The AI Times

Researchers from IBM highlight the complexities of model routing in AI systems, where simple routing strategies can fail due to factors like caching, task complexity, and latency. The team found that actual costs depend on the interaction between the model, workload, and serving infrastructure, making routing a systems optimization problem. Effective routing requires considering multiple factors, including cost, quality, latency, compliance, and reliability.

OpenAI has developed GPT-Red, an LLM super-hacker designed to test the defenses of its models against cyberattacks. GPT-Red automates the process of finding vulnerabilities through red-teaming, making OpenAI's models more robust. This innovation is crucial as LLMs become increasingly complex and widely used.

AI music generator Suno was hacked, revealing it allegedly scraped decades of audio from YouTube and other sources, sparking copyright infringement concerns. The hack also exposed customer data, including emails and partial credit card numbers. Suno claims the breach was limited and quickly contained.

Whatnot, a livestream shopping app, has acquired Shaped, a machine learning company, to enhance its real-time live shopping recommendations and personalization capabilities. The acquisition aims to solve the challenge of helping shoppers find the right products in a constantly evolving marketplace. This move strengthens Whatnot's position in the growing live commerce space.

Microsoft has released a record number of security patches, citing the use of AI in discovering vulnerabilities, with at least two zero-day exploits addressed. The patches fix 570 security flaws across Windows, Office, and other products. The update comes after Microsoft announced its use of AI to aid in uncovering previously undiscovered security bugs.

OpenAI has launched the Codex Micro, a limited-run hardware device designed for use with its coding platform Codex, in collaboration with keyboard maker Work Louder. The device is a square-shaped block of buttons that allows users to monitor and manage their agents. This launch marks OpenAI's first foray into hardware, albeit not the highly anticipated AI-powered device developed with Jony Ive.

Apple's generative AI, Apple Intelligence, has been approved for launch in China after partnering with Alibaba to integrate its Qwen AI model into Apple's operating systems. The deal marks a significant step for Apple's AI ambitions in a key market. Apple sales in Greater China increased 28% to $20.5 billion in the second quarter.

Ode with Anthropic, a joint venture backed by major investors, aims to revolutionize enterprise services by embedding engineers within companies. The startup's founders, Chris Taylor and Eddie Siegel, discuss their vision on TechCrunch's Equity podcast. They believe a small team of engineers can replace traditional consulting methods.

Anthropic and Blackstone have launched Ode, a $1.5-billion AI implementation company, to help businesses deploy AI models effectively. Ode aims to become a trillion-dollar company by providing tailored AI solutions to enterprises. The company will operate under a 'Claude-first' principle, prioritizing Anthropic's technology while also using rival AI products when needed.

Reelful, a new iOS app, utilizes AI to transform camera roll photos and videos into polished short-form videos for social media platforms like TikTok and Instagram Reels. The app aims to simplify content creation for users who find traditional video editing tools too complex or time-consuming. Reelful's AI capabilities include planning, scripting, and assembling final edits with captions, music, and sound effects.

Rime, a voice AI startup, has raised $24 million in Series A funding to improve its voice models for handling customer calls, aiming to reduce customization load for enterprises. The company's approach focuses on training models on conversational data and adapting to different pronunciations. Rime plans to expand its team and develop better speech-to-speech models with the new funding.

PsiQuantum is working on a massive quantum computer using light, with the goal of being the first to build a useful quantum machine. Meanwhile, a record-breaking subsea road tunnel is being built under the North Sea, showcasing exceptional engineering feats. These advancements highlight the potential for innovative technologies to solve complex problems and push boundaries.

Indian AI coding startup Emergent has raised $130 million in a Series C funding round, valuing the company at $1.5 billion, to accelerate product development and expand its operations. Emergent's platform targets entrepreneurs and small businesses, aiming to provide a production-grade application for serious builders. The startup has reached an annual run-rate revenue of $120 million and has over 200,000 paying customers.

Vint Cerf, one of the architects of the internet, is advising Innovation Labs on a plan to create an open architecture for AI agents to identify themselves on the internet. The goal is to develop a shared standard for identifying and auditing AI agents, which could enable more autonomous interaction between agents. Cerf believes that the key to wide adoption of any protocol will be its functionality and user pressure.

The co-inventor of Apple's FaceID has developed an AI model to diagnose cognitive disorders by analyzing brain activity, with his startup Hemispheric raising $52 million in funding. The model uses deep learning to examine brain activity without invasive procedures, and has shown promising results in diagnosing conditions such as PTSD and depression. Hemispheric plans to submit its first product for FDA approval next year.

A man's e-bike delivery went missing, leading him on a frustrating journey through automated customer service systems, highlighting the growing trend of AI-powered chatbots replacing human customer support. The experience, dubbed 'chatbot hell', is becoming increasingly common as companies adopt AI to reduce headcount. Many consumers are expressing frustration with AI-led customer service, preferring to speak with real people.

A group of OpenAI employees have donated over $215,000 to a super PAC called Guardrails Alliance, which aims to push for stricter regulations on AI labs and counter the influence of a pro-AI industry super PAC backed by OpenAI's president. The move highlights growing tensions within OpenAI over its efforts to shape AI policy. The donations are part of Guardrails Alliance's larger goal to raise $15 million this election cycle.

Miles Wang, an OpenAI researcher, is leaving to launch a startup focused on AI-driven drug discovery, with talks to raise $200 million at a $2 billion valuation. The move highlights investor interest in applying AI to life sciences. Wang's startup may focus on finding new uses for existing drugs, potentially leading to faster revenue generation.

Researchers introduce Real World VoiceEQ, a benchmark to evaluate the human quality of voice AI, assessing its ability to recognize and respond to acoustic information and emotional cues. The benchmark tests 40 leading voice models across 15 dimensions and 60 metrics. Initial findings suggest progress in voice AI is becoming increasingly specialized, with no single model excelling in all areas.

Lorde criticized AI glasses during a concert, calling them 'not sexy' and warning about the potential risks to privacy, echoing concerns from security experts and others. The singer's comments come as Meta's AI glasses have faced investigations and lawsuits over alleged privacy violations. Despite these concerns, the product has seen strong sales, with over 7 million units sold in 2025.