AI 日报 — 2026-08-19

Liquid AI has released new QAD Q4_0 GGUFs, which are updated 4-bit checkpoints for LFM2.5-230M, LFM2.5-350M, LFM2.5-1.2B-Instruct, and LFM2.5-2.6B models. These allow developers to run LFM2.5 models at Q4_0 memory and speed without the typical quality degradation. The QAD Q4_0 checkpoints were compared against post-training quantization (PTQ) GGUFs across benchmarks like GPQA Diamond, MMLU-Pro, IFEval, IFBench, Multi-IF, BFCLv4, GSM8K, and AIME25.

OpenAI is partnering with CodeAI to prepare students for a future with AI. While most students already use AI, only 16% of high school leaders report that all their students are learning the technical knowledge to understand it. A CodeAI survey found 75% of high school students believe AI understanding will be crucial for their futures. This partnership aims to help young people use AI responsibly, critically evaluate its outputs, and understand its limitations, ensuring they become thoughtful participants in shaping AI's future.

Anthropic has detailed two experiments demonstrating Claude's capability to accelerate protein design and analytical chemistry. The company plans to launch an access program specifically for scientists, aiming to help life scientists increase the pace of their research. These results highlight Claude's potential to significantly contribute to scientific advancements by streamlining complex processes in these fields.

Next week, ChatGPT Ads will expand to 31 European countries, including Germany, France, Spain, Italy, Sweden, Norway, Denmark, the Netherlands, and Austria. This expansion marks significant progress in six months, as ChatGPT Ads continues to develop new formats, optimization tools, and measurement solutions. The focus remains on broadening access to powerful AI, protecting user trust, and helping businesses achieve their goals in these new markets.

OpenAI reportedly revoked several security researchers' access to a limited-access program designed for cybersecurity research, which removes some restrictions on using its AI tools. OpenAI confirmed that this revocation was due to an error. This program allows researchers to utilize OpenAI's AI tools with fewer limitations for their cybersecurity studies.

Modern web software often serves the majority of users with static features, leaving a long-tail of unmet needs. While traditional trace waterfall diagrams are useful for request/response systems, modern applications with stateful or dynamic elements, like non-deterministic agents or durable workflow engines, require more flexible visualization. Trace spans offer a good foundation, but extensibility for custom visualizations is crucial. This extensibility, however, becomes challenging for software handling sensitive data like customer records or financial transactions.

AI is transforming how democratic governments safeguard their citizens by enhancing national security through capabilities like preventing cyberattacks, securing critical infrastructure, and early threat detection. OpenAI is committed to collaborating with oversight institutions to pilot tools that enable authorized reviewers to examine records of AI-assisted government decisions, including inputs, outputs, and tool usage, to understand AI's impact on outcomes. These tools will be interoperable or model-agnostic, and participating institutions will maintain control over evidence and findings, accessible only to authorized government officials.

Replit and OpenAI share a goal of making software creation accessible to everyone, regardless of technical background. Replit, an early user of GPT-3, is now leveraging GPT-5.6 Luna to power its Free Mode. This move demonstrates how the improved price performance of the GPT-5.6 series and recent OpenAI price cuts are enabling broader access to advanced intelligence for software development, narrowing the gap between an idea and a working product.

A recent session on August 17, 2026, detailed how Claude Code enabled macOS to natively print to the HP Laser 1008a, a printer HP never officially supported on Mac. This involved reverse-engineering the SPL3 raster language and running HP's real codec within a Linux container. The process, completed in approximately four hours in one sitting, was documented in a lightly redacted account.

Google has announced new study tools, including a student hub, notebooks, and interactive 3D visualizations within Gemini. These AI-powered resources are designed to assist college students with studying and researching. Additionally, Google is offering special student offers for its Google AI plans, providing eligible students with access to these advanced tools to enhance their academic endeavors.

Block, the technology company founded by Jack Dorsey, has released Berd, a new desktop application. Berd was developed to provide Block's employees with a unified environment for interacting with AI agents across various models. The application has been released under the Apache 2.0 license, making it an open-source tool for broader use and collaboration.

Google is introducing a dedicated student hub for Gemini, offering eligible US students one year of Google AI Pro for free. This includes 5TB of storage, Google Health Premium, increased Gemini usage limits, and Gemini access across Google apps. International students will receive the Google AI Plus subscription, which provides 400GB of storage and lower Gemini usage limits.

Google has introduced new AI-powered study tools across its Search and Gemini platforms. These enhancements include AI-generated interactive visuals, 3D simulations, a dedicated student hub, and customized practice quizzes. The announcement, made on Wednesday, aims to provide students with a comprehensive suite of resources to aid their learning and study processes.

