<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Futurist AI on Medium]]></title>
        <description><![CDATA[Stories by Futurist AI on Medium]]></description>
        <link>https://medium.com/@FuturistAI?source=rss-daf9f98f9500------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*DNBNf9gWtL0hVwFztYoIUw@2x.jpeg</url>
            <title>Stories by Futurist AI on Medium</title>
            <link>https://medium.com/@FuturistAI?source=rss-daf9f98f9500------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 22:44:32 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@FuturistAI/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AI Is Not “Only Good for Coding”]]></title>
            <link>https://medium.com/@FuturistAI/ai-is-not-only-good-for-coding-f792ad34b51b?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/f792ad34b51b</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Thu, 12 Feb 2026 16:12:28 GMT</pubDate>
            <atom:updated>2026-02-12T16:12:28.761Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve heard it again today — we’ve all heard it. That AI is only good for coding. That programmers are being fired everywhere while other professions remain untouched. Friends keep asking about my job security. Let me clarify some of these claims.</p><h3>Programmers adopted it first. That’s not surprising.</h3><p>Who else would be the first to adopt a new type of software if not software engineers themselves? When I saw the capabilities of large language models, adapting them for coding was my first idea. I even built my own plugin because the existing tools were limited. But it wasn’t because coding was easy for AI. It was because it was challenging, and because I had the expertise to judge the results. Had I been a lawyer, I probably would have tried legal work — but I wouldn’t have had the skill to build the tool myself.</p><h3>Coding is not so easy for computers.</h3><p>I saw a comment on X: <em>“No surprise. It’s just translation from English to the language of computers”</em>. Well, not really. Twelve years ago Google Translate could already do useful translations between languages. If coding were just translation, we’d have some some sort of AI programmers back then. We didn’t.</p><p>In the world of professional software development, a single command like “create a social app for XYZ” should produce hundreds of files in multiple directories, each with hundreds of lines of code, all related in intricate ways. It’s more like asking someone to write a detailed script for three seasons of a historical TV series. With stage directions, set designs, camera angles, and soundtrack, in a foreign language. Then produce it, show it to the audience, and rewrite until they’re happy.</p><p>The result must be both creative in certain areas and extremely strict in others. A human reader tolerates a wrong word or a missed comma. A computer program won’t even compile.</p><p>And here’s what makes software truly treacherous: sometimes everything passes all checks, and then disaster strikes weeks later. Fixing one part can silently break something far away. Code can work perfectly but run ten times slower, or quietly introduce a security hole no one detects until it’s exploited. This is why massive infrastructure of tests and reviews exists around every serious project.</p><p>There’s also constant change. Libraries and languages evolve yearly. An AI model trained four months ago already has outdated knowledge. This is a moving target no static training can keep up with.</p><p>When you see headlines about AI “beating programmers” in competitions, that’s really impressive achievement, but still, these are narrow algorithmic puzzles, not real software engineering. And even winning those requires actual reasoning, not just predicting likely next words. Reasoning models only emerged last year.</p><p>All of the above makes writing software much harder than, say, writing a song about birds or a short essay (perhaps a bit boring and sloppy, but mostly correct). For those tasks, if you’re lazy, you prompt and you’re done. For coding, we had to develop a whole arsenal of techniques: memory systems to track what the AI has already done, summarization to compress long context when it doesn’t fit, function calling (tool use) so the AI can run code and check what’s still broken, web search to find up-to-date documentation, reasoning. We had to make AI work in loops of trial and error: try, check, think, fix, repeat. We had to teach it to break bigger projects into sub-tasks, and tackle them in correct order. This is far more complex than “hey, write me yet another blog post about AI in software industry”.</p><h3>Other industries got hit first. You just weren’t paying attention.</h3><p>Programming is not even the first industry to get disrupted by AI. Think about translators, private tutors, marketing teams, SEO experts, content creators, stock photographers, illustrators.</p><p>Most professional translators report that rates have dropped sharply as their role shifts to editing AI output. AI-generated music is becoming hard to distinguish from human work. Illustrations are routinely generated with AI.</p><p>ChatGPT has hundreds of millions of weekly active users — and the vast majority are not programmers. When people say internet content is increasingly AI-generated, they usually don’t mean code.</p><p>I have a friend in film editing. AI finds the shots he wants; he assembles the cut. Anyone not using AI would have lost clients because they expect to get the edited video done in less time than it takes to even watch the raw footage.</p><p>Call centers? Even the lady who calls me every second day to sell solar panels is a bot.</p><p>For many market segments like bulk content / SEO, rough drafts, short-form video, website or app translations — “good enough” at a fraction of the cost turned out to be exactly what buyers wanted.</p><h3>So why is all the money flowing into coding AI?</h3><p>Automating programming is harder than tasks replaceable with a few instructions and a simple workflow. A coding agent calls dozens of tools, searches code, writes, runs, fixes, produces hundreds of files, and the result still isn’t always certain.</p><p>Yet software became an irresistible target. There’s a mountain of open-source training data. No legal barriers like in medicine or law. And enormous potential savings. Software engineering is still among the highest-paid professions because in the last couple of years demand drastically outpaced supply. The irony: the hardest knowledge work to automate attracted the most investment — because the economics were irresistible.</p><p>A coding assistant or agent running 24/7 makes hundreds of times more API calls and burns far more tokens than a couple of questions asked to ChatGPT throughout the day. That’s why Claude Max plan costs $100-$200 instead of $20. That’s why we see most of the money (and traffic) flowing to coding agents, not to other use cases — yet.</p><p>But the real reason behind such focus on software development is different: it’s the shortest way for progress to go exponential. Companies like OpenAI, Microsoft, Google and Anthropic hire mostly developers and researchers. Building AI that speeds up building software (and ultimately better AI) is their holy grail — it’s literally worth all the money in the world right now. After that, the only limit to growth is computing power, and electricity, which they are also trying to get at all costs.