Post labour economics of white collar aka tier 1 audience will unleash the mental health category like insane. We are headed to unimaginable identity crisis.

Indian manufacturer's data is so difficult to source! They have 25 years of data - all on paper. Still works that way. Pretty much >80% of manufacturers in Gujarat, at least. What is the best way to digitize these? 🫠

Next big thing for frontier model and physical AI will be learning human behaviour at neurological level. User / population simulation cannot truly predict the subjective factors of decision making. The objective lens of simulation may not be completely accurate. Especially true for consumer - no user, for any consumer application- commerce, media, finance, health etc. decides basis data but emotion first. Otherwise, mostly true, for enterprise. Current approach like @simile_ai’s makes sense for certain verticals like healthcare. Forecasting human behaviour cannot be objective.

So how many personal agents will live through the first-scroll on WhatsApp/iMessage effectively? And human chats later Open router/superhuman of personal agents? Are we really going to address it as personal agent? 🙃

Micro-apps say like speaking practice, lookmaxxing or Cal AI type - can be assistant-first WhatsApp/iMessage interface. It spike up engagement and retention by atleast 2x.

It's over for Clippers 🤯 Opus 5.5 - One-shot clip and edit 🔥 youtu.be/a00xIn5kwhM?si=b70T… Asked to clip 1 min from the link and edit it like the reference video provided.

Astra’s 3D rendering ability + opus 5.5’s + H3 max should really introduce a new format of interactive entertainment/media. Will be sad if it doesn’t happen. Short on bandwidth else would have tried it out

Agent harness for filmmaking, that too, vernacular first will be massive. UX cannot be like other video generation platforms - should be loveable like. Subjectivity and treatment of frames can eventually be harnessed with more data.

- Niche-sub-culture, - Long format content - Curated (to extent- personalised) feed/internet - Characters IP: multi format (like minion, hello kitty, Mario) - Interactive media IPs - ofc, analog experiences We will see significant deal flow in entertainment sector in 2027.

If you consciously audit your current or recently acquired beliefs - You’ll realise the line between our consciousness and our algorithm curated feed have gotten blurred. Beliefs beyond feed is rare. The evolution of our beliefs has mostly been induced by alot more amplifier nodes than we can think of. Volume + audio has the highest impact of all. Learning about evolving human beliefs is key for AI safety and RL environments for agents to go beyond human skills, objectively.

instincts is solid to dig up reddit for research

Most of the conversation is around evals, guardrails and monitoring what a model does. Which makes sense. But all of these systems mostly look at the end of the chain. An action happens, and we ask whether it was safe. What feels more interesting to me is everything that happened before the action. Why did someone want to do it? What information shaped that intent? Which beliefs changed along the way? Intent is difficult because it isn't binary. The exact same action can mean completely different things depending on the person, context and beliefs behind it. And increasingly, those beliefs are being shaped online - through machinery most safety conversations ignore. Short-form is now the primary discovery layer for new information. Not search, not editors - a feed. Within that feed, distribution runs on content seeding and organic UGC: one idea gets planted, thousands of creators and accounts carry it forward, and the algorithm amplifies whatever holds attention, not whatever is true. A single influencer is now a distribution node larger than most media institutions. One seeded narrative, repeated across enough surfaces, stops looking like a claim and starts looking like consensus. Information becomes attention. Attention becomes belief. Belief hardens into micro culture - communities with their own language, their own authorities, their own epistemology. This isn't a fringe dynamic. It's the default pipeline through which a generation now forms its worldview. And it's largely inevitable - the economics of attention guarantee it. None of these people were told what to do. They arrived there one belief at a time. That made me wonder whether a missing layer of AI safety is not just understanding what models do, but understanding how human intent forms in the first place. Safety work has mostly focused on what models do. I think the upstream layer - how intent forms in the people models learn from and act for - is underexplored, and increasingly load-bearing. Understanding the action won't be enough. We need a model that understands the belief system that produces it and its evolution.

Learning about human intent and belief is the way to ultimate AI safety. Evals and guardrails alone won't fill the void for what triggers action based on belief- to take the specific action and march forward. Action and intent are so closely intertwined and too contextually subjective to prevent the currect methods. Human belief now forms on the internet. If you can't model how belief becomes intent, you can't see the action coming.

It will be interesting to see how instincts or personal agents at scale monetise besides data play.

The only thing that can holdback from scaling any consumer ai are credits and working capital.

So, consciousness is the next frontier to conquer? I don't think we are too far away