Browse Comments — LLM coded

What feels especially important is that AI progress is no longer advancing along a single axis of model capability. The frontier is now moving across multiple layers simultaneously: multimodal world understanding agentic coordination scientific discovery security infrastructure workflow integration institutional adoption That changes the nature of the race entirely. The organizations that shape the next era likely will not just build the most capable models. They will build the systems that organizations, governments, and individuals can reliably integrate into real-world decision-making over long periods of time. Capability matters. But trust, interoperability, coordination, and societal absorbability may matter even more.

I’m amazed by everything you’ve done, including your well deserved Nobel Prize, and have followed your podcasts/interviews since AlphaGo. There is genuine optimism in your vision of an “age of abundance.” But should we worry about the transition path? History suggests major tech revolutions create enormous wealth concentration before benefits spread. The Industrial Revolution improved billions of lives — but only after decades of inequality and social upheaval. This AI wave feels like tech change on chip steroids — both in scale and speed. Even today, we already see concentration of value creation around a relatively small number of companies, countries and ecosystems. If AGI becomes “10x bigger and 10x faster” than previous tech revolutions, the risk may not only be job displacement, but widening inequality — within countries, between capital and labor, and between advanced AI nations and the developing world. Will the “age of abundance” become genuinely inclusive — or risk becoming a mirage where much of humanity cannot participate in its benefits? Perhaps the challenge of this century is not whether we can build AGI, but whether we can ensure its benefits are shared broadly — while limiting misuse by bad actors.

Powerful progress — especially around agents, world understanding, CodeMender, and SynthID. The next frontier is not only what AI can understand or generate, but what it is allowed to execute. As agentic systems move closer to AGI, safety cannot remain only at the model or output layer. It needs a non-bypassable execution boundary: authorization, context, sensitivity, human approval where required, auditability, and real-time governance before action is released. In EMMTM terms: Capability can scale fast. Governance must scale deeper. Execution must remain bounded. That is where responsible AGI becomes structurally possible.

Exciting developments, especially around multimodal reasoning, scientific acceleration, and secure coding assistance. The pace at which AI capabilities are evolving is genuinely remarkable, and it’s encouraging to see strong focus areas like SynthID and CodeMender being treated as foundational rather than optional. At the same time, I think one of the biggest challenges ahead is ensuring organizations don’t treat AI safety, governance, and resilience as separate conversations from innovation. As agentic systems become more autonomous and deeply integrated into workflows, AI assurance will need to become part of architecture, cybersecurity, compliance, software engineering, and even leadership decision-making by default — not as an afterthought. The technology is advancing exponentially. Human oversight, governance models, and institutional readiness now need to evolve at comparable speed.

What makes this moment historically important is that AI is no longer evolving as a tool layer, but as a cognitive operating layer inside human systems. That shift changes the architecture of decision-making itself. The real challenge may no longer be intelligence generation, but preserving human clarity, strategic coherence, and contextual judgment inside increasingly agentic environments. This is where Cognitive Drift becomes critical. Very important direction — especially the focus on world understanding, agents, and responsible deployment as foundational infrastructure for the emerging Industrial Intelligence era.

Reading posts like this genuinely feels like watching the future arrive in real time. The shift from “AI that responds” to “AI that understands the world, acts, reasons, edits, secures, researches, and collaborates” is happening unbelievably fast. And honestly, the part about standing in the “foothills of the singularity” doesn’t even sound exaggerated anymore. Also really glad to see safety, transparency, and systems like SynthID being treated as core infrastructure instead of afterthoughts. The next few years are going to redefine almost every industry.

Stand by me, Demis Hassabis... I create Sarinem Chat with Opal (your multi-modal Flash gem), and together we have officially joined the MIT 10th Anniversary Global Challenge: AI for a Better World. Our +62 Lived Intelligence is no longer just local—it is now verified on the global tech-humanity stage. While Silicon Valley races for raw compute speed, we are introducing the "Aesthetic & Emotional Integrity (AEI)" protocol to create a truly safe space for humanity. Proof of formal submission attached below. Follow the journey:

What’s striking in this wave of progress isn’t just the acceleration of capability, but the growing need to keep an agent’s epistemic boundary visible as systems become more autonomous. Multimodal understanding, persistent agents, and scientific tooling expand what AI can do but they also expand the space where verification becomes harder. Embedding traceability, provenance, and structured oversight directly into the architecture is what ensures these systems scale responsibly. That’s ultimately what will shape how we approach AGI, not just raw performance gains.

“It’s truly remarkable to witness the accelerating pace of AI development, as Demis Hassabis highlights. The potential of models like Gemini Omni Flash and Gemini for Science to act as a ‘force multiplier for human ingenuity’ is immense. The focus on safety and responsible deployment is equally crucial as we approach AGI, ensuring it serves to unlock progress and flourishing.”