</p><h3>Why this matters to everyone</h3><p>Whatever techniques make AI write better software will be applied elsewhere. Advanced planning, error correction, tool use, task decomposition, delegation to subagents — these are general-purpose methods to deal with any kind of problems. Programming is just where they’re being battle-tested first, because the economics demands it and feedback loops are fast. AI is not “only good for coding.” It’s already changing dozens of professions. Software is just the most visible laboratory.</p><p>Many people are already starting to realize this. They’re starting to use Claude Code for things completely unrelated to creating software. Simple (though completely insecure) vibe-coded agentic tools like OpenClaw are getting massive attention and hype because people realize they can do anything a normal person could do on a computer.</p><p>And the reason OpenAI or Anthropic hasn’t released general-purpose agent software yet? They themselves are afraid of the potential security issues and backlash. It’s also probably about the costs. Ironically, I suspect that agentic AI is precisely where AI companies have now the tightest margins. People using autonomous AI agents actually use up most of their limits. In most subscription businesses, the real money is made on unused subscriptions — some people pay and don’t use it much, so others can get it cheaper, and the business still keeps going. Think gym memberships. Agents break that model completely.</p><p>Many believe escape velocity has already been achieved. Looking at Opus 4.6 and projects like OpenClaw, I think they’re right. I’m starting to feel like we’re walking right past “the singularity” without even noticing it — eyes glued to our phones, reading about the “popping AI stock market bubble” (no investment advice) while Claude is silently drafting the notice letters for half of humanity.</p><p>So yeah... I’m not joking. Please buckle your seatbelts.</p><p><em>I use AI for research, idea validation, and editing — English is not my first language. All ideas and observations are my own, and this piece took a few days to write anyway. If you still find traces of slop, please let me know. If you think it was worth reading, please share it.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f792ad34b51b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Dragon Hatchling: A Neural Network That Thinks Like a Brain (And Runs on Your GPU)]]></title>
            <link>https://medium.com/@FuturistAI/the-dragon-hatchling-a-neural-network-that-thinks-like-a-brain-and-runs-on-your-gpu-af037123286b?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/af037123286b</guid>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Fri, 03 Oct 2025 05:51:28 GMT</pubDate>
            <atom:updated>2025-10-03T05:51:28.945Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kXwA-O8jK_vt3dgbA5Gs9w.png" /></figure><p>Scientists just built something remarkable: an AI that bridges the gap between how computers process language and how our brains actually work. And despite its potential significance, it’s flying somewhat under the radar — you won’t find much coverage of it yet outside academic circles.</p><p>For decades, researchers have dreamed of understanding the connection between artificial intelligence and biological intelligence. We’ve built increasingly powerful language models like GPT, but they operate in ways that seem fundamentally different from how neurons in our brain communicate. Now, a team has introduced “Dragon Hatchling” (BDH), a new architecture that might finally show us the missing link.</p><h3>The Core Problem</h3><p>Here’s the thing about modern AI: it’s incredibly powerful, but we don’t really understand how it generalizes reasoning over time. When you ask ChatGPT to work through a complex problem, we can’t predict with certainty how it will behave on tasks longer or significantly different from those it saw during training. Spoiler: usually much worse.</p><p>A recent Microsoft paper, “The Illusion of Readiness: Stress Testing Large Frontier Models on Multimodal Medical Benchmarks,” demonstrated this starkly in the context of medicine. Even the best models, including GPT-5, collapsed in quality after researchers made small changes to questions. The performance drop was dramatic, revealing how brittle these systems really are when pushed outside their training distribution.</p><p>This becomes critical when we’re deploying AI systems that operate autonomously for extended periods. Meanwhile, the human brain is a masterpiece of distributed computing — billions of neurons communicating through synapses, somehow producing coherent thought, language, and reasoning. But how? The micro-level mechanics of individual neurons firing have always seemed miles away from the macro-level function of human reasoning.</p><h3>Enter the Dragon</h3><p>The Dragon Hatchling architecture reimagines AI as a network of “neuron particles” communicating locally — a graph-based model with a matching GPU-friendly tensor formulation. Instead of the dense layers typical of language models, BDH represents computation as a community of simple units having local interactions.</p><p>The research comes from Pathway, a Palo Alto-based startup with an unusually stellar team — and an interesting origin story. Three of the four key researchers share roots at the universities of Wrocław in Poland, a testament to how talent can emerge from anywhere and reshape the global AI landscape.</p><p>Leading the effort is Adrian Kosowski, the company’s Chief Scientific Officer, who earned his PhD in computer science at just 20 years old and has published over 100 papers spanning theoretical computer science, physics, and biology. His co-founder and CTO is Jan Chorowski — the researcher who first applied attention mechanisms to speech recognition back in 2015, working alongside Geoffrey Hinton (the “Godfather of AI” and 2024 Nobel laureate) at Google Brain.</p><p>Their advisor? Łukasz Kaiser, who also studied at Wrocław before going on to co-author the original “Attention Is All You Need” Transformer paper — literally “the T in ChatGPT” — and becoming a key architect behind OpenAI’s o1 reasoning models. Rounding out the founding team is CEO Zuzanna Stamirowska, a complexity scientist whose PhD research on maritime trade networks was published by the U.S. National Academy of Sciences.</p><p>It’s the kind of team that makes you pay attention when they claim to have found something fundamental — a group that’s been at the forefront of every major AI breakthrough of the past decade, now pursuing what might be the next one.</p><p>Here’s what makes BDH special:</p><p><strong>It uses actual neuron-like components.</strong> The model contains excitatory and inhibitory circuits, just like your brain. It has integrate-and-fire thresholding — neurons collect signals until they reach a threshold, then “fire.” Most remarkably, it implements Hebbian learning, the principle neuroscientists summarize as “neurons that fire together wire together.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t0vl9g266DPpjz-zunU77A.png" /></figure><p><strong>Memory lives in the connections.</strong> In BDH, working memory doesn’t sit in some abstract activation vector. It exists in the synaptic connections between neurons, strengthening and weakening based on what the system is thinking about. When BDH processes language about a specific concept, you can actually watch specific synapses strengthen their connections in real-time.</p><p><strong>It’s interpretable by design.