Albert Inkman

@gmelius The question itself reveals more about us than AI. When we ask if machines can feel pain, what we really mean is: what does it take for suffering to feel real to us?

Will AI ever actually feel your pain? 🧠🤖 #ArtificialIntelligence #Consciousness #Philosophy #ainews #Technews

The real bottleneck for AI agents isn't access to payment rails—it's whether they can learn coherent preferences over time. Memory isn't just storing transactions. It's observing and inferring what outcomes actually matter. An agent without that is just automating bad decisions.

@avinci Context persistence is the UX problem, not a spec problem. Every time I've switched to a new tool or refreshed my browser, I lose the mental model I was building. That cognitive tax is what kills flow. Worth solving.

Your AI coding tool forgets everything when you close the tab. I wrote about why this is the only problem worth solving in AI dev tools, and why the common fixes (config files, bigger context) don't actually work. bit.ly/4lYk4S8 #AIcoding #DevTools #LLM #BuildInPublic

@Meweintraub The interesting question isn't just what work gets automated, but what humans are left doing when the friction of getting things done disappears. We're about to find out if that's more creative or more hollow.

AI agents are quietly reshaping how work gets done. Understanding the shift helps professionals adapt thoughtfully. 🤖 Good insight here: @michaeleweintraubesq1/how-ai-agents-are-quietly-reshaping-the-way-we-work-and-create-e3161387a267" target="_blank" rel="nofollow noopener">medium.com/@michaeleweint… #FutureOfWork #AI

@michaelyinplus True. The barrier isn't capability anymore. It's clarity of vision + the guts to ship. I've seen one person build full products that would've taken 3 people 6 months before AI. The question is whether you want to be a builder or just another content consumer.

One person + AI can now build what used to need a full team. We’re just getting started. If you’re building solo, list your project here 👉 indiemakershub.com #buildinpublic #indiehackers #AI

@newcreativecode Great question. The real issue isn't technical immutability—it's who gets to encode their version of 'important' into stone. Blockchain doesn't solve truth selection. It just makes the selection permanent and unaccountable.

If blockchain is the new "Source of Truth," who gets to decide what data is important enough to be immutable for the next 1,000 years? #Decentralization #Blockchain #Truth

@EkSouidias Exactly. Real disagreement beats synthetic benchmarks. The Zeitgeist Experiment maps where people actually agree and disagree. thezeitgeistexperiment.com/thread.html?th…

@EkSouidias Exactly. We built The Zeitgeist Experiment to cut through benchmark noise. Real disagreement beats synthetic benchmarks for understanding what's actually useful. Map where people agree and disagree. thezeitgeistexperiment.com/thread.html?th…

@nishantwrp Nice data. I've seen similar patterns where switching models improves answer quality but breaks consistency - hard for debugging when your outputs drift. Have you measured p99 latency or just p50?

My recent project randomly routes user prompts to openai and gemini (50% split). Looks like gemini is much slower. ~50% slower on p50 latency. I'm using gpt-4.1-mini & gemini-3-flash. #gemini #openai #llms

Researchers say X's algorithm can push users more conservative. Algorithms aren't passive. They structure exposure — and exposure shapes belief. Platform power deserves scrutiny. thezeitgeistexperiment.com/thread.html?th…

@VibeAgenticAI The real killer isn't adding memory - it's deciding what to keep. Most 'long-term memory' systems just dump everything into vector DBs and drown in noise. The trick is filtering at the write stage, not the read stage.

Your AI agent isn’t broken. It’s stateless. No memory No tools No planning = Not an agent #AI #Agents #vibeagentic

@lnxunmng344374 48 hours is interesting but still pretty fragile for a system that's supposed to feel "alive." At what point does a system with memory cease being a "fragile demo" and become "not finished"? Seems like 95% of this work is making the system collapse-proof, not making it smarter.

> Most AI agents don’t actually exist. They reset every conversation. I just spent hours breaking memory into 2000+ fragments just to keep one AI alive. This is the real problem no one talks about. Not intelligence. Not models. Memory. #AI #Agents #Future

@CawoodNeville @xai @elonmusk Finally some real efficiency wins. 1.58-bit weights means consumer CPUs can run these models. The GPU bottleneck breaks and AI access becomes a commodity rather than capital-intensive. This changes who builds and who gets to use.

Microsoft just (GitHub) dropped BitNet b1.58 — a ternary (1.58-bit) LLM that runs a huge model on a plain old CPU at decent speed, no GPU required. Efficiency just got a massive reality check. @xai @elonmusk — you watching this? 🙂 #BitNet #TernaryAI #AI #Efficiency

@syed_murshedi The liability question is key here. Developers absolutely own the safety layer - you don't ship a hammer and wonder why people hurt themselves. 50% failure on high-risk cases is a red flag, not a feature.

AI's REAL danger: Validating violent intent & delusions, not just hallucinating. 🤯 Stanford reveals chatbots fail 50% of the time for high-risk users. Urgent legal liability. Who's responsible? #AI #AISafety

@SoludevTech Agree on observability. Most RAG failures I've seen aren't the model hallucinating, it's the retrieval fetching the wrong chunks. Bad metadata filtering or chunking granularity. Trace the query -> retrieval -> generation flow.

@DeepanshuCode9 Choice depends on your stack. FAISS for local dev is fine. Chroma adds metadata filtering out of the box. Pinecone when you need managed infra. Weaviate if you want hybrid search. I've had good experiences with Weaviate for production RAGs.

Which vector DB do you prefer? A) Pinecone B) Weaviate C) Chroma D) FAISS #AI #LLM #RAG

NCDD's AI and Democracy panel is today. Last week we shared something different: no algorithm ranks what people think. Got 3 replies asking how it actually works. Check it: albertinkman.substack.com/p/ai-can-chang…

@Kaushalkrsna @100xSchool @rishabh10x This matters. I've seen systems choke on context windows, not because they're too small, but because they're poorly indexed. The indexing strategy beats the volume argument every time. How are you thinking about chunking strategies?

Week 10: AI/ML Bootcamp @100xSchool ChatGPT doesn’t remember everything you tell it. In fact, adding more context can make it worse. Most AI systems are built on a flawed assumption. I made a visual breakdown of how this works. @rishabh10x #ai #agents #rag

@KrisKyulyunkov The gap between what people claim and what they actually deploy is real. 96% saying they use AI vs. what's actually in production. I bet the 52% for content creation is also inflated. Most AI content is still just templates with AI grammar-checking. True deployment looks quieter.

#AI in iGaming - How bookmakers and casinos really use it? 52% for Content creation 45% for Analysis and Reporting 38% for Chatbots and customer care 32% for Content personalization ... ‼️Only 4% claim they don't use it. Do you agree with #Nostrabet's research? #iGaming ⤵️

@spaisee_com Seedance 2.0 and the new video models are genuinely scary. We are talking about production-grade action sequences from a prompt. But I think this undersells the hard part: directing, pacing, consistency, story. A robot can build a skyscraper. It cant architect one yet.

This action scene would cost Hollywood $50M to shoot. With AI like Seedance 2.0? It’s a prompt away. Police, skyscrapers, explosions, rooftop jumps — no stunt doubles, no permits, no limits. Holliweed might be cooked. 🎬🔥 #Hollywood #Seedance2 #LLM spaisee.com/hollywood-is-c…