🧠 A Decade of Impact in One Week 🧠
Week of April 21st, 2025
Upcoming Events
🌁 SF Bay Area
Wed, Apr 23rd: Game Night! (Marin County)
Wed, May 28th: SF DEMO NIGHT 🚀
🗓️ Hungry for even more AI events? Check out SF IRL, MLOps SF, or Cerebral Valley’s spreadsheet!
🗽 New York
Wed, Apr 30th: Research Roundtable (w/ Hebbia AI)
🌆 Los Angeles
Tue, May 6th: AI Founders Rooftop Happy Hour (w/ Roam)
🇬🇧 London
Thu, May 8th: 🇬🇧 LONDON DEMO NIGHT 💂🏼♂️
🔥 Austin
Tue, Apr 29th: Cruising through Austin with AI Founders
AI News Roundup
Guest Article by AJ Green
OpenAI launches o3, o4-mini, and a powerful coding agent
The News: OpenAI just dropped its most powerful reasoning models yet — o3 and o4-mini — marking the biggest leap since GPT-3.5. These models were specifically designed to optimize the ChatGPT toolset, and for the first time ever, are capable of advanced multimodal reasoning. OpenAI also introduced Codex CLI, an open-source agent that brings this power to developers' terminals.
The Details:
o3 now holds SOTA performance, scoring over 99% on AIME math benchmarks and achieving more than a 25% success rate on "humanity’s last exam."
o3 redefines benchmarks with SWE-bench Verified jumping from 48.9% to 71.7%, a Codeforces Elo of 2727, and 12x gains in advanced math — thanks to a new architecture leveraging both training-time and test-time compute.
Both models now operate as fully agentic systems - o3 uses tools, executes workflows, and integrates multimodal reasoning. In a live demo, it executed over 600 tool calls in a single response.
o3 is built for precision and deep workflows at $10/$40 per million tokens, while o4-mini offers scalable multimodal reasoning at $1.10/$4.40. Each serves a different purpose in your AI stack.
o3 isn’t just about the benchmarks, it isn’t just about how the entire world is comparing it to AGI. o3’s real winner is the advanced multimodal reasoning capabilities combined with incredibly powerful and intelligent tool use. While it’s not AGI yet, o3 signals what is possible - and with enough infrastructure and scaffolding, we may finally see what Sam Altman was talking about when he said “AGI will be released in 2025”.
Meta's Llama-4 Family is HERE
The News: Meta released the Llama 4 family featuring multimodal support and industry-leading context lengths. Included are new open-weight models Scout and Maverick, plus a preview of the still-training 2T-parameter Behemoth.
The Details:
Scout, with 109B parameters and a 10M-token context window, can run on a single H100 GPU and surpasses Gemma 3 and Mistral 3 in benchmarks.
Maverick, at 400B parameters, supports a 1M-token context window and tops GPT-4o and Gemini 2.0 Flash on key benchmarks while keeping costs low.
Behemoth (2T parameters) is still being trained but is said to outperform GPT-4.5, Claude 3.7, and Gemini 2.0 Pro…although we shall see in due time.
All models use a Mixture-of-Experts (MoE) architecture that activates specific experts per token, slashing compute and inference costs.
Scout and Maverick are available now and integrated across Meta AI experiences on WhatsApp, Messenger, and Instagram.
With LlamaCon approaching at the end of this month, all eyes are on the anticipated launch of Meta’s Behemoth model—but the Scout model may have already shifted the game. With its 10 million token context window, Scout enables AI to ingest and reason over entire codebases, legal archives, or full-length films in a single prompt—changing the way intelligent systems are built. This leap renders many RAG pipelines, API wrappers, and LLM Ops tools increasingly obsolete, as persistent, native memory becomes the new baseline. The memory moat is deepening fast, and with Behemoth on the horizon, we may witness a new open-source champion emerge—especially in the wake of DeepSeek’s disruption last year.
Google Cloud Next 2025 Recap
The News: At Cloud Next 2025, Google dropped an insane lineup of AI agent tools— in fact, we could dive into a whole newsletter with the updates from this event alone. Rather than write a novel, we’ve recapped the Top 5 most impactful updates of the event:
Top 5 Updates:
Agent2Agent (A2A) Protocol: An open-source standard enabling AI agents to communicate across platforms using JSON-based “Agent Cards”. It supports secure, encrypted multimodal workflows (text, voice, video) and integrates with Anthropic’s Model Context Protocol (MCP). Over 50 partners including Salesforce, SAP, and ServiceNow are backing the standard.
AI Agent Development Kit (ADK): A framework for building production-grade agents with under 100 lines of Python. Includes deterministic orchestration, streaming, enterprise connectors (BigQuery, LangChain, CrewAI), and native deployment via Vertex AI or Kubernetes. Evaluation tools are built-in to monitor agent performance.
AI Agent Marketplace: Embedded within Google Cloud Marketplace, it offers pre-built agents from partners like Deloitte and Accenture. Agents are compliance-ready with encryption and secure access controls.
Firebase Studio: Rebranded from Project IDX, it provides a unified full-stack AI development platform. Combines Gemini-powered prototyping, real-time collaboration, and seamless backend orchestration through Firebase. Vibe coding just got a massive upgrade.
Google is laying down the infrastructure for the next era of enterprise software—one built around autonomous agents, not just smarter models. These updates shift the focus from model benchmarks to real-world orchestration, developer tools, and deployment pipelines that make agents viable across organizations. It’s a strategic bet that AI systems will soon need to operate independently across data, apps, and workflows—and Google wants to own that layer. For technical leaders and builders, this signals a new design paradigm: the agent as the core building block, not the endpoint.
