How Google AI’s New Method Reduces LLM Training Data by 10,000x

Google Research has unveiled a groundbreaking method for fine-tuning large language models (LLMs) that reduces required training data by up to 10,000 times. This innovative approach utilizes active learning to focus expert labeling on the most informative examples, especially “boundary cases” where model uncertainty is highest. In experiments with Gemini Nano models, this technique matched or surpassed the quality of models trained on 100,000 random labels with as few as 250 to 450 targeted examples. This development promises to make AI model development significantly leaner, more agile, and more cost-effective.

Sources:

Tesla Discontinues Dojo Supercomputer, Pivots AI Training Strategy

Tesla is shutting down its ambitious in-house Dojo supercomputer project, reassigning personnel and parting ways with project lead Peter Bannon. Launched in 2021, the Dojo supercomputer was engineered around Tesla’s custom D1 chip to process vast video datasets from its vehicle fleet for training full self-driving (FSD) models. This decision marks a significant strategic pivot away from developing its AI training infrastructure internally. The company is now expected to increase its reliance on external hardware suppliers like Nvidia and AMD for its AI computing needs.

Sources:

KPIT Invests $10M in Helm.ai to Advance Autonomous Driving Software

KPIT Technologies Ltd. has announced a strategic investment of up to $10 million in Helm.ai, a developer of advanced AI software for autonomous vehicles. This investment aims to empower original equipment manufacturers (OEMs) to create safe, scalable, and cost-effective autonomous driving systems. Helm.ai’s sophisticated software is designed to accelerate the development of various levels of driver assistance technologies, enhancing vehicle safety and capability.

Sources:

Aurrigo Shares Drop 35% Amid Revenue Warnings and Project Delays

Shares of transport technology firm Aurrigo International plummeted by 35% following a warning that its full-year revenues will fall short of market expectations. The company’s automotive division has been negatively impacted by US tariffs on UK vehicle production, creating sales volatility. Additionally, its autonomous vehicle division is experiencing delays in testing programs and contract tenders, pushing new projects into the next year. Despite these setbacks, the autonomous division reported a 41% revenue increase in the first half, bolstered by its ongoing project at Singapore’s Changi Airport.

Sources:

OpenAI Set to Launch GPT-5, Promising ‘PhD-Level’ Expertise

OpenAI is expected to launch GPT-5, the highly anticipated next generation of its flagship AI model, this week. Arriving over two years after GPT-4, the release is poised to demonstrate significant advancements in generative AI. During a livestream, OpenAI CEO Sam Altman hailed the new model as a “significant step along our path to AGI,” likening its capabilities to having a “PhD-level expert in anything… on demand.” Speculated features for GPT-5 include superior efficiency, enhanced problem-solving skills, and more advanced logic-based reasoning. This launch follows the recent release of two new open-weight models from the company, GPT-OSS-120b and GPT-OSS-20b.

Sources:

Google Rolls Out Major August Updates for Vertex AI and Google Photos

This August, Google has deployed a suite of updates across its generative AI products. For developers on Vertex AI, the prompt optimizer is now generally available, and supervised fine-tuning has been enabled for Gemini 2.5 Flash-Lite and Gemini 2.5 Pro models. The Model Garden now also provides access to OpenAI’s gpt-oss models. For consumers, Google Photos has gained new creative tools powered by the Veo 3 model, allowing users to transform photos into eight-second video clips with sound. A new ‘Create’ tab is also rolling out in the U.S. to centralize these tools. In a separate development, Google DeepMind has created an AI model named Aeneas to help historians restore and interpret ancient texts.

Sources:

SEALSQ Commits $30M to Develop Quantum-Resistant AI Security

SEALSQ Corp has established a $30 million investment capacity for its SEALQuantum.com Lab, dedicated to the convergence of quantum computing, cybersecurity, and AI. As a specialist in semiconductors and post-quantum technology, SEALSQ aims to pioneer quantum-resistant cryptographic solutions to safeguard AI models and data against emerging cyber threats. A key initiative includes a collaboration with ColibriTD to tackle sub-7nm chip manufacturing challenges using quantum-powered modeling. SEALSQ has also invested in ColibriTD and acquired semiconductor design firm IC Alps to bolster its post-quantum security offerings.

Microsoft and IBM Spearhead Quantum Computing Breakthroughs

Recent milestones from tech giants Microsoft and IBM underscore the rapid acceleration of the quantum computing industry. Microsoft, in collaboration with Atom Computing, achieved the first operational deployment of a Level 2 Quantum computer in July 2025 and aims to ship commercial quantum computers with over 1,000 physical qubits in 2025. The company is integrating quantum with AI via its Azure Quantum platform. Meanwhile, IBM has expanded its global footprint with the deployment of its Quantum System Two in Japan, following the successful 2021 launch of its 127-qubit Eagle processor. IBM provides quantum access through the IBM Cloud and its popular Qiskit software, fostering a network of over 250 partner organizations.

Sources:

OpenAI Releases gpt-oss-120b and gpt-oss-20b Open-Source Models

On August 5, 2025, OpenAI released two new open-source large language models: gpt-oss-120b and gpt-oss-20b. This marks the first time the company has open-sourced a language model since GPT-2, representing a major contribution to the open-source AI community and equipping developers with powerful new foundational tools.

Sources:

Windows 11 August Update Integrates AI into Settings and Enhances Recall

Microsoft is deploying its August 2025 Security Update for Windows 11 version 24H2, introducing several new AI-powered features. The update adds an AI agent to the Settings app, allowing users to find and adjust settings with natural language commands. For Copilot+ PCs, the Windows Recall feature now allows data to be exported or completely reset. Other improvements include a Quick Machine Recovery feature and user interface refinements to Snap Layouts.

Sources:

Microsoft Issues Urgent Patch for Actively Exploited SharePoint Flaws

Microsoft has released urgent security patches for SharePoint after previous fixes for critical vulnerabilities were successfully bypassed by attackers. The flaws, tracked as CVE-2025-53770 and CVE-2025-53771, have been actively exploited in ransomware campaigns. The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added the vulnerabilities to its Known Exploited Vulnerabilities catalog, mandating that federal agencies apply the updates immediately. Microsoft also advises organizations to rotate the machine keys on affected servers as an additional precaution.

Sources:

Google Launches OSS Rebuild to Bolster Open-Source Software Security

Google has introduced OSS Rebuild, a new open-source project designed to enhance the security and trustworthiness of open-source package ecosystems. The project works by reproducing upstream software artifacts to verify that the distributed code perfectly matches the original source code. This verification process helps to detect and expose potentially malicious code injected into widely used open-source packages, strengthening the software supply chain.

Sources: