A Major Week for AI: Google’s Gains, Rocket League Excitement, and a Celebration at Sea

Reinforcement learning (RL) is emerging as a dominant theme, Google is making significant strides, and the party atmosphere has reached new heights. These were the key takeaways from this year’s NeurIPS held in San Diego.

NeurIPS, short for the “Conference on Neural Information Processing Systems,” began in 1987 as a strictly academic gathering. Over the years, it has transformed into a major industry event fueled by the AI hype, where research labs seek talent and investors scout for the next big AI startups.

Although I couldn’t attend NeurIPS this year, I was eager to learn about the discussions happening in San Diego. I reached out to engineers, researchers, and founders for their insights. The responses below include thoughts from notable figures like Andy Konwinski, cofounder of Databricks; Thomas Wolf, cofounder of Hugging Face; and representatives from various leading tech companies.

I posed three questions to everyone: What was the hottest topic at the conference? Which labs are thriving or struggling? And which party was the best?

The consensus was unmistakable. “RL RL RL RL is taking over the world,” said Anastasios Angelopoulos, CEO of LMArena. The industry is rallying around the notion that fine-tuning models for specific applications, rather than merely scaling pre-training data, will propel the next wave of AI advancements. When it comes to lab momentum, Google is clearly in the spotlight. “Google DeepMind is feeling good,” noted Wolf.

The party scene was equally intense. Konwinski’s Laude Lounge became a key gathering spot, attracting top researchers like Jeff Dean and Yoshua Bengio. The Model Ship cruise, an exclusive event for 200 researchers, showcased an unprecedented commitment to dancing, according to organizer Nathan Lambert. Roon had a more subdued take on the festivities: “You can learn more from Twitter than from literally being there… mostly my on-the-ground feeling was ‘this is too much.’”

Here’s what attendees shared about NeurIPS this year:

**What was the buzziest topic among attendees that you think more people will be talking about in 2026?**
– Andy Konwinski: “I heard agentic AI, RL, and world models as overhyped, but also as areas that are up-and-coming and interesting to watch.”
– Thomas Wolf: “AI x science, interpretability, RL long rollouts.”
– Roon: “The buzz is about continual learning… my on-the-ground feeling was ‘this is too much.’”
– Maya Bechler-Speicher: “Tabular foundation models are gaining traction, and I expect this momentum to continue into 2026. Physical AI is also drawing attention with many open research questions.”
– Anonymous researcher: “AI for the physical world (robotics, engineering, etc.) looks like it’s finally taking off.”
– Nathan Lambert: “It was accepted that we’re in ‘The Age of Research’ rather than just scaling. No single area stood out as the most important this year.”
– Brian Wilt: “The buzz was about the research happening in frontier labs versus academia, with a sense that we need another breakthrough beyond scaling current architectures.”
– Evgenii Nikishin: “Continual learning was certainly among the buzziest topics.”
– Paige Bailey: “Sovereign open models, especially deploying them on-prem with fine-tuning + RL, will be big in 2026.”
– Sachin Dharashivkar: “Designing RL environments and training agents was the most discussed topic.”
– Ronak Malde: “Continual learning will require new architectures and data scalability models.”
– Deniz Birlikci: “More teams are focusing on creating a dense taxonomy and labeling for their data, especially in RL.”
– Richard Suwandi: “Discussions around building truly creative AI systems are expected to become a major research frontier in 2026.”
– Anastasios Angelopoulos: “RL RL RL RL is taking over the world.”

**Which labs feel like they’re surging in momentum, and which ones feel more shaky?**
– Nathan Lambert: “Gemini and Anthropic are on the rise, while OpenAI seems to be struggling. I didn’t hear much about xAI.”
– Evgenii Nikishin: “The Big 3 frontier labs (GDM, Anthro, OAI) are doing well overall, but many LLM/image startups from 2022-2024 are quietly fading away.”
– Andy Konwinski: “Surging momentum: Alibaba/Qwen, Moonshot/Kimi, Arcee, Reflection AI, and Google with Gemini 3.”
– Anonymous researcher: “Reflection had a massive booth for a young startup.”
– Brian Wilt: “Alphabet/Google had the most accepted papers this year.”
– Paige Bailey: “Periodic Labs and Reflection AI are surging with interesting missions.”
– Ronak Malde: “Several new labs launching in 2026 will shake up research. DeepMind is still excelling.”
– Richard Suwandi: “Google DeepMind is pushing a new research agenda, focusing on continual, long-term memory rather than just bigger transformers.”
– Thomas Wolf: “Google DeepMind is feeling good.”

**What was the best party you attended or had FOMO over?**
– Nathan Lambert: “Model Ship was the quintessential NeurIPS party, featuring 200 top researchers and a commitment to the dance floor that was unprecedented.”
– Andy Konwinski: “I missed events organized by Robert Nishihara and others, which was a bummer.”
– Roon: “I liked the a16z party because I got to meet Lex Fridman, but even the parties I tried to avoid were huge.”
– Maya Bechler-Speicher: “The Meta party was impressive, and G-Research invited a small group to a Michelin-star restaurant, which was exceptional.”
– Brian Wilt: “A small gathering at comma.ai was my favorite event. I use their open-source driver assistant.”
– Evgenii Nikishin: “The OpenAI party was great.”
– Paige Bailey: “I had FOMO over the ML for Systems workshop and the ‘Claude and Gemini Play Pokemon’ workshop.”
– Ronak Malde: “Radical VC bringing Jeff Dean and Geoffrey Hinton together was a highlight.”
– Anastasios Angelopoulos: “Laude Lounge was a standout.”
– Thomas Wolf: “The Hugging Face party had over 2,500 registrations, and I enjoyed the Prime-intellect event.”
– Dylan Patel: “Mine haha.”

Yes, some attendees considered keynotes to be parties, proving that academia still thrives at NeurIPS.