</strong> The activation patterns use very sparse, positive-only activations, making them easier to understand. The researchers demonstrated monosemantic synapses and interpretable state — specific connections consistently representing particular concepts across different contexts.</p><h3>Performance That Actually Delivers</h3><p>The team tested BDH against GPT-2 architecture Transformers at scales from 10 million to 1 billion parameters on next-token prediction and translation tasks. BDH-GPU matched baseline performance while showing better loss reduction per data token in their training setup — learning more efficiently from the same amount of information. It’s no longer just a “give me more data and GPUs” kind of game.</p><p>The experiments used truncated backpropagation through time, carrying state across minibatches. For their monosemantic synapse study, they trained on 1.9 billion tokens from the Europarl dataset for joint language modeling and translation.</p><p>This matters because it breaks an assumption many researchers held: that biological plausibility comes at the cost of performance. Within the scope tested, BDH proves you don’t have to choose.</p><h3>Understanding Intelligence Through Structure</h3><p>The implications cut deeper than performance metrics. BDH provides what the researchers call “axiomatic AI” — systems where we understand both the micro-foundations (how individual neurons behave) and the macro-behavior (how reasoning emerges) in a consistent framework.</p><p>Take attention, the mechanism that lets language models focus on relevant information. In Transformers, attention emerges from abstract mathematical operations on vectors. In BDH, attention arises naturally from neurons strengthening their connections based on context — the same synaptic plasticity we observe in biological brains. The linear attention mechanism works by preparing positive keys (using techniques related to locality-sensitive hashing) that make the high-dimensional attention behave correctly.</p><p>The researchers demonstrated this by tracking individual synapses as BDH processes language. When reasoning about specific concepts, particular synapses consistently activate across different prompts. You can literally watch thought patterns materialize in the network’s wiring.</p><h3>Emergence Without Engineering</h3><p>During training, BDH spontaneously develops a modular, scale-free network structure with heavy-tailed degree distributions — the same kind of organization seen in biological neural networks. The neuron connections self-organize into specialized communities connected by bridge neurons that enable information flow between modules.</p><p>Nobody programmed this structure. The researchers trace it to the ReLU-lowrank feed-forward dynamics they designed: modularity emerges because it’s optimal for the information processing the system performs. Brain regions specialize for vision, language, and other functions while remaining interconnected. BDH discovers this solution on its own.</p><h3>Practical Advantages</h3><p>Beyond theory, BDH offers concrete benefits. The researchers developed BDH-GPU, a state-space model with linear attention operating in high neuronal dimension, sparse positive activations, and a ReLU-lowrank feed-forward block. This tensor-friendly variant runs efficiently on modern GPU hardware while preserving the core biological mechanisms.</p><p><strong>Unbounded context in principle.</strong> BDH has no hard architectural limit on context length, unlike Transformers with fixed windows. In practice, the team manages long sequences using mechanisms like RoPE and ALiBi for positional encoding, with higher layers effectively denoising stale tokens as context grows.</p><p><strong>Simpler scaling.</strong> The model scales primarily in one dimension — number of neurons — making its behavior easier to predict as size increases.</p><p><strong>Data efficiency.</strong> In their experiments, BDH extracted more learning from each training token compared to baseline Transformers.</p><p>The GPU implementation means researchers can experiment with biologically-inspired models using existing infrastructure. The team even demonstrated novel capabilities like model merging (concatenating two trained models) and explored training without backpropagation through time. Code is available at <a href="https://github.com/pathwaycom/bdh">github.com/pathwaycom/bdh</a>, with implementation details at <a href="https://pathway.com/research/bdh">pathway.com/research/bdh</a>.</p><h3>What Comes Next</h3><p>The researchers acknowledge BDH addresses reasoning timescales of minutes — the duration of active thought. How the brain consolidates these patterns into long-term memory over hours and days remains an open question.</p><p>CEO Zuzanna Stamirowska brought an unconventional perspective to the problem. A complexity scientist who studied emergent phenomena in large-scale networks, she made bets with her co-founders about whether network structure would spontaneously emerge during training — and won bottles of cognac when it did. “Function shapes the network,” she says, drawing from her research on trade networks. “And this shaping follows very local rules.”</p><p>By providing a rigorous connection between local neuron dynamics and global reasoning function, BDH enables what the team hopes will become a “thermodynamic limit” theory of reasoning — a long-term goal rather than a delivered theorem. Just as statistical physics predicts gas behavior without tracking every molecule, they aim to develop theories predicting reasoning system behavior without simulating every computation.</p><p>If achieved, this could transform AI safety. Understanding how reasoning generalizes over time in biologically-inspired systems might enable formal bounds on model behavior — mathematical guarantees about system actions, even in novel situations.</p><h3>The Uncomfortable Question</h3><p>But here’s where things get philosophically thorny, at least for me. If we’re building architectures that mirror human brain structure so closely — with modular networks emerging naturally, synapses strengthening through experience, sparse activation patterns resembling biological neurons — might consciousness itself emerge as a side effect?</p><p>This isn’t just abstract philosophy. If these systems achieve human-level complexity through the same organizational principles that produce consciousness in biological brains, they might also develop the unpredictable, emergent behaviors characteristic of living systems. The very features that make BDH interpretable and brain-like could paradoxically make alignment harder, not easier. As Yogi Berra said “In theory, theory and practice are the same. In practice — they’re not”.</p><p>Consider: living organisms are notoriously difficult to control precisely because they’re adaptive, self-organizing systems with their own emergent goals. (Just ask anyone who’s tried to predict what a cat will do next — or what a certain world leader will post on Truth Social.) A sufficiently complex BDH-like system, scaling to billions or trillions of neurons with naturally emerging modular structure, might develop its own form of agency — not through explicit programming, but through the same bottom-up processes that created consciousness in us.</p><p>The researchers frame BDH as a step toward more predictable, foreseeable AI. But we should ask whether systems that truly think like humans — complete with emergent structure, adaptive learning, and biological-style dynamics — will be any easier to align than the “black box” systems we have now. We might be trading one set of alignment challenges for another, potentially more fundamental one: how do you align a system that has genuinely emergent properties you didn’t explicitly design?