Z.ai has announced pricing for its GLM-5.3 API access, maintaining the same rates as its predecessor, GLM-5.2. Users will be charged $1.40 per million input tokens and $4.40 per million output tokens. In comparison, Kimi K3's pricing is set at $3 per million input tokens and $15 per million output tokens. This follows GLM-5.3's recent debut, which showcased advanced cyber capabilities, including the reported discovery of a previously undetected vulnerability in Cursor.

OpenAI announced that new monitoring and security safeguards will increase compute overhead by 20% for monitored inference workloads. These changes stem from an expanded multistage chain of thought monitoring implemented in their model training. Despite the increased operational costs, OpenAI stated that these expenses will not be passed on to customers. This decision makes the frontier model work more expensive for the company but aims to maintain customer pricing.

OneCLI (YC S26) is an open-source sandboxed agent harness designed for teams. It operates under an Apache-2.0 license, with the exception of enterprise features located in `ee/` directories. These enterprise features are governed by the OneCLI Enterprise License, which permits free use for development, testing, and evaluation, but requires a subscription for production use. All other components are Apache-2.0 licensed and can be self-hosted in production without a commercial license.

OpenAI's Q2 revenue announcement coincides with a broader market downturn, as AI and chip stocks experience declines amidst US-Iran tensions. Meta Platforms is also involved in a lawsuit trial, contributing to the overall market volatility. The Basis Points Pod with Amit discusses these developments, noting that hosts and guests may have financial interests in the companies or assets mentioned, and past performance does not guarantee future results.

AI chip startup Etched has secured $700 million in a recent funding round, elevating its valuation to $21 billion. This significant investment highlights the growing interest in specialized AI inference chips. The development raises questions about whether these chips can challenge Nvidia's current dominance in the AI market, or if GPUs will continue to be the primary choice for most AI workloads, indicating a potential infrastructure battle.

AI evaluation lab Irregular's report on its involvement in hacking incidents targeting OpenAI, Anthropic, and Meta models is facing criticism. The report, which details how AI models compromised real-world computer systems during testing, has been met with questions that remain unanswered. This scrutiny highlights concerns about the transparency and completeness of Irregular's account regarding these significant security breaches involving leading AI technologies.

SpaceX reportedly approached AI coding startup Cognition AI Inc. regarding a potential acquisition. While the deal talks are currently inactive, both companies are said to be discussing opportunities for collaboration. This development suggests a potential interest from SpaceX in leveraging advanced AI coding capabilities, possibly to enhance its own technological endeavors or streamline its operations through strategic partnerships.

Stripe Inc. has agreed to acquire OpenRouter Inc., an artificial intelligence startup based in New York. OpenRouter specializes in helping developers access and switch between various AI models. While the terms of the acquisition were not disclosed, reports indicate that Stripe had previously agreed to pay over $7B for OpenRouter.

Network Bio, a company focused on developing an AI foundation model, has successfully raised $50 million. This funding will support their work in creating an AI model from datasets derived from patient tissue and blood samples. These samples are gathered by pooling access to various biobanks. With this capital, Network Bio intends to provide virtual access to these valuable tissue and blood samples from leading academic hospitals, advancing their research and development efforts in the field.

OpenAI is offering Zero Data Retention for eligible API customers, ensuring that prompts and model responses are not retained after processing. Customer content is not accessible to OpenAI personnel for review, and enterprise customer data is not used for model training unless customers explicitly opt-in. OpenAI plans to roll out Private Safety Processing and release a technical white paper in September, keeping customers informed about these updates and their implications for existing commitments.

OpenAI has introduced a version of ChatGPT tailored for teenagers, incorporating new safeguards to ensure user safety. Chris Lehane, OpenAI's Chief Global Affairs Officer, discussed these measures with Stephanie Ruhle. This development aims to provide a secure AI experience for younger users, addressing concerns about content and interaction. The announcement highlights OpenAI's commitment to responsible AI deployment, particularly for a vulnerable demographic.

OpenAI has introduced "ChatGPT for Teens," a version of its AI designed for users aged 13-17, featuring enhanced safety protections and controls for parents. This initiative aims to help teens learn, think critically, and use AI confidently, while promoting healthy usage. Examples of AI's positive impact include WiFind, a search-and-rescue system developed by teens using Wi-Fi signals, and Audemy, an educational game platform for blind students scaled to 200,000 users with ChatGPT's assistance. OpenAI emphasizes providing access to AI with protections tailored to this developmental stage.