Welcome to the fourth edition of Neoteric AI News Digest! Curious what we have for you this time? From Google’s efforts to fix its new AI-powered search feature to Elon Musk’s bold stance against Apple’s AI integration plans, there’s plenty to catch up on.
Regarding Apple, there’s one hot news topic you won’t find below: their WWDC24. The reason is simple — it’s everywhere, dominating almost every front page, so there’s no point in repeating it all again here. Instead, we focused on news that might have slipped your notice amid the flood of Apple’s announcements.
Grab a coffee and join us as we unravel some intriguing AI stories of the past two weeks!
Table of Contents
Google AI Overview Needs Work: Addressing Early Failures
In the last issue of Neoteric News Digest, we covered Google’s new AI-powered search feature’s (epic) fail. After Google Overview surprised its first users with a mix of odd, inaccurate, and unhelpful results, the internet was flooded with memes and criticism highlighting the poor quality and misinformation spread by the feature.
Liz Reid, Google’s VP and Head of Search, acknowledged these problems in a blog post titled “About last week.” She explained that errors arose from misinterpreted queries, nuances of language, and limited accurate information available online — despite extensive testing conducted by the company before the launch. Additionally, she pointed out that some viral examples were either faked or based on nonsensical queries, which the AI handled poorly.
Google assures it is now working on improvements, including better detection for nonsensical queries, limiting user-generated content for responses, and refining AI triggers for health and news searches. Reid noted that having millions of users helps identify these issues quickly, as Google continues to iterate and improve the feature.
While the giant’s mistakes are undoubtedly laughable, this story underscores an important challenge of AI implementation we all should learn from. Whether it’s a feature in an internal system or a large-scale project like Google’s, AI-powered software always requires thorough testing and continuous iteration.
For more on the initial rollout issues, you can read about it in our previous issue. For the latest updates, check out the full article on TechCrunch.
DuckDuckGo Enhances AI Privacy: Anonymous Access to GPT-3.5 and Other Models
Remember the concerns regarding data privacy and security in AI models we discussed in the previous issue? This week, we came across some promising news on a solution that addresses (at least some of) them.
DuckDuckGo — a privacy-focused search engine that prioritizes user anonymity — has introduced private AI chats, allowing users to use popular AI models anonymously. The method involves calling AI models on behalf of users, effectively masking their identities by removing IP addresses and metadata from queries.
Moreover, the company has agreements with AI providers to delete stored chats within 30 days and prevent their use in training models. This means that while AI providers may temporarily store chats, they cannot trace them back to individual users nor utilize the information they include to further train their models.
Currently, DuckDuckGo’s service includes access to AI models such as GPT-3.5 Turbo, Claude 3 Haiku, Meta’s Llama 3 70B, and Mixtral 8x7B. Users can try this feature for free within daily limits, with plans for a paid version offering higher limits and access to more advanced models in the future.
Wanna dig deeper? Read the full article on The Verge and DuckDuckGo’s blog post.
AI Powers Growth in Tech Manufacturing
The AI boom is driving significant growth for tech component manufacturers. Nvidia’s market cap recently soared to $3.01 trillion, making it the second most valuable company globally. This surge is fueled by the high demand for Nvidia’s AI hardware, particularly its H100 GPUs, which dominate the AI chip market with a 70-95% share. Nvidia plans to launch new AI chips annually, starting with the Blackwell B200 GPU.
Raspberry Pi is also making strides in the AI space. Partnering with Hailo, they introduced an AI add-on for the Raspberry Pi 5 microcomputer. This AI Kit, featuring Hailo’s Hailo-8L M.2 accelerator, allows the microcomputer to run AI applications natively. This move highlights a broader trend of integrating AI into portable devices, making AI capabilities more accessible and energy-efficient.
Meanwhile, tech giants like Google, Intel, Microsoft, Meta, AMD, Hewlett-Packard Enterprise, Cisco, and Broadcom have formed the Ultra Accelerator Link Promoter Group. This alliance aims to create an open standard for AI accelerator chips, challenging Nvidia’s proprietary NVLink. The first version, UALink 1.0, promises improved speed and reduced latency; however it is expected to be available in the next two years, which, as PC Gamer fairly pointed out, gives Nvidia quite a head start.
For more details, check out the full article on PC Gamer and these two on The Verge: one about Nvidia and one about Raspberry Pi.
Reachy2: An Open-Source Chore Robot
Recently, a post by Joanna Maciejewska with a comment on AI’s capabilities went viral, making people worldwide nod their heads. It goes: “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.” — and as much as we love building AI-powered software, we can’t help but admit it is a pretty good point. Luckily, there are people already dedicated to addressing this need!
