LWiAI Podcast #246 - Gemini 3.5 + Omni, Musk Loses, OpenAI vs Erdős

2026-05-20 · Source: Last Week in AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Intermediate, extended

Summary

Google I/O unveiled Gemini 3.5 Flash, outperforming its predecessor on benchmarks and powering Gemini Spark, an agentic browser running on dedicated cloud VMs. Gemini Omni, a new multimodal model, processes diverse inputs to generate and edit video, with a Flash variant already available. Google also launched Anti-Gravity 2.0 (CLI/SDK) and Gemini for Science, alongside a Genie world model update simulating Street View. Meanwhile, OpenAI lost its lawsuit against Elon Musk on statutory grounds. Anthropic secured a \$900 billion valuation, projecting Q2 profitability, and hired Andrej Karpathy for its pre-training team. OpenAI restructured product teams and faces tensions with Apple. AI chipmaker Cerebras saw a 90% IPO surge. Research highlights include OpenAI solving an 80-year-old Erdos problem, "negation neglect" in models, and the discovery of redundant neural network circuits. Benchmarks like Terminal World show agents' limitations, while autonomous AI cyber capabilities are doubling every 4.7 months. New deepfake legislation and image watermarks address safety concerns.

Key takeaway

For AI Engineers and strategists, the accelerating pace of AI capabilities, particularly in multimodal and agentic systems, demands continuous re-evaluation of model choices and infrastructure investments. Prioritize robust safety evaluations, especially concerning negation neglect and autonomous hacking, while exploring new tools like Gemini Omni and Cursor Composer 2.5 to maintain competitive advantage and ethical deployment. The rapid market shifts underscore the need for agile strategy.

Key insights

AI capabilities are rapidly accelerating across multiple modalities and domains, driving significant market revaluations and strategic shifts among leading labs.

Principles

• AI capabilities, particularly in cyber, are doubling every 4.7 months, indicating rapid acceleration.
• Multimodal data and world models offer significant differential advantages for AI development.
• Supervised fine-tuning can inadvertently reinforce false beliefs due to "negation neglect."

Method

The "All Circuits Lead to Rome" paper details a method for mechanistic interpretability, using continuous learning parameters on graph edges to discover sparse, functional neural network subsets.

In practice

• Utilize Gemini Omni Flash for advanced video generation and editing workflows, leveraging its multimodal capabilities.
• Leverage Cursor Composer 2.5 for cost-effective, high-performance coding assistance, undercutting competitors on price.
• Implement C2PA/Synth ID watermarks for verifying AI-generated images, enhancing content authenticity.

Topics

• Google Gemini
• Multimodal AI
• AI Agents
• AI Safety
• Cybersecurity AI
• AI Benchmarking

Best for: Machine Learning Engineer, NLP Engineer, Computer Vision Engineer, AI Scientist, AI Engineer, Director of AI/ML

Related on AIssential

• Gemini's busy agentic day at Google I/O — Gemini is evolving into a pervasive, agentic, and multimodal AI engine deeply integrated across Google's product ecosystem.
• GOOGLE IO | LET"S GOOOOOOOOOOOOOOOOOO!!! — Google is rapidly advancing AI capabilities across its product ecosystem, focusing on agentic systems, multimodal models, and foundational infrastructure.
• MLWhiz Weekly Recsys/ML/GenAI Newsletter # 8 - The week of Google I/O 2026 — Google's I/O 2026 announcements collectively position agents as the next interaction layer, with Google aiming to own the entry point.
• True Positive Weekly #141 — The AI and tech landscape in 2025 features new models, enterprise adoption trends, and tools for privacy and multimodal processing.
• Last Week in AI #341 - Musk loses to OpenAI, Google's IO updates, OpenAI solves Erdős — AI is rapidly advancing in capabilities and market presence, driving significant legal, ethical, and competitive shifts.
• LWiAI Podcast #234 - Opus 4.6, GPT-5.3-Codex, Seedance 2.0, GLM-5 — The AI sector is rapidly advancing with new models, generative media tools, and significant venture capital investments.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Last Week in AI.