|
| Together with |
 |
|
|
|
|
|
Hi there, this is your daily ☕️ Techpresso.
|
|
|
In today's Techpresso: |
|
🤖 Amazon launches an AI agent-building platform
📞 Google's AI can now make phone calls for you
🤝 OpenAI taps Google Cloud to power ChatGPT
⚠️ Top AI firms have 'unacceptable' risk management, studies say
🛒 OpenAI will take a cut of ChatGPT shopping sales
📉 Scale AI cuts 14 percent of staff after Meta deal
🎁 + 15 other news you might like
🔮 + 6 handpicked research papers and tools
|
|
FROM OUR PARTNER
7 predictions on AI, LLM, and Observability
Read the 7 key takeaways from Google's Director of AI and Datadog's VP of Engineering as they break down their predictions of the future:
- Smarter AI and LLM strategies for your org
- Building customer trust in AI outputs
- Scaling your tooling as LLM expertise grows
Navigate the complexities of AI adoption and avoid costly missteps with this expert guidance.
Future proof your AI strategy
| 🤖 Amazon launches an AI agent-building platform LINK
|
- Amazon Web Services launched Amazon Bedrock AgentCore, a new platform for businesses to build connected AI agents that can analyze internal data and write code.
- The service lets agents run for up to eight hours and supports MCP and A2A protocols, allowing them to communicate with agents outside a company's network.
- It was introduced as a tool to help organizations adopt agentic AI, freeing up employees from repetitive work to focus on more creative and strategic tasks.
| 📞 Google's AI can now make phone calls for you LINK
|
- Google Search can now call local businesses on your behalf to check prices, availability, and even make appointments or book reservations for you.
- The free AI calling feature is available in 45 US states, but subscribers to Google AI Pro and AI Ultra plans will get higher usage limits.
- For quality control, the automated calls will be monitored and recorded by Google, and local businesses are given an option to opt out of receiving them.
| 🤝 OpenAI taps Google Cloud to power ChatGPT LINK
|
- OpenAI now uses Google Cloud for cloud infrastructure, adding a new supplier to get the computing capacity needed for its popular large language models.
- The deal shows OpenAI's evolving relationship with Microsoft, which is no longer its exclusive cloud provider and is now considered a direct AI competitor.
- Google joins other OpenAI partners like Oracle and CoreWeave, as the company actively seeks more graphics processing units to power its demanding AI workloads.
| ⚠️ Top AI firms have 'unacceptable' risk management, studies say LINK
|
- A new study by SaferAI found that no top AI company, including Anthropic and OpenAI, scored better than "weak" on their risk management maturity.
- Google DeepMind received a low score partly because it released its Gemini 2.5 model without sharing any corresponding safety information about the new product.
- A separate assessment found every major AI lab scored a D or below on "existential safety," lacking clear plans to control potential future superintelligent machines.
| 🛒 OpenAI will take a cut of ChatGPT shopping sales LINK
|
- OpenAI reportedly plans to take a commission from sellers for sales made through ChatGPT, creating a new way to earn money from shopping features.
- The company is looking to integrate a checkout system directly into its platform, letting people complete transactions without navigating to an online retailer.
- Getting a slice of these eCommerce sales allows the AI startup to make money from its free users, not just from its premium subscriptions.
| 📉 Scale AI cuts 14 percent of staff after Meta deal LINK
|
- Scale AI is laying off 14 percent of its workforce, or 200 employees and 500 contractors, just one month after Meta purchased a major stake.
- CEO Jason Droege explained they ramped up GenAI capacity too quickly, which created inefficiencies, excessive bureaucracy, redundancies, and confusion about the team's mission.
- The data labeling company is now restructuring its generative AI business from sixteen pods to five and reorganizing the go-to-market team into a single unit.
|
|
FROM OUR PARTNER
Framer lets you ship fast-no devs, no bottlenecks.
Build polished, high-converting pages in hours using a real design canvas that publishes to the web instantly.
Perfect for waitlists, launch pages, MVPs, and more.
Get 25% off for 3 months with code TECHPRESSO
|
|
|
|
Other news & articles you might like
- Google brings Gemini 2.5 Pro and Deep Search to AI Mode LINK
- How Nvidia's Jensen Huang Persuaded Trump to Sell A.I. Chips to China LINK
- Denmark aims to host world's most powerful quantum computer LINK
- Intel to layoff more than 5,500 employees in US LINK
- Apple's first foldable iPhone tipped to feature 7.8-inch display, A20 Pro chip, and 48MP cameras LINK
- Slack is getting a host of new AI tools LINK
- Google finds custom backdoor being installed on SonicWall network devices LINK
- Anthropic draws investor interest at more than $100 billion valuation LINK
- Tesla is launching a longer, six-seat Model Y SUV in China this autumn LINK
- Jack Dorsey pumps $10 million into a nonprofit focused on open-source social media LINK
- Mercedes-Benz will let you use an in-car camera in Microsoft Teams while driving LINK
- 1 in 5 new video games on Steam now uses generative AI, report says LINK
- xAI is hiring an engineer to make anime girls LINK
- Perplexity CEO's warning to startups: big tech will "copy anything that's good" LINK
- Linux surpasses 5% market share on US desktops for the first time LINK
|
|
Trending research and tools
|
Wttr: a console-oriented weather forecast service that gives you weather information for any location directly in your terminal or web browser. LINK
|
|
Hand: an open-source robotic hand that provides dexterous finger movements at a moderate cost, with all motors contained within the hand itself. LINK
|
|
Metaflow: a framework designed to help scientists and engineers build and manage real-life AI and machine learning systems, streamlining the process from initial prototyping to production deployment. LINK
|
|
WaitLock: a command-line tool for shell scripts that provides locks to prevent multiple operations from conflicting with each other, with automatic cleanup when processes die. LINK
|
|
How Many Instruction Can LLMs Follow at Once?: this paper finds that large language models struggle to follow all instructions when given more than a few at once, especially those in the middle of the prompt. LINK
|
|
Mixture-of-Recursions: Learning Adaptive Token-Level Computation: this paper introduces a technique that lets a language model learn to adaptively apply more or less computation to each individual token based on its complexity. LINK
|
|
|
|
|
Want to get the latest news differently? |
|
|
|
|
