Last Week in AI
#226 - Gemini 3, Claude Opus 4.5, Nano Banana Pro, LeJEPA
Last Week in AI
•November 30, 2025•1h 11m
0
Last Week in AI•Nov 30, 2025
Key Takeaways
- •Gemini 3 achieves record 37.4 on Humanity's Last Exam.
- •Nano Banana Pro adds accurate image editing with AI-generated infographics.
- •Anthropic Opus 4.5 surpasses Gemini 3 on benchmarks, cheaper.
- •OpenAI Codex Max enables 24‑hour continuous coding tasks.
- •NVIDIA reports $500B chip orders, maintains high profit margins.
Pulse Analysis
Google’s Gemini 3 burst onto the scene with a headline‑grabbing 37.4 score on the notoriously tough Humanity’s Last Exam benchmark, signaling a decisive leap over earlier Gemini 2.5 Pro releases. Trained entirely on Google’s in‑house TPUs, the model rolled out seamlessly across Google’s ecosystem, accompanied by the Antigravity IDE aimed at competing with tools like Cursor. The DeepThink variant pushes research‑grade capabilities further, while internal leaks suggest OpenAI feels pressure as Sam Altman’s memo hints at a shifting competitive balance.
On the visual front, Google unveiled Nano Banana Pro, an upgrade to its image‑editing model that now leverages Gemini 3’s reasoning power. The tool can transform raw photos into polished infographics, maintain layout consistency across multiple images, and even solve math problems embedded in pictures. Crucially, it embeds an imperceptible Synth ID watermark, allowing users to verify AI‑generated content—a response to growing concerns over deep‑fake detection. These advances broaden creative workflows, letting professionals generate slide decks, comics, and detailed graphics without traditional design skills.
Anthropic’s Opus 4.5 entered the arena shortly after, outscoring Gemini 3 on several benchmarks, including a 43 on Humanity’s Last Exam, while slashing pricing by roughly one‑third. OpenAI countered with Codex Max, a coding‑focused model promising 24‑hour uninterrupted task execution and superior memory compaction. Both releases highlight a race toward more efficient, higher‑capacity LLMs for enterprise and developer use. Meanwhile, NVIDIA’s earnings beat expectations, reporting $500 billion in chip orders through 2026 and maintaining profit margins near 70%, reinforcing the hardware backbone that fuels these AI breakthroughs. Together, these developments underscore a rapid escalation in model performance, cost competitiveness, and real‑world applicability across coding, visual creation, and enterprise productivity.
Episode Description
Our 226th episode with a summary and discussion of last week's big AI news!
Recorded on 11/24/2025
Hosted by Andrey Kurenkov and co-hosted by Michelle Lee
Feel free to email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
In this episode:
New AI model releases include Google's Gemini 3 Pro, Anthropic's Opus 4.5, and OpenAI's GPT-5.1, each showcasing significant advancements in AI capabilities and applications.
Robotics innovations feature Sunday Robotics' new robot Memo and a $600M funding round for Visual Intelligence, highlighting growth and investment in the robotics sector.
AI safety and policy updates include Europe's proposed changes to GDPR and AI Act regulations, and reports of AI-assisted cyber espionage by a Chinese state-sponsored group.
AI-generated content and legal highlights involve settlements between Warner Music Group and AI music platform UDIO, reflecting evolving dynamics in the field of synthetic media.
Timestamps:
(00:00:10) Intro / Banter
(00:01:32) News Preview
(00:02:10) Response to listener comments
Tools & Apps
(00:02:34) Google launches Gemini 3 with new coding app and record benchmark scores | TechCrunch
(00:05:49) Google launches Nano Banana Pro powered by Gemini 3
(00:10:55) Anthropic releases Opus 4.5 with new Chrome and Excel integrations | TechCrunch
(00:15:34) OpenAI releases GPT-5.1-Codex-Max to handle engineering tasks that span twenty-four hours
(00:18:26) ChatGPT launches group chats globally | TechCrunch
(00:20:33) Grok Claims Elon Musk Is More Athletic Than LeBron James — and the World’s Greatest Lover
Applications & Business
(00:24:03) What AI bubble? Nvidia's strong earnings signal there's more room to grow
(00:26:26) Alphabet stock surges on Gemini 3 AI model optimism
(00:28:09) Sunday Robotics emerges from stealth with launch of ‘Memo’ humanoid house chores robot
(00:32:30) Robotics Startup Physical Intelligence Valued at $5.6 Billion in New Funding - Bloomberg
(00:34:22) Waymo permitted areas expanded by California DMV - CBS Los Angeles - Waymo enters 3 more cities: Minneapolis, New Orleans, and Tampa | TechCrunch
Projects & Open Source
(00:37:00) Meta AI Releases Segment Anything Model 3 (SAM 3) for Promptable Concept Segmentation in Images and Videos - MarkTechPost
(00:40:18) [2511.16624] SAM 3D: 3Dfy Anything in Images
(00:42:51) [2511.13998] LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering
Research & Advancements
(00:45:10) [2511.08544] LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
(00:50:08) [2511.13720] Back to Basics: Let Denoising Generative Models Denoise
Policy & Safety
(00:52:08) Europe is scaling back its landmark privacy and AI laws | The Verge
(00:54:13) From shortcuts to sabotage: natural emergent misalignment from reward hacking
(00:58:24) [2511.15304] Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models
(01:01:43) Disrupting the first reported AI-orchestrated cyber espionage campaign
(01:04:36) OpenAI Locks Down San Francisco Offices Following Alleged Threat From Activist | WIRED
Synthetic Media & Art
(01:07:02) Warner Music Group Settles AI Lawsuit With Udio
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
Show Notes
0
Comments
Want to join the conversation?
Loading comments...