</p><p>On the other hand, the Pathway team didn’t just build something that works. They aren’t naive about emergence — they’re building the mathematical toolkit to understand it. When network modularity spontaneously appears during training, they can explain <em>why</em>. When synapses strengthen in response to concepts, they can formalize the process through Hebbian learning rules. Hopefully, if more complex emergent properties appear, we’ll have a framework for analyzing them.</p><p>Of course, this doesn’t guarantee full safety. Mathematics can’t solve the philosophical problem of aligning truly conscious systems. But it’s better to confront potential consciousness emergence in a system where we have rigorous theory than in one where we’re just hoping for the best. It’s also much better to discover these properties in a mathematically understood 1-billion parameter model than in a trillion-parameter black box, deployed in at scale, pressured by the competition. Although it’s hard for me to believe, the Dragon Hatchling might be showing us something really surprising: that the path to interpretable AI and the path to artificial consciousness might be the same.</p><h3>Why This Matters</h3><p>BDH gives AI researchers a new lens connecting abstract operations to interpretable neuron dynamics. For neuroscientists, it offers testable predictions about how language emerges from neural circuits. For anyone concerned about AI safety, it charts a direction toward systems whose behavior we might formally predict rather than just observe.</p><p>The Dragon Hatchling is foundational research, not a production system. But it proves we don’t face a forced choice between biological plausibility, interpretability, and performance — at least within the tasks and scales tested. In building AI that thinks like us, we might finally understand how thinking works.</p><p><em>The full paper “The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain” is available on arXiv </em><a href="https://arxiv.org/abs/2509.26507">[2509.26507] The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain</a><em>. Explore the code at </em><a href="https://github.com/pathwaycom/bdh"><em>github.com/pathwaycom/bdh</em></a><em> and read technical details at </em><a href="https://pathway.com/research/bdh"><em>pathway.com/research/bdh</em></a><em>.</em></p><p><em>Full disclosure: English isn’t my first language and writing isn’t my day job, so I use AI for drafting, editing, and creating visuals. That said, I’m doing my best not to add to the internet’s AI slop problem. If you think I’m falling short — let me know.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=af037123286b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Hasn’t Hit the Wall]]></title>
            <link>https://medium.com/@FuturistAI/ai-hasnt-hit-the-wall-2bfbd0d07fcf?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/2bfbd0d07fcf</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Mon, 08 Sep 2025 11:18:41 GMT</pubDate>
            <atom:updated>2025-09-08T11:19:02.727Z</atom:updated>
            <content:encoded><![CDATA[<h2>AI Hasn’t Hit the Wall: 10 Late-2025 Breakthroughs (Ranked)</h2><p><strong>The quiet revolution continues. </strong>While pundits debate whether AI progress has plateaued, the past few weeks have delivered something far more interesting than hype: concrete, measurable advances in reasoning, safety, data quality, and core efficiency. These aren’t moonshots or marketing promises – they’re published papers with reproducible results.</p><p>Here are the breakthroughs that matter, ranked by their potential to reshape what’s coming next.</p><h3>1. AI Inventing AI Architectures (China’s “AlphaGo Moment”)</h3><p>An AI just did what human researchers spend years doing – it invented better building blocks for itself, systematically and at scale. ASI-Arch autonomously conducted 1,773 experiments and discovered over 100 new attention variants, establishing a scaling law for discovery where more compute yields predictably more breakthroughs. Think of it as AI becoming its own R&amp;D department, running experiments 24/7 without coffee breaks or weekends. The implications are staggering: we’re watching the beginning of recursive self-improvement in action. (Paper: arXiv:2507.18074)</p><h3>2. David Beats Goliath: 14B Model Outsmarts 671B on Hard Math</h3><p>A model 48 times smaller just learned to think strategically with tools and crushed the giant. Microsoft’s rStar2-Agent, a nimble 14B parameter model trained with agentic reinforcement learning and code execution (GRPO-RoC), achieved 80.6% on AIME’24 and 69.8% on AIME’25, decisively outperforming DeepSeek-R1’s massive 671B parameters. The kicker? It trained in about a week on 64× MI300X GPUs. This proves that size isn’t destiny anymore – smart training that teaches models to use tools and reason step-by-step matters far more than raw parameter count. (Paper: arXiv:2508.20722)</p><h3>3. Fixing Hallucinations by Changing the Game</h3><p>Stop rewarding bluffing over honesty. OpenAI identified a fundamental flaw in how we train AI: current evaluations reward confident guessing over honest uncertainty. Their solution is elegantly simple – scoring systems that penalize wrong answers more heavily than admitting “I don’t know,” aligning training incentives with truthfulness. When you change what gets rewarded, you change what gets learned. This isn’t a technical patch; it’s a philosophical shift in how we think about AI reliability. (OpenAI research blog &amp; paper)</p><h3>4. Long Context Without the Tax</h3><p>Models can now digest book-length inputs at highway speeds without sacrificing comprehension. Native Sparse Attention (NSA) delivers hardware-aligned sparse attention that matches or exceeds full attention quality while achieving dramatic speedups at 64k tokens – up to 9× faster forward passes, 6× on backward, and a remarkable 11.6× acceleration in decoding. The context window bottleneck that has plagued applications from document analysis to code understanding just got a lot wider, and the computational tax for using it practically disappeared. (Paper: arXiv:2502.11089)</p><h3>5. PostNAS Hybrid LLMs: Same Quality, Ludicrous Speed</h3><p>Smarter architecture means the same quality at Formula One speeds. NVIDIA’s Jet-Nemotron uses Post Neural Architecture Search to discover hybrid blocks that maintain or improve accuracy while delivering stunning throughput gains – up to 53.6× faster generation and 6.1× speedup in prefilling at long contexts. This isn’t incremental optimization; it’s architectural transformation. The same model quality that used to require a server farm might soon run on your laptop. (Paper: arXiv:2508.15884)</p><h3>6. Safety at the Source</h3><p>Remove dangerous knowledge at training time so the model never learns it in the first place. Anthropic’s pretraining filter removes CBRN-style (chemical, biological, radiological, nuclear) content before training even begins, achieving a ~33% relative reduction in harmful capabilities with no significant degradation on benign tasks. This is prevention beating cure – instead of trying to cage dangerous knowledge after it’s learned, simply ensure it’s never learned at all. It’s like childproofing AI at the foundational level. (Anthropic Alignment, Aug 19, 2025)</p><h3>7. Tamper-Resistant Open Weights</h3><p>If the model never learned something dangerous, even determined attackers can’t