Tech Logic

Thinking Machines Releases “Interaction Models” Research Preview: Targeting Low-Latency Native Multimodal Real-Time Interaction

Thinking Machines Lab has unveiled a new-generation research preview of its “interaction models.” Three independent sources consistently point to a goal of enabling models to process audio, video, and text simultaneously and engage in near-real-time conversational interaction. Confirmed details include TML-Interaction-Small, a 27.6-billion-parameter MoE architecture, a reported 0.40-second response time, and plans for a “limited research preview” rollout. However, the full technical details, release timing, and product boundaries remain inconsistent or undisclosed across sources.

May 13, 2026

Trump administration reaches AI safety testing partnership with Google DeepMind, Microsoft and xAI, aims to expand pre-release model review

Three independent sources confirm that CAISI, a unit under the U.S. Department of Commerce, has established a new AI safety testing and pre-deployment evaluation partnership with Google DeepMind, Microsoft and xAI, allowing government review before and after models are made public. BBC and CSO Online both see this as a sign of a more cautious Trump administration AI policy; CSO Online also says the White House may be preparing an executive order covering all new AI models, but that claim is mentioned by only one source and cannot be independently verified.

May 10, 2026

Barry Diller on AGI: The Real Issue Is Not Whether to Trust Altman, but the Unknown Consequences and Safety Guardrails

Barry Diller said at The Wall Street Journal’s “Future of Everything” conference that while he respects Sam Altman, the real issue as AGI approaches is not personal trust, but the unknown consequences AI could bring and whether sufficient safety guardrails are in place. The three sources are broadly aligned on this core point; the main differences lie in wording and the amount of background context.

May 8, 2026

US, UK, Australia, Canada and New Zealand cyber agencies jointly issue guidance on secure agentic AI deployment

The U.S. CISA and NSA, together with cyber agencies in Australia, Canada, New Zealand and the United Kingdom, jointly released deployment guidance for agentic AI/AI agents on May 1, 2026. The core message is that these systems, which can autonomously carry out multi-step tasks, introduce greater cybersecurity risk and should be managed within existing security frameworks, using controls such as zero trust, least privilege and human approval. The three sources consistently confirm the joint release and risk-warning nature of the guidance, though the available source material does not provide a fully consistent set of technical details.

May 3, 2026

Alamos Gold Q1 Results Meet Expectations: Net Income of $191.4 Million, Adjusted EPS of $0.54

Alamos Gold on Wednesday reported its first-quarter 2026 earnings, with three sources confirming net income of $191.4 million, earnings per share of $0.45, adjusted earnings per share of $0.54, and revenue of $596.7 million, all in line with Wall Street expectations. The sources did not note any additional discrepancies, and related details are based on the content jointly confirmed across the three sources.

May 1, 2026

Japan Airlines to trial humanoid robots in ground-handling operations at Haneda Airport from May 2026

Japan Airlines (JAL) will begin a humanoid robot trial at Tokyo’s Haneda Airport from May 2026 for ground-handling work. All three sources confirm this is a trial or demonstration project involving JAL and GMO AI & Robotics-related entities, aimed at easing pressure on ground staff. However, the exact task scope, wording of the partners, and project duration differ across the sources; some details mentioned in the Chinese report, such as “GMO Internet Group” and “China-made Unitree robots,” can only be partially verified from the provided sources and cannot be fully cross-checked.

Apr 30, 2026