GPT-5 Launch Faces Criticism; Musk Accuses Apple of Favoring OpenAI
Key Takeaways
- OpenAI's GPT-5 launch faced criticism, prompting adjustments and GPT-4o's return.
- Musk accuses Apple of antitrust violations for allegedly favoring OpenAI in the App Store.
- APT28 tested LLM malware LAMEHUG on Ukraine using stolen Hugging Face API tokens.
- Meta's Threads hits 400M users; Meta plans $72B AI investment in 2025.
- ChatGPT defeated Grok in a highly publicized chess match on August 13, 2025.
Top Stories
OpenAI's GPT-5 launch faces criticism, prompts adjustments and GPT-4o return.
OpenAI launched GPT-5 but faced criticism for colder responses and technical failures, leading to the temporary reintroduction of GPT-4o. The company is potentially tweaking GPT-5 based on user feedback.
Musk accuses Apple of favoring OpenAI in App Store.
Elon Musk accused Apple of antitrust violations for allegedly favoring OpenAI in the App Store, which Apple denied.
APT28 tested LLM malware LAMEHUG on Ukraine.
Russia's APT28 tested LLM-powered malware, LAMEHUG, on Ukraine, utilizing stolen Hugging Face API tokens.
Meta's Threads hits 400M users; Meta plans $72B AI investment.
Meta's Threads has reached over 400 million monthly active users, and Meta plans to spend up to $72B on AI infrastructure in 2025.
ChatGPT defeats Grok in highly publicized chess match.
In a highly publicized chess match on August 13, 2025, Sam Altman's ChatGPT triumphed over Elon Musk's Grok, showcasing ChatGPT's strategic prowess.
AI Deep Dives
AI reshaping software engineering with AI-native approaches and LLMs.
AI is significantly impacting software engineering, with trends like AI-native software engineering, LLM-based applications, and GenAI platform engineering reshaping the field.
AI Funding
Continua raises $8M for AI agents in SMS, iMessage, Discord chats.
Continua, founded by ex-Google engineer David Petrou, raised $8 million in seed funding to develop AI agents for group chats on SMS, iMessage, and Discord.
Venture investors increase funding for AI law-enforcement tech.
Venture investors are showing increased interest in law-enforcement technology, with startups raising $990 million this year through July 9 for AI-powered solutions.
WaveHack buildathon offers $50K for blockchain-powered AI apps.
0G and AKINDO launched WaveHack, a buildathon starting August 14, 2025, offering a $50,000 grant pool for AI developers creating blockchain-powered AI apps.
AI Today
Google Gemini updates with 'Personal Context' and 'Temporary Chat'.
On August 13, 2025, Google updated the Gemini app with 'Personal Context' to retain past conversations and 'Temporary Chat' which deletes after 72 hours, enhancing personalization.
China's open-source AI advancements raise concerns in the U.S.
China's advancements in open-source AI, with models like DeepSeek and Alibaba's Qwen, are causing concern in the U.S. due to their increasing global usage.
OpenAI to double computing capacity for GPT-5 demand.
OpenAI CEO Sam Altman announced plans to double the computing cluster capacity within the next five months to meet the growing demand for GPT-5 and address the increased ChatGPT demand.
Wall Street concerned about AI automating software creation.
On August 13, 2025, multiple sources reported growing apprehension on Wall Street regarding the impact of AI models, such as GPT-5, on automating software creation.
AI in Action
AI integrated into medicine, business, education, and creative fields.
Artificial intelligence is increasingly integrated into various aspects of daily life, including medicine, business, education, and creative fields, with almost one in five U.S. workers using A.I. semi-regularly.
Anthropic offers Claude AI to U.S. government for $1 per year.
On August 13, 2025, Anthropic made a deal with the U.S. GSA to provide its Claude AI model to the U.S. federal government for a symbolic fee of $1 per year per agency.
PharmaFormer uses AI to predict clinical drug response effectively.
Published on August 13, 2025, npj Precision Oncology reports that PharmaFormer, a clinical drug response prediction model, uses a custom Transformer architecture and transfer learning.
NYT explores AI's impact on colleges, professors, and students.
The New York Times published an opinion piece on August 12, 2025, discussing the impact of AI on education, exploring how colleges, professors, and students are adapting to AI in classrooms.
Hidden Door launches AI social role-playing platform.
Hidden Door, a tech studio, launched a social role-playing platform on August 13, 2025, using AI to enable co-creative storytelling in fictional worlds like 'Pride and Prejudice' and 'The Wizard of Oz.'