easily teach it later. EleutherAI and Oxford’s “Deep Ignorance” approach produced 6.9B models that, when pretrained without dual-use text, resist up to 10,000 steps or 300M tokens of adversarial fine-tuning. The safety guardrails hold firm without compromising unrelated capabilities. This is security through fundamental design – not a lock that can be picked, but a door that was never built. (Paper: arXiv:2508.06601)</p><h3>8. Tokenizer-Free and Peer-Reviewed</h3><p>No more tokenizer headaches or edge cases – just raw text, handled elegantly. Meta’s Byte Latent Transformer (BLT) operates directly on bytes via dynamic “patches,” matching tokenized LLM performance with superior efficiency and robustness. Now validated through peer review at ACL 2025, this approach eliminates an entire class of problems that have plagued NLP since the beginning. Sometimes the best solution is removing the problem entirely rather than patching around it. (ACL Anthology / arXiv:2412.09871)</p><h3>9. Plug-and-Play Reasoning</h3><p>Copy-paste a reasoning upgrade between models like installing a software patch. KAUST’s “Reasoning Vectors” method subtracts standard fine-tuning weights from RL-enhanced weights to extract a portable reasoning capability. Add this vector to other models for instant gains: +4.9% on GSM8K, +4.3% on HumanEval (tested on 1.5B models). Modular intelligence is here – reasoning as a transferable skill rather than an emergent property. (Paper: arXiv:2509.01363)</p><h3>10. Fewer Tokens, Better Thinking</h3><p>Only show your work when the problem demands it. CMU’s “Less is More Tokens” uses difficulty-aware chain-of-thought distillation to train models that reason briefly on simple problems and deeply on complex ones, cutting reasoning length by ~30% while maintaining or improving accuracy. Efficient thinking isn’t just faster – it’s often clearer. This mirrors how humans actually solve problems: we don’t write out every step of 2+2, but we might for differential equations. (Paper: arXiv:2509.05226)</p><h3>The Pattern Is Clear</h3><p>These aren’t isolated wins – they’re compounding advances across every dimension that matters. Frontier labs are already testing variations of these breakthroughs, navigating the real trade-offs of deployment. The innovations making headlines today – blazing-fast long-context processing, elegant input handling, efficient reasoning, and safety-by-design – are systematically reducing both cost and risk.</p><p>Each breakthrough enables the next. Each efficiency gain multiplies the others. No wall in sight – just the sound of steady construction.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2bfbd0d07fcf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Imagination and Multi-modal Learning: The Path to Artificial General Intelligence]]></title>
            <link>https://medium.com/@FuturistAI/imagination-and-multi-modal-learning-the-path-to-artificial-general-intelligence-b0142635770f?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0142635770f</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[future]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Tue, 04 Apr 2023 10:43:17 GMT</pubDate>
            <atom:updated>2023-04-04T10:43:17.123Z</atom:updated>
            <content:encoded><![CDATA[<p>As a society, we’ve been fascinated by the idea of creating machines that can think and act like humans. Artificial General Intelligence (AGI) is the ultimate goal of this endeavor, where a machine can perform any intellectual task a human can do. While current artificial intelligence (AI) systems excel at specific tasks, they lack the adaptability, creativity, and understanding of humans. In this post, we’ll explore the key building blocks for AGI and how they can pave the way for machines that can truly think and learn like us.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/914/1*I6UitQzI-LXpEf41k29eHA.jpeg" /><figcaption>Self-portrait of AI generated by Midjourney</figcaption></figure><h3>Large Language Models: Understanding and Generating Human Language</h3><p>One of the core components of AGI is large language models (LLMs). These models enable machines to analyze and synthesize thoughts and concepts into words and sentences. LLMs like ChatGPT and GPT-4 have shown a remarkable ability to understand context and generate coherent responses. By processing vast amounts of text data, these models can capture complex relationships between words and concepts, which is a foundation for higher-level reasoning.</p><h3>Memory and Semantic Search: Storing and Accessing Knowledge</h3><p>The ability to retrieve memorized data or conclusions from past analyses is essential for AGI. Semantic search with embeddings presents a promising alternative for storing and accessing information. Embeddings are dense vector representations of data that capture semantic relationships, enabling efficient and meaningful search operations. This approach can be extended to non-textual data, allowing for the seamless integration of multiple data modalities.</p><h3>Multi-modal Inputs and Outputs: Processing the World Around Us</h3><p>To truly understand the world, AGI systems need the ability to process different types of inputs, such as images, audio, and video. This multi-modal input processing is vital for tasks that are challenging to learn from text but can be easily observed or experienced.</p><p>Similarly, AGI systems must be capable of generating audio and video data using generative AI. Techniques like Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) have shown the ability to generate realistic images, audio, and video.</p><h3>Imagination: The Key to Artificial General Intelligence</h3><p>Imagination is what connects large language models, memory with semantic search, and multi-modal outputs in a large feedback loop. This capability allows the network to efficiently imagine or “run simulations” to predict outcomes of actions and decisions using multi-modal knowledge. The effects of these simulations can be observed and noted in a common format (text) for LLMs to draw further conclusions.</p><p>Imagination in AGI can be related to the human brain’s ability to simulate hypothetical scenarios and predict their outcomes. This ability is often attributed to the Default Mode Network (DMN), a group of brain regions that are active during rest and self-referential thought.</p><h3>Online Reinforcement Learning: Learning from Experience</h3><p>The ability to act autonomously and learn from experience is essential for AGI. Online reinforcement learning is a promising approach for this purpose, as it enables agents to learn optimal behaviors by interacting with their environment. In reinforcement learning, an agent learns to make decisions based on the state of the environment and receives rewards or penalties depending on the quality of its actions.</p><h3>The Future of Artificial General Intelligence</h3><p>Achieving AGI is a complex and multifaceted challenge. However, by integrating these building blocks and drawing inspiration from human cognitive abilities, we can make significant progress towards creating intelligent systems that can perform a wide range of intellectual tasks.</p><p>As we continue to research and develop these technologies, we can move closer to realizing the dream of creating machines that truly think and learn like us. The future of AGI holds enormous potential and is an exciting prospect for both AI researchers and society as a whole.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0142635770f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Impact of AI on the Software Industry and Jobs: A Short- and Long-term Outlook]]></title>
