Hello everyone, welcome to TechCrunch's regular AI newsletter. If you'd like to have this sent to your inbox every Wednesday, sign up here.
This week, the TechCrunch staff (including you) will be attending TC's annual Disrupt conference in San Francisco. We have speakers from the AI industry, academia, and policy, so instead of the usual editorials, we thought we'd preview some of the great content.
My colleague Devin Coldewey will be interviewing Perplexity CEO Aravind Srinivas on stage. This AI-powered search engine is thriving, recently reaching 100 million queries per week. But it is also being sued by News Corp's Dow Jones over what the publisher describes as “content theft politics.”
Kirsten Korosec, TC's transportation editor, chats by the fire with Zoox co-founder and chief technology officer Jesse Levinson. Levinson, who has been at the forefront of self-driving car technology for a decade, is now gearing up for his next big adventure with Amazon's robotaxi company, which he plans to report on in the future.
There will also be a panel discussion on how AI is flooding the web with disinformation, including Pamela San Martin, member of the Meta Monitoring Board, and Imran Ahmed, CEO of the Center to Counter Digital Hate. , Brandi Nonnek, founder of the CITRIS Policy Lab at the University of California, Berkeley. The three will discuss how, as generative AI tools become more widely available, they are being misused by a variety of actors, including nation-state actors, to create deepfakes and spread disinformation. .
And we'll hear from Cara CEO Jingna Zhang, AI Now Institute Co-Executive Director Sarah Myers West, and ElevenLab's Aleksandra Pedraszewska about the legal and ethical minefield of AI. The rapid rise of AI is creating new ethical dilemmas and exacerbating old ones, while the number of lawsuits continues to decline. This threatens both emerging and existing AI companies and the creators and workers who power the models. The panel tackles all this and more.
This is just a sampling of this week's decks. AI Safety Institute Director Elizabeth Kelly, California State Senator Scott Wiener, Berkeley AI Policy Hub Co-Director Jessica Newman, Luma AI CEO Amit Jain, Suno CEO Mikey Shulman, Splice CEO Kakul Srivastava, and more Experts are expected to appear.
news
Launch of Apple Intelligence: Through a free software update, iPhone, iPad, and Mac users will have access to the first set of Apple Intelligence features powered by Apple's AI.
Brett Taylor's startup raises new funding: Sierra, the AI startup co-founded by OpenAI Chairman Brett Taylor, raises $175 million in a funding round that values the startup at $4.5 billion. raised dollars.
Google expands AI Overview: Google Search's AI Overview, which provides a snapshot of information at the top of results pages, is starting to roll out in more than 100 countries and territories.
Generative AI and e-waste: Due to the vast and rapidly evolving computing requirements of AI models, the industry could dispose of e-waste equivalent to more than 10 billion iPhones annually by 2030. Researchers predict that there will be.
Open Source, a New Definition: The Open Source Initiative, a long-standing organization dedicated to defining and “governing” all things open source, this week released version 1.0 of its definition of open source AI.
Meta releases its own podcast generator: Meta releases an “open” implementation of its viral podcast generation feature in Google’s NotebookLM.
Hallucinatory transcription: Researchers say OpenAI's Whisper transcription tool has hallucination issues. Whisper reportedly introduced everything from racial comments to imagined treatments into the record.
This week's research paper
Google has announced that it has taught a model that converts handwritten photos into “digital ink.”
The InkSight model was trained to recognize words written on a page and output strokes that closely resemble handwriting. According to Google researchers who worked on the project, the goal was to “capture details of handwritten stroke-level trajectories” and allow users to save the resulting strokes in their favorite note-taking app.
Image credit: Google
InkSight isn't perfect. Google points out that there is a mistake. However, the company also claims that this model performs well in a variety of scenarios, including difficult lighting conditions.
I hope it's not used to forge signatures.
this week's model
Cohere for AI, a nonprofit research lab run by AI startup Cohere, has released a new family of text generation models called Aya Expanse. The model can write and understand text in 23 different languages, and Cohere claims it outperforms models including Meta's Llama 3.1 70B on certain benchmarks.
Kohia said a technique called “data arbitrage” was key to training Aya Expanse. Taking inspiration from the way humans learn by going to different teachers to gain unique skills, Cohere uses a particularly talented multilingual “teacher” to generate synthetic training data for Aya Expanse. Selected model.
Synthetic data is problematic. Some research suggests that over-reliance on this can lead to model quality and diversity deteriorating over time. However, Kohia says data arbitrage effectively alleviates this. We'll soon see whether this claim stands up to scrutiny.
grab bag
OpenAI's Advanced Voice Mode, a realistic-sounding voice feature for OpenAI's ChatGPT, is now available for free on the ChatGPT mobile app for users in the EU, Switzerland, Iceland, and Norway, in addition to Liechtenstein . Previously, users in these regions needed to subscribe to ChatGPT Plus to use advanced voice modes.
A recent article in the New York Times highlights the pros and cons of Advanced Voice Mode, including its reliance on metaphors and stereotypes to communicate in the way users want. Advanced Voice Mode has exploded in popularity on TikTok for its incredible ability to mimic voices and accents. But some experts warn that it could lead to emotional dependence on a system that has no intelligence or, for that matter, empathy.