Hi there, this is your daily ☕️ Techpresso.
In today's Techpresso:
💥 OpenAI unveils ChatGPT Atlas
🥽 Samsung launches its Vision Pro alternative
📉 iPhone Air orders slashed to almost ‘end of production’ levels
🛰️ SpaceX now sells laser tech to other companies
🔎 YouTube launches new AI tool to fight deepfakes
🤖 Meta downsizes its legacy AI research team
🎁 + 15 other news you might like
🔮 + 6 handpicked research papers and tools
💥 OpenAI unveils ChatGPT Atlas LINK
OpenAI unveiled ChatGPT Atlas, a new browser for macOS built on Chromium that layers ChatGPT support onto every website to provide contextual answers and inline writing suggestions.
A preview feature called agent mode allows the Computer-Use-Agent to perform complex tasks by clicking and navigating interfaces, though security experts have warned about the potential risks involved.
By launching a browser, OpenAI aims to embed ChatGPT into daily user habits, giving the company a direct channel to desktops similar to how Google distributes services through Chrome.
🥽 Samsung launches its Vision Pro alternative LINK
Samsung's new Galaxy XR headset is a Vision Pro alternative selling for $1,800 and was made in partnership with Google for the software and Qualcomm for its main chip.
The headset’s most impressive function is Google's Gemini assistant, which lets you control apps or generate video with your voice, a feature that the competing Vision Pro does not have.
A major downside is that Gemini runs in the cloud, transmitting everything you do on the device to Google's servers since the company lacks Apple’s private cloud technology.
📉 iPhone Air orders slashed to almost ‘end of production’ levels LINK
According to a Nikkei report, Apple is drastically cutting its production orders for the new iPhone Air, slashing them to levels that are typically seen only near a product's end of production.
Multiple sources indicate that the significant reduction in iPhone Air manufacturing is a direct result of weak demand in several markets, despite the device only recently becoming available to customers in China.
The weak performance of the iPhone Air is reportedly being compensated for by unexpectedly robust demand for other iPhone 17 models, keeping overall production in line with the company’s original forecasts.
🛰️ SpaceX now sells laser tech to other companies LINK
SpaceX is providing its mini laser terminals to startup Muon Space, giving its satellites a persistent optical broadband connection to the global Starlink constellation for real-time data.
SpaceX is offering companies a 25Gbps laser connection between two satellites up to 4,000 kilometers apart, a system designed to eliminate communication delays with their own hardware.
SpaceX will also provide its laser system to aerospace company Vast for its Haven-1 private space station, expanding Starlink's commercial offerings beyond just providing internet service.
🔎 YouTube launches new AI tool to fight deepfakes LINK
YouTube's new likeness detection technology has rolled out to eligible Partner Program creators, allowing them to find and request removal of AI-generated content that features their face or voice.
To use the tool, creators must go to the "Likeness" tab and verify their identity by submitting a photo ID and a brief selfie video through a QR code process.
Once the system grants access, a user can view all detected videos and then submit a removal request under privacy guidelines, make a copyright request, or archive the clip.
🤖 Meta downsizes its legacy AI research team LINK
Meta is axing around 600 roles from its legacy Fundamental AI Research unit, known as FAIR, and its AI product and infrastructure division to reduce the team's size.
While cutting these legacy positions, the company continues to hire workers for its newly formed superintelligence team, a separate group which is being called the TBD Lab.
The firm plans to integrate and scale many research ideas from FAIR into the larger model runs being conducted by the new TBD Lab, according to an internal memo.
Other news you might like
- From Prince Harry to Steve Bannon, hundreds call for ban on AI superintelligenceLINK
- Anthropic in talks with Google for $10B cloud computing dealLINK
- Folding iPad release may have been delayed until 2029LINK
- Netflix goes ‘all in’ on generative AI as entertainment industry remains dividedLINK
- Zuckerberg, Mosseri and Spiegel ordered to testify in child social media trialLINK
- Apple and Google may be forced to change app stores in UKLINK
- Jaguar Land Rover struggling 8 weeks after most expensive UK cyberattackLINK
- Baidu to test its robotaxis in Switzerland as the race for self-driving taxis in Europe heats upLINK
- Microsoft breaks, then quickly fixes Windows Recovery Environment bug that bricked USB input devicesLINK
- Several users reportedly complain to FTC that ChatGPT is causing psychological harmLINK
- Smart bed owners experience AWS nightmare as outage leaves them sweating and stuck in upright positionLINK
- Google embraces vibe coding with latest version of AI Studio app development platformLINK
- Unseeable prompt injections in screenshots: more vulnerabilities in Comet and other AI browsersLINK
- Airbnb CEO Chesky says ChatGPT isn't 'quite robust enough' to integrate into travel appLINK
- Hackers compile personal data on 22,000 US officials using stolen Salesforce recordsLINK
Latest research and tools
MinIO: has stopped distributing free Docker images for its latest security release, with users unable to find the new version on container registries like quay.io or DockerHub.LINK
rlsw: a small, software-based OpenGL renderer created for the Raylib game development library.LINK
The Dragon Hatchling: this paper introduces a model that acts as a link between artificial intelligence transformers and computational models of the brain.LINK
Knocker: a self-hosted tool that keeps home services private, granting temporary access only to authorized IP addresses after receiving a secret knock request.LINK
Why Can't Transformers Learn Multiplication?: these AI models struggle with multiplication because their architecture is designed to find patterns in data rather than learning the precise, algorithmic steps of arithmetic.LINK
OptPipe: Memory- and Scheduling-Optimized Pipeline Parallelism for LLM Training: this paper introduces a system that makes training large language models more efficient by optimizing memory use and task scheduling.LINK
Want to get the latest news differently? Find us on:
See you tomorrow for a new dose of ☕️ Techpresso!
.png){kind=small}
.png){kind=small}
.png){kind=small}