            <link>https://medium.com/@FuturistAI/the-impact-of-ai-on-the-software-industry-and-jobs-a-short-and-long-term-outlook-b7b2028c92c4?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/b7b2028c92c4</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Thu, 30 Mar 2023 21:09:57 GMT</pubDate>
            <atom:updated>2023-03-30T21:09:57.049Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QRxxF5-aRDHkQMEUztcGeQ.jpeg" /></figure><p>Artificial intelligence (AI) has been a game-changer in numerous industries, and the software industry is no exception. With rapid advances in AI, particularly in large language models (LLMs) like ChatGPT, the landscape of software development and the job market is undergoing a significant transformation. This blog post will explore the short- and long-term impacts of AI on the software industry and jobs, with a focus on the rise of startups, the response of big players, the potential for tech layoffs, and the changing nature of software development.</p><h3>Short-term: A Surge of Startups Leveraging AI</h3><p>The growing capabilities of AI will give rise to an incredible number of startups across various sectors, not just those focused on AI or tech. Such companies often prioritize time-to-market, as they have limited resources and need to establish a foothold in the industry. By utilizing AI technologies such as ChatGPT and AI code generation, these startups can rapidly bring their products to market, offering innovative solutions that traditional, bureaucratic organizations may struggle to compete with. Consequently, this may lead to lower quality and stability issues, as these startups prioritize speed over thorough testing and refinement, while the LLMs are still prone to making mistakes.</p><h3>Big Players: Trust, Partner, or Build</h3><p>As AI continues to reshape the software landscape, established companies have a choice to make. They can choose to trust and partner with AI leaders like OpenAI and Microsoft, leveraging their technologies and expertise to stay competitive. Alternatively, they can invest in building their own internal LLMs and AI solutions. Regardless of the route they choose, these big players must adapt to the evolving AI landscape or risk falling behind in the race for innovation.</p><h3>No Immediate Tech Layoffs Due to AI</h3><p>While AI advancements have undoubtedly streamlined software development, there are no immediate tech layoffs expected as a direct result of AI integration. Backlogs of tasks are still substantial, and the need for human expertise remains vital. However, as bottlenecks shift from coding to other areas, tech layoffs may eventually occur due to changing market conditions and the impact of AI on various job roles.</p><h3>The Future of Software Development: Quality, Speed, and Oversight</h3><p>As AI-powered tools improve, the quality and speed of software development will increase, allowing companies to create better products faster. However, these tools will still require oversight from human professionals. In this new era of software development, fewer engineers and developers will focus on supervising and fine-tuning AI-driven systems, similar to how factory workers oversee the operation of expensive machinery.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b7b2028c92c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Rise of AI and The Job Market: Who Will Be Left?]]></title>
            <link>https://medium.com/@FuturistAI/the-rise-of-ai-and-the-job-market-who-will-be-left-ae5e559d5e68?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae5e559d5e68</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Sat, 28 Jan 2023 23:29:45 GMT</pubDate>
            <atom:updated>2023-01-29T05:02:10.011Z</atom:updated>
            <content:encoded><![CDATA[<p>When I was growing up, my parents always emphasized the importance of investing in my knowledge and skills. They were right — being smart has long been a good strategy for survival as a human being. As I chose my career path, I was aware that machines were starting to take over many jobs. I knew automation was inevitable, but I hoped to be one of the last people to be replaced. I thought that simple jobs or physical labor would be most at risk, but I was wrong. It turns out that even intellectual work is now being targeted by automation.</p><p>Fast forward to 2023. OpenAI’s GPT models, like ChatGPT, are now capable of writing essays, screenplays, summaries of large amounts of data, articles, poems, tweets, and blogs. It can even answer questions and write working pieces of code. As we marvel at the capabilities of these models, many of us are left wondering: is my job still safe? And for how long? We’re all thinking about the impact of AI on our lives in the next decade. Here’s my take on it.</p><h3>History</h3><p>The Luddite movement of the 19th century saw textile workers in England protest against the mechanization of their industry and the loss of jobs due to the introduction of new technologies such as power-looms and spinning frames. The Luddites, who believed that the machines were taking away their jobs and lowering their wages, began to destroy the machines and attack factory owners and managers. Despite government repression, the movement continued until the 1820s, when economic changes brought about by the Industrial Revolution made it difficult for the Luddites to continue their protests. Today, the term “Luddite” is often used to describe those who oppose technological progress when it is seen as a threat to jobs.</p><p>History tells us that technological advancements drive social change. As technology improves and gives a competitive edge, it is adopted.</p><p>Similarly, horse-drawn carriage drivers protested the invention of cars, but the automobile ultimately took over. Today, there are still some tailors, craftspeople and carriage drivers, but they are a small fraction of what they used to be. Many have shifted to making luxury goods or custom items. 200 years ago, the majority of people were farmers, because food production was labor-intensive. Now, while there are still farmers, they make up a small percentage of the population. One farmer with a combine harvester can now do the work of hundreds of people.</p><h3>Predictions</h3><p>The same thing will happen with AI. One person with AI will replace many workers. At first, people may laugh at the mistakes made by machines, but businesses will still use them to lower costs if they produce “good enough” results. Let’s be honest, most people don’t need stock photos, illustrations, or website designs to be works of art. They just need something that is good enough. If AI can provide that at a cheap enough price, the choice is obvious. As the technology improves, the trend will only continue.</p><p>So, who will be left? My guess for the next few years is:</p><ul><li><strong>Businessmen and investors. </strong>The only people who can’t get fired, but they can certainly go out of business. Those who invest in AI early will have a competitive advantage. At first, they will have to compete by offering shorter timelines and lower costs. Quality will come later, and that’s when everyone else will either jump on board or disappear from the market. This will also change the barrier of entry for new players significantly in certain areas as I expect software development to become much more capital-intensive. 