OpenAI launched a new set of AI agent tools for developers

Hi pals! Here’s what you need to know about AI today:

👉 OpenAI launched new tools for developers to build sophisticated AI agents

👉 Hugging Face’s new self-driving dataset is here

👉 Meta is working on in-house AI chips

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 5 minutes

What’s going on: OpenAI has launched new tools, including the Responses API and Agents SDK, with the aim of empowering developers and enterprises to build sophisticated AI agents. These agents are designed to automate tasks such as web searches, file scanning, and website navigation, similar to OpenAI's Operator product. The Responses API, which will replace the Assistants API in 2026, provides access to AI models like GPT-4o search and GPT-4o mini search.

What does it mean: While these tools represent a significant step forward, OpenAI knows that there are very hard challenges in AI automation. AI-powered search tools, while more accurate than traditional AI models, are not immune to inaccuracies and may struggle with certain types of task/queries. The CUA model is still in early stages and prone to errors.

More details: 

• Developers can use the Computer-Using Agent (CUA) model to automate computer-based tasks, with enterprises having the option to run the CUA model locally.

• Agents SDK is an open-source toolkit for integrating models, implementing safeguards, and monitoring agent activities.

• Interested in details, benchmarks, pricing, and tutorials? Check out OpenAI’s official blog post.

What’s going on: Hugging Face, a very well-known AI development platform, is expanding its LeRobot platform through a collaboration with AI startup Yaak. This expansion introduces a new training set called Learning to Drive (L2D), designed for robots and cars capable of autonomous navigation in complex environments like city streets.

What does it mean: Building full self-driving cars needs a ton of high quality data. The L2D dataset, which exceeds a petabyte in size, includes data from sensors installed in German driving school cars. It captures a range of information, including camera footage, GPS data, and vehicle dynamics, as instructors and students navigate diverse scenarios such as construction zones, intersections, and highways. Unlike existing open self-driving training sets that often focus on specific planning tasks and require extensive annotations, L2D is engineered to facilitate end-to-end learning. This approach enables the prediction of actions directly from sensor inputs.

More details: 

• Several open datasets for self-driving cars, such as those from Alphabet's Waymo and Comma AI, exist. However, these datasets often emphasize planning tasks like object detection and tracking, which demand high-quality annotations.

• The data in the L2D was collected with identical sensor suites installed on 60 electric vehicles operated by driving schools in 30 German cities over the span of 3 years.

• To accurately represent functioning real-world driving situations, L2D creators have incorporated scenarios with a variety of environmental conditions, sensor malfunctions, construction areas, and inoperative traffic signals.

• Interested in the details? Read their official post on Hugging Face.

🖥 Meta is testing internally developed AI training chips, manufactured with TSMC, to decrease its dependence on suppliers like Nvidia and potentially reduce AI costs.

✍ OpenAI CEO Sam Altman announced the development of a new AI model that excels at creative writing in a recent long post on X, marking a potential expansion of OpenAI's focus beyond technical fields and into more artistic domains.

🤖 Dexterity, an AI robotics firm specializing in robots with human-like skills for industrial tasks, has raised $95 million, bringing its post-money valuation to $1.65 billion.

☁ Salesforce has announced a commitment to invest $1 billion in Singapore over the next five years to accelerate the adoption of its AI platform, Agentforce, which aims to enhance the country's digital transformation and address labor shortages.

👨‍🏫 Microsoft is supporting an AI transformation in Malaysia through cloud solutions and AI skills training initiatives like AIForMYFuture aiming to skill 800,000 people by the end of 2025.

⚖ Spain's government has approved a bill imposing fines of up to €35 million or 7% of a company's global turnover for failing to label AI-generated content with the aim of preventing the spread of deepfakes.

AI + Simulating Excel Sheets

Writer - Customer AI engineer

MongoDB - Senior AI Architect

Scale AI - ML Research Engineer, ML Systems

EarnIn - Staff Machine Learning Engineer (ML Platform)

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!