Good morning! This is your daily ☕️ Techpresso.
In today's Techpresso:
💥 Google found guilty of monopoly in search
🤖 OpenAI co-founder leaves for AI rival Anthropic
🦾 Figure unveils new sleeker and smarter humanoid robot
🎥 Nvidia used 'a lifetime' of videos everyday to train AI
🍎 Leaked code reveals Apple Intelligence's plan to prevent hallucinations
🎁 + 6 other news you might like
🔮 + 1 handpicked research papers and tools
💥 Google found guilty of monopoly in searchLINK
A federal judge has ruled that Google violated US antitrust laws by maintaining a monopoly in the search and advertising markets, specifically in general search services and text advertising.
This ruling marks a significant win for the Department of Justice, although some claims were dismissed, and indicates that Google is liable for monopolistic practices under Section 2 of the Sherman Act.
Future proceedings will determine potential remedies for Google's monopolistic behavior, with possibilities ranging from business practice changes to a breakup of its search operations, while Google plans to appeal the decision.
🤖 OpenAI co-founder leaves for AI rival AnthropicLINK
When OpenAI was founded in 2015, it had 11 founding members, but only three remain, including CEO Sam Altman, Wojciech Zaremba, and Greg Brockman.
Greg Brockman announced on Monday that he would take a sabbatical until the end of the year, marking the extended absence of one of the remaining cofounders.
John Schulman, another cofounder, recently left OpenAI to join its rival Anthropic, making him the eighth cofounder to exit since the company's inception.
🦾 Figure unveils new sleeker and smarter humanoid robotLINK
Figure has introduced its new humanoid robot, the Figure 02, which features improved hardware and software, including six RGB cameras and enhanced CPU/GPU computing capabilities.
Leveraging a longstanding partnership with OpenAI, the Figure 02 is equipped for natural speech conversations, featuring speakers and microphones to facilitate communication with human co-workers.
Figure 02's advanced AI and language processing aim to make interactions transparent and safe, which is crucial given the robot's potential use alongside humans in factory and commercial environments.
🎥 Nvidia used 'a lifetime' of videos everyday to train AILINK
Nvidia collected videos from YouTube and other sites to create training data for its AI products, as shown by internal documents and communications obtained by 404 Media.
Nvidia asserted that their data collection practices align with both the letter and spirit of copyright law when questioned about legal and ethical concerns regarding the use of copyrighted material.
A former Nvidia employee revealed that workers were directed to gather videos from sources like Netflix and YouTube to train AI for the company's 3D world generator project, internally referred to as Cosmos.
🍎 Leaked code reveals Apple Intelligence's plan to prevent hallucinationsLINK
Leaked code for macOS Sequoia 15.1 has revealed pre-prompt instructions for Apple Intelligence to minimize hallucinations and improve accuracy in responses.
These pre-prompt instructions include directives for Apple Intelligence to ensure questions and answers in mail assistance are concise and relevant to avoid false information.
Instructions also specify limitations for creating photo memories, prohibiting religious, political, harmful, or provocative content to maintain a positive user experience.
Other news you might like
X is reportedly closing its San Francisco office.LINK
Samsung starts mass production for memory chips for on-device AI.LINK
Microsoft is binding employee bonuses and promotions to security performance.LINK
Nvidia's subscription software empire is taking shape.LINK
Tesla rival Lucid gets another $1.5 billion cash injection from Saudi Arabia to keep going through 2025.LINK
Uber highlights autonomous vehicle efforts now that Tesla’s in its rearview mirror.LINK
Latest research and tools
Karpathy/Nano-Llama31: a minimal, dependency-free version of the Llama 3.1 model architecture, enabling simple training, finetuning, and inference with significantly lighter dependencies compared to the official Meta and Hugging Face implementations.LINK
Want to get the latest news differently? Find us on:
See you tomorrow for a new dose of ☕️ Techpresso!
-min.png){kind=small}
.png){kind=small}