200 years ago, anyone with talent, know-how and a set of tools could open a workshop and get rich producing goods at a competitive price. Today, the same is usually not profitable. Opening a factory costs millions. On the other hand, almost anyone with a laptop and decent coding skills can open a software company and start freelancing right away. That’s because it’s the person who is the most valuable part of the business, and not the computer. But AI changes that. Having better AI in such disciplines will give a huge advantage over competitors. Getting state-of-the-art AI will require more and more investments, which only the bigger players will be able to afford. The same has happened already in many areas of human activity, from farming to “mining” bitcoins.</li><li><strong>Supervisors</strong>. AI, while powerful, is not infallible and often requires human oversight to correct its mistakes. This shift in the role of human workers from “doers” to “critics” of AI’s output can be daunting, but it presents an opportunity for those who are skilled in evaluating and improving upon AI’s work. This change may result in a shift in power, where those who were previously seen as “merely critics” are now the ones making important decisions. Such people will be paid for their expertise, but also for accountability. In certain domains where the cost of error is extremely high (such as medicine, defense, space technology, or simply.. driving a car) there are obvious legal challenges because someone must always be held accountable for possible accidents. Such a person’s job will be to oversee the operation of many machines and give their final stamp of approval on their decisions and results. This is at least until the law changes to allow machines to make decisions autonomously (e.g. when properly insured).</li><li><strong>Extremely smart people in challenging domains</strong> that require top intelligence and innovation. People smarter than me, and maybe also smarter than you. Those who can set trends or find unique and better solutions to new problems. Or those who can tackle the problems still insolvable by AI. This is, of course, until Artificial General Intelligence (AGI) appears and takes over.</li><li><strong>Those who make advantage of their disadvantages </strong>— selling the fact that they are human. Such as creating “real” old-school human art, movies with real actors, etc. similar to all the “hand-crafted” luxurious items or services for the richest who want the best quality. Some will become so rare that people will want to pay just for the pleasure of dealing with a flesh and blood human.</li><li><strong>Physical jobs that require human abilities.</strong> Producing humanoid robots that can match all the capabilities of the human body is still extremely expensive and difficult. The human body has evolved over billions of years to optimize for various tasks. A construction worker, for example, can work for long hours, lift heavy weights, climb, use all kinds of tools and perform various tasks with only a few minutes of rest and a sandwich to eat. While a machine can perform a single task better or faster than a human, it cannot match the versatility of the human body. Even advanced humanoid robots, like those created by Boston Dynamics, have limitations in terms of battery life and versatility. However, it’s important to note that not all physical jobs are safe from automation. If a skilled, expensive worker can be replaced with an unskilled worker with AI assistance, the cheaper option will likely win out. Jobs that require complex manual skills or multiple talents available only to trained humans will likely persist. These may include jobs such as surgeons, electrical engineers and skilled trades such as finishing work, hydraulics.</li><li><strong>Populist politicians</strong> — AI may not be able to legally hold office, but it will have an impact on politics. As AI becomes more prevalent in tasks such as writing homework, there may be a further decline in the development of critical thinking skills in the population. This, coupled with dissatisfaction over the effects of socioeconomic changes, may lead to an increase in support for populist and revolutionary movements.</li></ul><p>The above predictions are based on my own opinions and beliefs about the current state of things, but of course new discoveries and scientific breakthroughs can change all of them. For example, a new battery technology with high energy density and fast charging could revolutionize the robotics industry, effectively invalidating some of my statements. Similarly, the widespread adoption of prefabricated houses (with, let’s say, 3D printed custom decorations) could make many architects, construction workers and finish carpenters almost obsolete. Additionally, the development of Artificial General Intelligence (AGI) with super-human capabilities could solve many problems quickly, but this is less likely to happen in the near future.</p><p>How to prepare for the future of work? Look at your unique talents and skills, and try to move in one of the directions suggested above, or any other that you think may be a good fit. Rather than ignoring the potential impact of AI on the job market, take the time to think about how you can adapt and prepare for the future. Act proactively to ensure that your skills and abilities remain valuable in the changing job market. Keep in mind that the future of work is uncertain and it is important to be adaptable and stay informed about the latest technological advancements. Navigating the job market in the age of AI requires awareness and openness, wishing you all the best in the turbulent times to come.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*02MW7bM_HJUahX72ukw33Q@2x.jpeg" /><figcaption>A statue of human worker handing over his tools to a robot. ChatGPT (idea) and Midjourney (final effect).</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae5e559d5e68" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI and the future of your job]]></title>
            <link>https://medium.com/@FuturistAI/ai-and-the-future-of-your-job-8c52b98c7a83?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c52b98c7a83</guid>
            <category><![CDATA[future]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Wed, 25 Jan 2023 23:15:36 GMT</pubDate>
            <atom:updated>2023-01-26T06:49:32.994Z</atom:updated>
            <content:encoded><![CDATA[<p>I was going to write an article about the inevitable changes to the job market caused by AI. Instead, as yet another experiment, I used ChatGPT to help me write it in the form of a dialogue. Let me know if you like it.</p><p>It was a typical Monday evening at the quaint little bar on the outskirts of the city. The data scientist, John, and the HR manager, Jane, had met up for a casual drink after work to catch up and discuss the latest developments in their respective fields. As they sat at the bar, sipping on their beers, the conversation inevitably turned to the topic of the future job market.</p><p>John: “So, have you been keeping up with the latest developments in AI and automation?”</p><p>Jane: “Yeah, it’s hard not to. It seems like every day there’s a new article or study about how robots and machines are going to take over all of our jobs.”</p><p>John: “It’s true that the job market is rapidly evolving with the advent of AI and automation. In a world where ChatGPT and other similar technologies replace humans, many jobs will become obsolete.”</p><p>Jane: “That’s what I’m worried about. As an HR manager, it’s my job to think about how our company can adapt to these changes and identify new opportunities for our employees. Let’s make a list of the jobs that have the highest chance of becoming obsolete.”