In an exciting collaboration, Hugging Face and Pollen Robotics have unveiled Reachy2, an open-source humanoid robot designed to assist with household chores. This innovative project is part of Hugging Face’s Le Robot program, led by former Tesla robotics engineer Remi Cadene. Reachy2, developed by Pollen Robotics in Bordeaux, France, is equipped to perform various tasks such as placing cups on a dish rack and handing objects to humans.
The training process for Reachy2 is as fascinating as its capabilities. Initially, a human controlled the robot via a virtual reality headset, guiding it through chores (placing cups on a dish rack and handing apples to people). These sessions were recorded, creating a dataset that a machine learning algorithm used to train the robot. After extensive training, Reachy2 can now independently perform these tasks, showcasing the potential of combining VR and AI in robotics training.
Cadene shared the project’s progress on X (see the above screenshot), emphasizing the open-source nature of the dataset and the trained model (available on Hugging Face and GitHub). This initiative aims to democratize advanced robotics, allowing enthusiasts and developers to replicate and enhance Reachy2’s capabilities on smaller robots.
This development comes at a time of significant investment in practical robotics, with companies like Figure and Cobot also advancing the field. Hugging Face’s open-source approach could pressure these competitors by providing advanced robotics technology for free.
For more details on Reachy2 and its development, check out the article on VentureBeat.
Frustrated Artists Flock to Cara: The Anti-AI Social Platform
If anyone can wholeheartedly agree with the viral post mentioned earlier, it’s artists and creators. For them, the problem extends further with numerous cases of AI-related breaches of their intellectual property rights. A huge part of this problem is companies like Meta, allowing public posts to train generative AI systems, despite artists using these platforms (for years now) to showcase their work.
There wouldn’t be an issue if the giant showed some respect and offered an easy opt-out option. The thing is, they allow such luxury only to European users — and not out of respect, but due to the European GDPR laws.
Fed up with Meta’s AI rules, artists are migrating to Cara, an artist-run, anti-AI social platform. Founded by photographer Jingna Zhang, it offers a unique, safe space for all artists to share their work without fear of it being used to train AI models. Due to the recent migration spike, Cara surged from 40,000 to 650,000 users in one week, quickly topping the App Store charts.
Zhang, who has a history of fighting for artists’ rights (including her own), designed Cara to combine features of Instagram and X, allowing artists to host portfolios and post updates. The platform also partners with the University of Chicago’s Glaze project, adding protection against AI scraping.
Cara’s rapid growth hasn’t come without challenges. Fully bootstrapped and supported by volunteers, the platform faced a $96,280 bill from web hosting company Vercel due to the sudden user increase. Despite the hurdles, Cara’s rise shows a strong demand for an artist-friendly platform where AI isn’t a threat — and perhaps suggests Meta should rethink its approach and start supporting users rather than disregarding them.
Curious to know more? Read the full article on TechCrunch.
Musk Threatens to Ban Apple Devices Over OpenAI Integration
Elon Musk has threatened to ban Apple devices from his companies if OpenAI’s AI software is integrated into Apple’s operating systems, calling it a security risk. This follows Apple’s announcement that customers will soon access OpenAI’s ChatGPT through Siri as part of new AI features.
“If Apple integrates OpenAI at the OS level, then Apple devices will be banned at my companies,” Musk wrote on X, the social network he owns. “That is an unacceptable security violation,” he continued, adding that visitors would need to check their Apple devices at the door, storing them in a Faraday cage to block electromagnetic fields.
Apple’s presentation assured users that privacy protections are built in, stating that ChatGPT requests are not stored by OpenAI and users’ IP addresses are obscured. Despite these assurances, Musk remains skeptical, claiming Apple cannot effectively manage data once it’s handed over to OpenAI.
It may be a fairly valid concern, but it’s worth noting here that Musk, who co-founded OpenAI but is no longer affiliated with it, is currently developing his own AI startup, xAI, and a chatbot named Grok. In light of that, a question arises: are Musk’s threats coming out of genuine concern, or rather out of hope to slow down the competition?
For more on this developing story, read the full article on Bloomberg.
***
That wraps up this edition of Neoteric AI News Digest! We hope these stories have given you valuable insights into the current trends and challenges in AI. If so, don’t hesitate to share this issue with your network and come back for more in two weeks!
Want to make sure you don’t miss out?
Sign up below, and we’ll notify you of new releases.