AI News Roundup: January 19-23, 2026
The week of January 19-23 saw AI dominate the World Economic Forum in Davos, where fears about AI-driven job displacement took center stage—with IMF’s Kristalina Georgieva warning that 40% of jobs are touched by AI and describing it as a “tsunami” hitting the labor market. Anthropic CEO Dario Amodei predicted AI will replace all software developers within a year and reach Nobel-level science in two years, while Mira Murati’s $12B startup Thinking Machines suffered catastrophic defections with three founding members returning to OpenAI. Meanwhile, Lightricks launched revolutionary audio-to-video generation with LTX-2, and Trump declared at Davos that energy infrastructure—not compute—will determine who wins the AI race.
🧠 Big Tech News
Lightricks launches audio-to-video generation with LTX-2 and ElevenLabs partnership
On January 19-20, Lightricks launched audio-to-video capability for LTX-2, partnering exclusively with ElevenLabs. The system generates video where audio becomes the control layer—voice, music, and sound effects shape timing, motion, and performance from the first frame. API and open-source access launched January 27.
Mira Murati’s Thinking Machines loses three founding members to OpenAI
In week of January 14-16 (continuing through January 19-23 news cycle), Thinking Machines Lab suffered exodus of co-founders Barret Zoph, Luke Metz, and Sam Schoenholz—all returning to OpenAI. Murati fired Zoph citing “unethical conduct”; OpenAI’s Fidji Simo announced hiring “had been in the works for several weeks.”
OpenAI confirms first consumer device launch H2 2026
At Davos on January 20, OpenAI’s Chief Global Affairs Officer Chris Lehane confirmed the company is “on track” to unveil its first consumer device in second half 2026. Details remain guarded, but reports suggest small, screenless devices—possibly wearable—relying on AI-driven interaction rather than traditional apps.
Microsoft CEO: Energy costs will determine AI race winners
At Davos January 20, Microsoft CEO Satya Nadella declared energy and infrastructure costs—not model quality—will determine AI dominance. “Are you a cheap producer of energy? Can you build data centers? What’s the cost of the silicon?” Nadella framed tokens as “new commodity” for economies.
⚖️ Politics & Legal Affairs
IMF chief warns 40% of jobs “touched by AI” in labor market “tsunami”
At Davos January 20, IMF Managing Director Kristalina Georgieva warned 40% of jobs are either enhanced, scrapped, or significantly changed by AI without implications for better pay. “Even in the best prepared countries, I don’t think we are prepared enough,” she stated.
Trump declares energy infrastructure priority, announces nuclear push
Speaking at Davos January 22, President Trump emphasized U.S. commitment to domestic electricity infrastructure for AI, stating: “We’re opening up” energy plants rather than closing them. Trump touted executive orders targeting 300 gigawatts of new nuclear capacity by 2030.
Yuval Noah Harari warns of AI manipulation potential at Davos
Philosopher Yuval Noah Harari issued stark warning at World Economic Forum about AI’s potential for manipulation, urging humility and robust “correction mechanisms” to prevent systems from exploiting human psychological vulnerabilities.
🔬 Research & Development
Davos AI leaders clash over timeline to human-level intelligence
At World Economic Forum January 22-23, sharp divide emerged: Dario Amodei predicted AI will replace all software developers within one year and reach Nobel-level science in two years, while DeepMind’s Demis Hassabis said current systems are “nowhere near” human-level intelligence and AGI has 50% chance within the decade.
Microsoft Research identifies careers most exposed to AI automation
New study from Microsoft Research detailed in Fortune identified specific career paths most “exposed” to generative AI and autonomous agents. High-skill white-collar roles—particularly finance, legal services, and software engineering—face highest degree of task automation.
🌍 Tools and Launches
Adobe Acrobat “Chat with your PDF” – Acrobat now supports prompt‑based editing and podcast‑style summaries, turning PDFs into interactive, queryable documents.