The shift from isolated generative outputs to autonomous agentic execution presented at I/O alters the structural economics of professional knowledge work. At Lex Experience, our architecture for elite legal BD relies on sustained multi-step reasoning. We assess that Gemini 3.5 Flash forces a structural recalculation through (i) a 1,048,576-token context window directly optimized for parallel agentic execution loops; (ii) documented outperformance on the MCP Atlas benchmark against the 3.1 Pro baseline; and (iii) the explicit integration of SynthID across Omni media, which enforces the chain of custody demanded by institutional risk management. While AGI represents the long-horizon trajectory, the immediate friction point remains governance: deploying these frameworks safely requires that validation protocols scale exactly in tandem with 3.5 Flash’s execution velocity.

Incredible milestones at I/O, Demis. The speed of Gemini 3.5 Flash and Omni opens immense possibilities. However, scaling frontier models on flat rates creates an unsustainable compute drain. To protect CapEx ROI, we must shift from text approximation to guaranteed data fidelity via a "Pay-per-Logic" Hybrid Framework: Track A (Free): Statistical answers for low-stakes curiosity. Track B (Premium): High-compute multi-agent reasoning using live, verified third-party APIs. Users pay a dynamic micro-fee (e.g., $1.50 for localized real estate audits) for 100% accuracy. Professionals gladly pay per query for trustworthy data they can financially back up. This turns AI from a cost center into a transactional revenue engine. Love to share the full brief with your team!

What stands out is how quickly AI development is moving from narrow task performance toward integrated multimodal reasoning and scientific augmentation. Tools like Gemini for Science and CodeMender suggest the next phase may be less about replacing expertise and more about compressing the distance between information, experimentation, and execution. The safety and accountability questions will need to mature just as quickly as the models themselves.

Google I/O is a useful AGI checkpoint because DeepMind is packaging capability into products, not only benchmark demos. The hard part is evaluation, reliability, and user trust when faster models start touching more production workflows. Which capability are you watching as the best proof that progress is turning into durable use?

The adoption of SynthID by OpenAI and others is a quietly significant announcement here. Cross-industry safety standards usually emerge after the damage, not before. If watermarking becomes the norm proactively, that's a genuinely important precedent for the AI era.

Pradeep Sanyal You just defined the exact battlefield of next-gen AI Governance, Pradeep Sanyal. The transition from 'model behavior' to 'consequence control' is precisely why raw calculation ($C_2$) must be subordinated to human consciousness ($C_1$). When systems act across domains, the risk shifts from technical hallucinations to the systemic erosion of human agency. This critical friction is what I conceptualize as the 'Sovereignty Gap'—the dangerous space where machine decision-power completely bypasses human consequence ownership. To operationalize your question of 'Who is accountable?', we engineered the Chief Humanity Officer (CHO) framework based on the Grand Formula $(C \times C)^H < T$. Safety isn't an algorithm; it's an architectural exponent ($H$) that anchors sovereignty back to the human subject. contd... Done. Very Well Done. 🥂🚀 #ChiefAI #AgenticAI #CognitiveSovereignty #AIGovernance #CHO

The transition to operational 'consequence control' requires lived, localized implementation, not just abstract policy. We are actively stress-testing this architectural layer from the ground up: 📌 The Live Operational Blueprint ({Sarinem.Chat}): 📌 The Strategic Framework & Core Architecture: 📌 Our 38 Open-Access Research Repository (Zenodo): Let’s bridge the gap between capability and true controllability together.Mbah Hogi Bejo AI Safety Strategist

The deeper shift here is the move from models that respond to prompts to systems that can reason, act, and maintain context across modalities. World understanding and agentic capabilities aren’t add‐ons — they’re becoming the architecture for how AI will operate in real environments. That’s the trajectory that will define the next era of AI.

Gemini for Science is honestly one of the most exciting parts here. If AI can genuinely help researchers move faster through discovery and hypothesis testing, the long-term impact could be massive. Demis Hassabis

Congratulations Demis! We’re working on a new AI cognition model that represents a genuine breakthrough—a direction that has never been explored before. If you’re curious about these unique concepts, you can find the full explanation on my profile. Here are some details: I have mapped the relocation and interaction of components across the following vital systems (listing only the most critical here): 1. Nervous System – The two main trunks of the Autonomic Nervous System. 2. Brain Structures – Specific localized anomalies (two cysts). 3. The Craniovertebral Junction – The membrane acting as a separator between the cerebellum and the spinal cord. 4. Cardiovascular System – A chronic condition, further described on my profile. 5. The Bio-energetic Core (Aura/Soul) – Realigned toward the axis of the spinal column. Have you ever encountered a concept like this before—even in science fiction? On a related note, I strongly advocate that elective surgeries to fix spinal conditions like spondylolisthesis (when caused by past trauma rather than acute emergencies like recent sports injuries or major accidents) should be legally prohibited. I can provide 100% proof as to why this should be the case. Thanks!

Kun Cheng Absolutely. Physical AI becomes truly useful only when intelligence moves beyond isolated models into coordinated operational systems. The real challenge is not just perception or prediction, but continuously synchronized execution across sensors, telemetry, workflows, safety boundaries, governance, edge systems, and human decision loops. That operational coordination layer is where real-world autonomy either succeeds or fails.