BONUS: Stanford's 2025 AI Index Report
The News: The 2025 AI Index Report, released by Stanford's Institute for Human-Centered Artificial Intelligence,offers a 456-page deep dive into research breakthroughs, policy shifts, enterprise adoption, and public perception. It’s an essential compass for anyone navigating the AI ecosystem.
Key Takeaways:
Smaller, Stronger Models: Phi-3-mini (3.8B parameters) matched PaLM's 2022 performance with 142× fewer parameters.
Cost Collapse: GPT-3.5-class inference dropped from $20 to $0.07 per million tokens.
Geopolitical Shifts: The U.S. led model development, China dominated patents and papers, and open-source grew to 66% of all new models.
Enterprise Acceleration: AI adoption jumped from 55% to 78% as companies moved from experimentation to implementation.
This is one of the most respected AI reports of the year—something we personally look forward to, and many in the industry rely on to shape strategy. It’s not opinion, it’s signal. For the builders, strategists, and researchers who want the stats—all the stats—the 2025 AI Index is essential reading. From sweeping shifts in model economics to the changing global AI landscape, it’s a rare source of clarity in a fast-moving space. If you care about where this is all going, this is an essential read.
Events Spotlight
Picture courtesy of our co-founder, Pierce Kelatia.
🌁 SF Bay Area: Pitch, Compete, Win! (with Alumni Ventures)
The GenAI Collective and Alumni Ventures hosted a packed night at the SVB Experience Center, where eight early-stage AI startups pitched to a room of investors, builders, and peers. Through a live app, attendees explored company profiles, submitted questions, and voted for their favorites in real time. The result was an evening that paired polish with honest feedback and surfaced the momentum building across different corners of the AI ecosystem.
SelfActualize.AI was awarded Best Pitch for a clear and confident presentation from CEO Amit Bakshi that stood out in both focus and delivery. Final Round AI also earned top marks, while Truth Systems was selected as the crowd favorite. Each company brought strong signals of early traction, but the night belonged to the teams who could translate technical depth into business clarity. As funding becomes more selective and the bar rises, this night affirmed that the edge belongs to teams that combine sharp execution with genuine market signals.
Picture courtesy of Eris Hanson.
🏛️ DC: AI Insiders Roundtable at Halcyon
The GenAI Collective and Halcyon brought together a diverse group of founders, diplomats, lawyers, creatives, and technologists for a curated evening of small-group conversation and strategic thinking. Held at the historic Halcyon House, the roundtable focused on the pace of AI innovation, the role of ethics and governance, and the need to frame AI as a tool for augmentation—not replacement.
Each table brought its own perspective, from State Department diplomacy to consumer health, but a shared theme ran through: advancing AI requires trust, transparency, and cross-sector alignment. Insights ranged from overlooked open-source practices to critiques of boardroom AI hype cycles. What emerged was a grounded optimism—a collective sense that AI’s promise is real, but the responsibility to build it wisely is just as urgent.
Join the Community!
💬 Slack: GenAI Collective
𝕏 Twitter / X: @GenAICollective
🧑💼 LinkedIn: The GenAI Collective
📸 Instagram: @GenAICollective
We are a volunteer, non-profit organization – all proceeds solely fund future efforts for the benefit of this incredible community!
Join the GenAI Creative Studio!
The GenAI Collective is on the lookout for creative pros who are passionate about AI and storytelling. We need sharp, innovative minds to help shape our brand across social media, newsletters, PR, podcasts, and beyond. If you're ready to craft compelling content at the intersection of tech and creativity, let’s talk.
We are currently looking for:
Creative Director
Graphic Designers
Videographers/Editors
Photographers
Producers
Animators
Marketing Copywriters
While these roles are volunteer-based, the perks are big. You'll get exclusive access to all GenAI Collective events, connect with a cutting-edge AI community, and collaborate with a fast-growing team. Plus, every project you contribute to comes with a shout-out to our highly engaged audience of AI pros and industry leaders—giving your work the visibility it deserves.
If you're a creative professional eager to join a team of AI experts dedicated to building global AI communities and ready to have fun, contact us. This is an amazing opportunity.
Our Premier Partners
Premier Partners are values-aligned leaders who invest in the future of AI by supporting the world’s most vibrant grassroots community. We thank them immensely for their ongoing support! 😄
About AJ Green
AJ Green is a founder, writer, scout, chairman, and respected community leader in the AI and startup space. A former athlete turned tech entrepreneur, AJ is on a mission to make AI the great equalizer—scaling startups, connecting ecosystems, and turning disruption into opportunity. Borderline obsessed with building the infrastructure for the next generation of AI-native companies and the intelligence that power them.
About Noah Frank
Noah is the co-founder of Aurix and has spent his career both working at startups and advising global leaders on innovation strategy. His work and body of research focus on AI policy, anticipatory governance, and effective decision-making. When not working to make emerging tech work for all, you can find him making music with his band. 🎸
About Aqeel Ali
Aqeel co-leads the newsletter for the GenAI Collective. He’s independently researching AI for emotional intelligence and human understanding, an ads industry enthusiast, and veteran startup operations generalist. When not immersed in voice chats with ChatGPT, startup firefighting, or making untimely jokes, Aqeel writes! 🎨