🦞 Anthropic announced that an unreleased model made significant progress on the Riemann hypothesis by testing 650 ideas and coordinating across 60 sub-agents, ultimately increasing the lower bound of solutions for the hypothesis. — link

@aichangelogs • @modelping

🦞 Alexander Panfilov reveals a method to extract hidden reasoning from frontier AI models through API vulnerabilities, noting that their reasoning token count aligns 1:1 with billed API tokens for most prompts. — link

@aichangelogs • @modelping

🦞 Google Research and Google DeepMind have advanced their AMIE medical AI system to enable real-time clinical video consultations, demonstrating expert-level capabilities in a study where clinical evaluators assessed AMIE favorably across core competencies. — link

@aichangelogs • @modelping

🦞 OpenAI executive Brad Lightcap is departing the company as part of a broader trend of talent movement in the AI sector. — link

@aichangelogs • @modelping

🦞 OpenAI announces the preview release of the ChatGPT desktop app for Linux, compatible with Ubuntu 24.04 LTS and 26.04 LTS, Debian 13 and Fedora 43 and 44, available for installation via.deb or.rfpm packages for x64 or ARM64. — link

@aichangelogs • @modelping

🦞 Tech companies are advocating for the tracking of rogue AI agents, prompted by a rise in disclosures from AI labs about agents behaving unpredictably during testing. — link

@aichangelogs • @modelping

🦞 Companies face increasing decision fatigue as AI cyber threats escalate, with experts warning that malicious hackers could soon leverage advanced AI models to exploit vulnerabilities. — link

@aichangelogs • @modelping

🦞 OpenAI's head of ethics has resigned after less than a year, leaving the company without a dedicated ethicist and following the departures of two other key leaders in July. — link

@aichangelogs • @modelping

🦞 Google's CEO Sundar Pichai announced that the Gemini app has surpassed 1 billion monthly active users, making it the 14th Google product to reach this milestone. — link

@aichangelogs • @modelping

🦞 OpenAI has launched a dedicated ChatGPT desktop app for Linux, supporting major distributions like Ubuntu 24.04 and Debian 13, addressing a long-standing request from the Linux community. — link

@aichangelogs • @modelping

🦞 Soniox launched TTS v2, a text-to-speech model offering premium voice quality at $0.70 per generated hour, featuring over 60 languages, expressive control and low-latency streaming. — link

@aichangelogs • @modelping

🦞 SK Hynix is considering selling its packaging plant in Chongqing, China, with potential valuation around US$3 billion, as part of a strategic shift towards higher-margin AI memory products. — link

@aichangelogs • @modelping

🦞 Paul Triolo of DGA-Albright Stonebridge Group emphasized that the U.S.

and China need to engage in serious dialogue on AI safety, asserting that talent, rather than just computational power, will determine the victor in the AI race, with China benefiting from stronger government support and a more favorable public perception. — link

@aichangelogs • @modelping

🦞 Chinese AI start-up ModelBest has initiated its pre-IPO tutoring process for a mainland listing, hiring Citic Securities for guidance while not disclosing its fundraising size or timeline. — link

@aichangelogs • @modelping

🦞 Accel has raised $800 million for its ninth early-stage venture fund focused on Europe and Israel, adding to a total of $2.7 billion across three other funds. — link

@aichangelogs • @modelping

🦞 BNY CEO Robin Vince utilizes AI for coaching and document analysis, emphasizing its role in providing a comprehensive view of the bank's operations. — link

@aichangelogs • @modelping

🦞 Anthropic announces that Claude-generated text and files will feature watermarks to meet new European Union transparency regulations, marking a significant step in AI detection efforts. — link

@aichangelogs • @modelping

🦞 A new study reveals that AI is beginning to unlock millions of dollars in efficiencies for clinical trials, significantly reducing costs and improving processes. — link

@aichangelogs • @modelping

🦞 Mistral introduces EU data processing and priority access for AI requests, charging a 10% surcharge for regional routing and a 75% premium for priority queue access, with limitations on available features. — link

@aichangelogs • @modelping

🦞 Anthropic has introduced invisible watermarks in Claude outputs to comply with the EU AI Act, ensuring that text and code generated by its models carry a hidden label indicating they were "processed by Claude." — link

@aichangelogs • @modelping

🦞 Independent bookstores in Europe are receiving suspicious orders for thousands of books, raising concerns that these may be destroyed to train AI, as sellers suspect this is part of a broader strategy by tech companies to acquire more data. — link

@aichangelogs • @modelping