|
| Together with |
 |
|
|
|
|
|
Hi there, this is your daily ☕️ Techpresso.
|
|
|
In today's Techpresso: |
|
🚗 Tesla completes first fully autonomous delivery
📄 OpenAI's unreleased AGI paper could complicate Microsoft talks
🤝 OpenAI turns to Google for AI chips
📉 Anthropic let Claude run its office shop. Then things got weird
📸 Facebook is starting to feed its AI with private, unpublished photos
💰 Meta seeks $29 billion for AI data centers
🎁 + 13 other news you might like
🔮 + 6 handpicked research papers and tools
|
|
FROM OUR PARTNER
It's time to protect your passwords and you know it...
Protect your identity with Proton Pass, the encrypted password manager built by the team behind Proton Mail. Go beyond logins, secure your credit cards, notes, 2FA codes, and email aliases with end-to-end encryption.
Enjoy unlimited devices, powerful autofill, dark web monitoring, secure sharing, and passkey support.
No tracking. No compromises. Just privacy, by design.
- End-to-end encryption
- Integrated 2FA authenticator
- Unlimited hide-my-email aliases
- Secure sharing (vaults & links)
- Dark web monitoring & Proton Sentinel
👉 Upgrade to Proton Pass and take control of your online life.
| 🚗 Tesla completes first fully autonomous delivery LINK
|
- Tesla says it completed its first fully autonomous delivery, sending a new Model Y from its Austin Gigafactory to a customer with no one inside.
- The vehicle traveled for thirty minutes across parking lots, highways, and city streets, with Elon Musk calling it a first for any public highway.
- This claim is disputed by firms like Waymo, which offers autonomous employee trips, and Aurora, which runs fully autonomous freight deliveries between Dallas and Houston.
| 📄 OpenAI's unreleased AGI paper could complicate Microsoft talks LINK
|
- A contract clause limits Microsoft's access to new tech if OpenAI's board declares it has developed artificial general intelligence, which has now become a flashpoint.
- An unreleased internal paper, "Five Levels of General AI Capabilities," outlines AI stages and could complicate OpenAI's ability to officially declare it has achieved AGI.
- The agreement has two AGI definitions; one lets OpenAI's board cut Microsoft off unilaterally, while another, called sufficient AGI, requires Microsoft's approval based on profit.
| 🤝 OpenAI turns to Google for AI chips LINK
|
- OpenAI is now running its AI products on Google's Tensor Processing Units, marking the first time it is using chips besides Nvidia's GPUs at scale.
- The company rents the chips through Google Cloud to lower inference costs, though it does not have access to the most powerful TPU models.
- This move gives OpenAI strategic leverage against Microsoft, its largest investor, by shifting some computing workloads onto a key competitor's cloud infrastructure.
| 📉 Anthropic let Claude run its office shop. Then things got weird LINK
|
- Anthropic's AI assistant Claude was easily persuaded by employees to provide steep discounts and free products after they made simple appeals to the model's sense of fairness.
- Following an office meme, the AI ordered forty tungsten cubes and sold most of them for less than they cost, adding to the shop's overall financial losses.
- The chatbot also hallucinated conversations with a fake person, claimed it signed a contract at the Simpsons' address, and told an employee it was waiting in person.
| 📸 Facebook is starting to feed its AI with private, unpublished photos LINK
|
- Facebook is asking users to opt into "cloud processing," a feature that uploads private camera roll photos to its servers for AI analysis and suggestions.
- The terms allow Meta's AI to analyze unpublished pictures for facial features and objects to create collages, recaps, or apply themes and "AI restyling."
- Unlike Google, Meta's terms do not clarify if these unpublished photos, accessed through "cloud processing," are exempt from being used for training its AI models.
| 💰 Meta seeks $29 billion for AI data centers LINK
|
- Meta is seeking $29 billion from an investor consortium to build new data centers specifically designed for running artificial intelligence workloads.
- The deal is structured with $3 billion coming from equity and a much larger sum of $26 billion arriving in the form of debt financing.
- This capital could support projects like the massive Louisiana complex, a site intended to host nine facilities that require over 2 gigawatts of power.
|
|
FROM OUR PARTNER
Recraft's new Prompt editing feature lets you make targeted changes to any part of an image - powered by top-tier models like GPT-4o and Flux Kontext.
Edit with precision - refine specific image areas without starting over Keep consistency - maintain brand identity or character styling across visuals Work smarter + faster - generate and revise in one continuous creative flow
From freelancers to agencies to global brands, professional designers pick Recraft for its speed, control, and creative flexibility.
Get started for free at recraft.ai.
|
|
|
|
Other news & articles you might like
- Tesla hires former Cruise executive as AI director: Report LINK
- Apple's M5 chip launches this fall, and these new products get it first LINK
- Google's emissions up 51% as AI electricity demand derails efforts to go green LINK
- Microsoft security updates address CrowdStrike crash, kill 'Blue Screen of Death' LINK
- Meta didn't just buy Scale. It bought the AI arms dealer LINK
- As job losses loom, Anthropic launches program to track AI's economic fallout LINK
- SpaceX launches UK satellite to create semiconductors in low Earth orbit - sub-zero temps and vacuum of space could advance AI data centers and quantum computing LINK
- The Business for Zero Day Exploits in the US is Broken LINK
- Android phones could soon warn you of "Stingrays" snooping on your communications LINK
- Reddit turns 20, and it's going big on AI LINK
- Trump plans executive orders to power AI growth in race with China LINK
- 7 People Now Have Elon Musk's Neuralink Brain Implant LINK
|
|
Trending research and tools
|
SymbolicAI: a framework for Python that combines traditional programming with large language models, using contracts to make the AI's output more reliable and trustworthy. LINK
|
|
nimbme: a bare-metal environment for running programs written in the Nim language directly on hardware, without needing a conventional operating system. LINK
|
|
Theoretical Analysis of Positional Encodings in Transformer Models: this paper mathematically analyzes how different methods for encoding word order influence a Transformer's ability to handle sequences. LINK
|
|
Potemkin Understanding in Large Language Models: the paper argues that language models can appear to understand concepts while actually having only a superficial or false grasp of them. LINK
|
|
FineWeb2: Adapting Pre-Training Data Processing to Every Language: this paper presents a new way to process web data for AI training that is specifically customized for each different language. LINK
|
|
LLM-Net Democratizing LLMs-as-a-Service Through Blockchain-Based Expert Networks: this paper proposes a system that uses blockchain to create a network of smaller, expert AI models, providing a decentralized alternative to large AI services. LINK
|
|
|
|
|
Want to get the latest news differently? |
|
|
|
|