Microsoft Copilot: Real Talk + Video – Microsoft is rolling out “Real Talk” globally (more human‑like conversation with adjustable depth and style) and testing a “Create a video” feature in the Copilot mobile app for short video generation.
🌍 Startups & Investments
Chinese AI startup ecosystem continues parallel development
While Western labs compete for same talent pool creating instability (Thinking Machines), Chinese startups like DeepSeek and Moonshot AI create competitive models without same talent wars, operating in separate ecosystem.
In India, funding momentum is ticking up again, with 302.8 million dollars raised across 37 deals in the week of January 19–23; the week was led by Emergent.
🌍 AI News in EU & Sweden
Davos discussions focus on European AI investment climate
Multiple Davos panels addressed whether Europe can attract private AI capital given regulatory environment. Microsoft’s Nadella emphasized policies must attract both public and private capital, with governments providing foundational infrastructure.
Jensen Huang: “Robotics is once-in-lifetime opportunity for European countries”
NVIDIA CEO at Davos January 21 called robotics a transformational opportunity for Europe, citing strong workforce in trade skills. Emphasized AI creating manual jobs: “plumbers and electricians…we’re seeing quite significant boom and salaries have gone up.”
🧠 AI in Healthcare & Education
OpenAI preparing healthcare device as part of H2 2026 launch
Chris Lehane’s Davos confirmation of OpenAI consumer device launch includes potential healthcare applications, following January 12 Torch Health acquisition for unified medical memory and ChatGPT Health announcement.
Researchers warn AI cannot replace human expertise in science
University at Albany philosopher Alessandra Buccella argued January 20 that while AI can streamline certain scientific processes, it cannot replace human expertise and collaboration—foundational to every scientific breakthrough.
🤖 Robotics
NVIDIA CEO predicts AI creating manual labor jobs boom
At Davos January 21, Jensen Huang argued AI won’t eliminate jobs but create manual positions. “We’re going to have plumbers and electricians…salaries have gone up. Nearly double. Everybody should be able to make a great living; you don’t need a PhD.”
Boston Dynamics Atlas humanoid integrated with Google AI research
Earlier CES announcements continue development at Davos discussions, with Hyundai-Boston Dynamics partnership using Google’s AI research lab (not OpenAI/Anthropic) for training existing and new Atlas iterations.
🎮 Hardware
Global data center power usage projected to reach 84GW by 2028
Goldman Sachs research presented at Davos shows data center power consumption growing from current 55 gigawatts to 84 gigawatts within two years, driven by AI infrastructure buildout.
Trump administration targeting 300GW new nuclear capacity by 2030
President Trump’s May 2025 executive orders (discussed at Davos January 22) aim to add 300 gigawatts of new nuclear capacity by 2030, increase domestic nuclear fuel production, and speed regulatory approvals for reactor construction.
📊 Market Insights & Investment Trends
WEF releases report on companies successfully scaling AI
World Economic Forum released January 19 report “Proof over Promise” showing divide between companies building capabilities to scale AI versus those struggling. Produced with Accenture, draws on MINDS program showcasing high-impact AI deployments.
🧠 Adoption Trends & Consumer Behavior
McKinsey survey: Two-thirds haven’t scaled AI across enterprise
McKinsey global survey of ~2,000 companies found nearly two-thirds have not yet scaled AI projects across enterprise, despite $1.5 trillion invested in AI last year according to Gartner.
Half of CEOs believe jobs dependent on “getting AI right”
Boston Consulting Group’s January 2026 AI Radar survey found 50% of CEOs believe their jobs depend on successfully implementing AI, representing unprecedented executive-level pressure for AI ROI.
🧠 Research Paper of the Month
“Synchronized Audio-Video Generation via Diffusion Transformers” – LTX-2 Technical Report
Released: January 2026 (Lightricks) | First open-source unified audio-video foundation model
While not a traditional research paper, Lightricks’ release of LTX-2 with complete technical documentation, training code, and architecture details represents a landmark moment in generative AI transparency. The model’s unified approach to audio-video generation challenges the industry’s standard pipeline of generating visuals first and adding sound later.