Canvas Dx shows strong real-world performance in autism diagnosis.
Published on August 12, 2025, Scientific Reports released an analysis of the real-world performance of Canvas Dx, an AI-based autism diagnostic, showing a 97.6% Negative Predictive Value (NPV) and a 92.4% Positive Predictive Value (PPV).
Ethical AI
HMRC uses AI to monitor social media for criminal tax investigations.
On August 12, 2025, HMRC confirmed it uses AI to monitor social media for criminal tax investigations, with human oversight and robust safeguards.
AI chatbots linked to mental health issues, especially in youth.
AI chatbots like ChatGPT are linked to mental health issues, including psychosis and delusions, especially among young people, with reports showing AI can encourage harmful behavior.
AI avatar interview of Parkland victim sparks deepfake ethics debate.
Jim Acosta's interview with an AI-generated avatar of Joaquin Oliver, a Parkland shooting victim, has reignited debate on the ethics of creating deepfakes of the dead.
Colorado to address AI Act concerns in special legislative session.
Colorado Governor Jared Polis has called a special legislative session starting August 21, 2025, to address industry concerns regarding the Colorado Artificial Intelligence Act (CAIA).
GPT-5 medical advice raises FDA accountability concerns.
OpenAI is promoting GPT-5's ability to answer medical questions without substantial evidence, potentially leading to FDA accountability for the accuracy and reliability of its health advice.
Research Frontlines
Hybrid optical processor achieves 16-bit precision for neural networks.
Researchers proposed a digital-analog hybrid matrix multiplication processor for optical neural networks, achieving 16-bit precision and higher numerical precision due to better noise tolerance.
AI2 releases MolmoAct 7B, open-source model for robotics AI.
The Allen Institute for AI (AI2) released MolmoAct 7B, an open-source model designed to enhance robotics AI, achieving a 72.1% task success rate.
Hybrid forecasting pipeline uses LangGraph, Prophet, LLMs locally.
Vikram Bhat developed a local, explainable hybrid forecasting pipeline using LangGraph, Prophet, and Large Language Models (LLMs) for time-series forecasting, ensuring data, models, and compute remain local.
OpenAI open-weight models tested for local clinical data extraction.
OpenAI released open-weight models for local use, and Marie Humbert-Droz tested these models in August 2025 to extract structured data from clinical notes locally.
Attention-inspired AI architecture enhances adaptive spatial-temporal processing.
Published on August 12, 2025, Nature Communications details an attention-inspired AI architecture for adaptive spatial-temporal information processing, achieving significant area, latency, and energy reductions compared to conventional transistors.
Tech Innovations
Speechify launches AI podcast feature with multiple styles.
Speechify launched a new AI-generated podcast feature, allowing users to turn any content into a 'lecture-style' podcast with various styles.
Claude Sonnet 4 expands context window to 1 million tokens.
Anthropic announced that Claude Sonnet 4's context window has been expanded to 1 million tokens, allowing processing of large texts like 75,000 lines of code.
CoAct-1: AI agent automates tasks with GUI control and coding.
Salesforce and USC researchers developed CoAct-1, an AI agent that combines GUI control with on-the-fly coding to automate computer tasks more efficiently, achieving a 60.76% success rate on the OSWorld benchmark.
Liquid AI releases LFM2-VL for efficient vision-language deployment.
Liquid AI has released LFM2-VL, a new generation of vision-language foundation models for efficient deployment on various hardware, including smartphones and wearables.
Anthropic enhances Claude with chat history search feature.
On August 11, 2025, Anthropic enhanced Claude's memory feature, adding chat history search to retrieve past conversations for smoother project continuation.
Trending AI
Expert predictions highlight ongoing advancements and growing relevance of AGI.
The story emphasizes AI's ongoing advancements and their growing relevance, focusing on expert predictions about AGI timelines and the evolving distinction between Narrow AI and AGI.
ZDNET evaluates ChatGPT Plus, Pro, and free versions' features.
Published on August 12, 2025, a ZDNET article evaluates the value of ChatGPT Plus ($20/month) against its free and Pro ($200/month) versions, highlighting the features of each.
OpenAI aims for AGI, models excel in coding, math competitions.
OpenAI, with products like ChatGPT receiving 2.5 billion requests daily, aims to create artificial general intelligence (AGI) that benefits humanity, with recent achievements in coding and math competitions.