AI, the nonprofit research organization established by AI startup Cohere, has launched a groundbreaking family of text-generating models named Aya Expanse. These models are capable of composing and interpreting text in 23 distinct languages, with Cohere asserting that they surpass models like Meta’s Llama 3.1 70B in specific performance benchmarks.
The innovative technique dubbed “data arbitrage” played a crucial role in training Aya Expanse. Drawing inspiration from human learning methodologies, Cohere strategically selected highly proficient multilingual “teacher” models to generate synthetic training data for Aya Expanse.
While synthetic data has its drawbacks, some studies highlight that excessive reliance on it can lead to diminishing quality and diversity in models. However, Cohere believes that their data arbitrage approach effectively addresses these concerns. It will be interesting to see whether this claim withstands scrutiny over time.
Grab bag
OpenAI’s Advanced Voice Mode, the company’s sophisticated voice feature for ChatGPT, has now been made available for free in the ChatGPT mobile app across the EU, Switzerland, Iceland, and Norway in addition to Liechtenstein. Previously, only ChatGPT Plus subscribers could access this advanced functionality.
A recent article in The New York Times examined the pros and cons of Advanced Voice Mode, noting its tendency to lean on common tropes and stereotypes when attempting to engage with user requests. This voice mode has gained significant traction on TikTok for its impressive ability to mimic various voices and accents. However, experts express concerns regarding the potential for emotional dependency on such systems, which lack actual intelligence or empathy.
Or the text-generating models. This approach has enabled the Aya Expanse family to not only generate coherent and contextually relevant text but also to improve their understanding and interpretation of nuanced language across various cultures.
Cohere’s Aya Expanse models are designed with versatility in mind, making them suitable for a range of applications, from creative writing to customer support automation. As AI continues to evolve, models like Aya Expanse highlight the potential for enhanced multilingual capabilities that bridge communication gaps worldwide.
As the discussions at the TC Disrupt conference unfold and new innovations emerge, the AI landscape will undoubtedly continue to shift. We look forward to bringing you more insights from the event and the ongoing developments in the world of technology and artificial intelligence.