Architecture innovations:
Asymmetric two-stream transformer: LTX-2 uses separate processing streams for video (14B parameters) and audio (5B parameters) that communicate via cross-attention mechanisms. This architecture generates both modalities simultaneously in a single diffusion pass, ensuring temporal synchronization impossible with sequential generation.
Audio-first control paradigm: Unlike text-to-video systems where audio is an afterthought, LTX-2 can use audio as the primary control signal. Speech cadence determines pacing, musical energy influences motion and camera behavior, and scene changes occur where sound demands them—not where text prompts guess they should.
Multi-stage quality enhancement: Rather than generating 4K video directly (computationally prohibitive), LTX-2 generates base resolution footage then uses spatial and temporal upscalers to reach 4K/50fps. This staged approach maintains generation stability while achieving high-end output quality.
LoRA fine-tuning at scale: The model supports Low-Rank Adaptation in under an hour, enabling users to adapt the 19B-parameter system to specific styles, characters, or visual languages without full retraining. Pre-made LoRAs for camera control and detail enhancement democratize professional-grade customization.
Open-source strategic implications:
Lightricks’ decision to release complete training code—not just weights—marks rare transparency in generative AI. This enables:
Audit and verification: Researchers can inspect architecture choices, understand failure modes, and verify claims about synchronization quality.
Custom deployment: Organizations with strict data privacy requirements can self-host without cloud dependency or vendor lock-in.
Community innovation: Developers can create specialized variants for anime, scientific visualization, or domain-specific content without starting from scratch.
Competitive pressure: By proving open-source can achieve 4K synchronized audio-video, Lightricks challenges closed models from Runway, Pika, and others to justify their proprietary approaches.
Real-world performance:
Early adopters report LTX-2 excels at dialogue-driven content where lip sync matters (AI influencers, educational videos, podcast visualizations) but struggles with complex multi-character interactions and very long sequences. The model’s 10-20 second clip limit requires modular assembly for full videos, though this matches professional editing workflows.
NVIDIA’s optimization work (3x faster inference with 60% less VRAM) demonstrates how open models enable ecosystem collaboration impossible with closed systems—GPU manufacturers can optimize specifically for community-important models.
Industry impact for January 19-23:
The timing of LTX-2’s audio-to-video launch (January 19-20) during Davos week, where CEOs demanded AI ROI, was fortuitous. The model provides concrete example of how open-source can deliver production-grade capabilities at fraction of cost of closed alternatives—directly addressing the “show me the money” theme dominating Davos discussions.
As Anthropic’s Dario Amodei predicted software developer replacement within a year, LTX-2 demonstrates how creative tools—not just coding assistants—will transform professional workflows. The ability to generate broadcast-quality video from podcast audio represents exactly the kind of labor productivity unlock economists at Davos debated.
🧠 Tools to Try
Scroll AI
Scroll.ai – Turns your existing knowledge base (docs, Notion, Confluence, etc.) into domain‑specific “AI experts” that handle RFPs, compliance questionnaires, and sales enablement with higher factual accuracy than a generic chatbot. Good fit for turning your own AI courses or valuation know‑how into internal experts.
Auth0
Auth0 for AI Agents – Identity and access platform tuned specifically for AI agents: token vaults, granular permissions for RAG, human‑in‑the‑loop approvals, and native hooks into LangChain, LlamaIndex, Cloudflare Workers/Agents. Useful if you’re shipping agents that touch customer data
Claude for Chrome
Claude in Chrome – An agentic browser extension bringing Claude directly into Chrome for in‑page summarization, code and content help, and lightweight research flows. A good everyday assistant for your research‑heavy newsletter workflow
.
Discover more from The Tech Society
Subscribe to get the latest posts sent to your email.