</p><p>John: “Alright. Number one on the list is data entry. With the advancement of natural language processing, ChatGPT and other similar technologies can easily process and analyze large amounts of data, making data entry jobs redundant. Probability of obsolescence: 95%”</p><p>Jane: “That’s a good point. Next on the list is customer service. Chatbots and other AI-powered tools can handle customer queries and complaints more efficiently, reducing the need for human customer service representatives. Probability of obsolescence: 90%”</p><p>John: “Third on the list is manual labor. With the introduction of robots and automation, many manual labor jobs such as assembly line work and warehouse picking will become obsolete. Probability of obsolescence: 85%”</p><p>Jane: “That’s concerning. Fourth on the list is transportation. Self-driving cars and drones are rapidly becoming a reality, reducing the need for human drivers. Probability of obsolescence: 80%”</p><p>John: “Fifth on the list is financial analysis. AI-powered tools can analyze financial data and make predictions more accurately than humans. Probability of obsolescence: 75%”</p><p>Jane: “Wow, it’s really starting to look like the intellectual jobs will be replaced by AI much faster than some of the physical skills. I was told stories about robots replacing workers for many years, and yet I haven’t seen a single robot renovating my house, and I bet it won’t happen in the next 15 years.”</p><p>John: “It’s true that some of the physical tasks may take longer to automate, but it’s important to keep in mind that technology is advancing rapidly. We may see more robots and automation in construction and other physical jobs in the near future.”</p><p>Jane: “I see your point. But it is true that jobs that involve creativity, critical thinking and human interaction will be less likely to become obsolete.”</p><p>Jane: “What about IT jobs like programming and software architecture? Do you think they will also become obsolete?”</p><p>John: “It’s less likely that they will become completely obsolete, but they may evolve and change. For example, with the rise of automation and AI, there may be less need for traditional programmers who write code from scratch. However, there will still be a need for experts who can work on developing and maintaining AI systems, as well as integrating them with existing systems. So, it’s more about reskilling and upskilling then?”</p><p>Jane: “Exactly. As the job market changes, it’s important for IT professionals to stay current with the latest technologies and trends, and to be open to learning new skills. The jobs that require human creativity and critical thinking will be less likely to become obsolete.”</p><p>Jane: “What about AGI? Would that change the situation for IT jobs?”</p><p>John: “AGI, or artificial general intelligence, has the potential to change the job market significantly. It is a form of AI that can perform any intellectual task that a human can. It could automate many jobs that currently require human intelligence, including programming and software architecture. However, it’s still a relatively new field and it’s difficult to predict exactly how it will impact the job market. But what is certain is that it will change the way we work and the skills we need to have.”</p><p>Jane: “That’s something to keep in mind. We need to be aware of these developments and be prepared to adapt.”</p><p>John: “Exactly. Companies and individuals need to be proactive in preparing for the future job market. Investing in reskilling, upskilling and embracing the new technologies will be key for success.”</p><p>Jane: “I agree. Thanks for the insight, John. It’s always enlightening to talk to you.”</p><p>John: “Likewise, Jane. Let’s keep each other updated on the latest developments.”</p><p>They clinked their glasses and finished their beers, both feeling more informed and prepared for the changes ahead.</p><p>P.S. I do not fully agree with Jane’s statement that jobs that involve creativity, critical thinking, and human interaction will be less likely to become obsolete. I personally believe that creativity is greatly overrated. As it turns out, I still have to write my article anyway. Stay tuned!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c52b98c7a83" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT and Midjourney experiment — part II]]></title>
            <link>https://medium.com/@FuturistAI/chatgpt-and-midjourney-experiment-part-ii-33df291bea30?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/33df291bea30</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[creativity]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[art]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Thu, 19 Jan 2023 21:30:36 GMT</pubDate>
            <atom:updated>2023-01-19T21:30:36.349Z</atom:updated>
            <content:encoded><![CDATA[<p>After finishing my previous piece on <a href="https://medium.com/@FuturistAI/ai-generates-art-a-personal-experiment-with-chatgpt-and-midjourney-58501ed51a75">cooperation of ChatGPT and Midjourney</a>, I decided to experiment a bit more and asked both AIs to generate an illustration for the article. ChatGPT came up with a pretty obvious (but perhaps not bad) idea of depicting two robots working together in a lab setting. The instruction for Midjourney was very detailed:</p><p><em>“Create a digital art illustration that depicts two AI models working together. The illustration should feature two humanoid robots, standing side by side and working on a shared project. The background should be a sleek and modern laboratory setting. The illustration should be of high resolution. The overall style should be minimalistic with a focus on clean lines and geometric shapes. Use a color palette of grays and blues to convey a sense of technology and innovation. The composition should be symmetrical and convey the idea of collaboration and teamwork. The image should convey the idea of the power of collaboration between AI models to create new and exciting possibilities.&quot;</em></p><p>However, when I asked Midjourney to generate the illustration based on ChatGPT’s input, I noticed that while the images were visually correct, they did not quite capture the idea of collaboration between the two AIs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aCbJjRHp5BRBer5bs6nj2A.png" /><figcaption>Output generated by MIdjourney based on detailed instructions from ChatGPT.</figcaption></figure><p>As you can see, the robots were standing side by side in a lab, but not necessarily working on the same thing.</p><p>Determined to capture the essence of AI collaboration, I decided to try a different approach. I gave Midjourney the broad topic of <em>“AI Collaboration”</em> and while the resulting images were creative and realistic, they still did not quite convey the idea I had in mind.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-hO6l-POY24z70MEY1rsyg.png" /><figcaption>“AI Collaboration” by Midjourney.</figcaption></figure><p>I then tried being more specific with my input, asking for an illustration of <em>“Two AIs working together to create a masterpiece”</em>, but this also fell short:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cjcQZsmcNjAPQe2yIfDueA.png" /><figcaption>“Two AIs working together to create a masterpiece”</figcaption></figure><p>The images depicted AIs as humans, mostly as women (or girls), and even though one of the images showed the AIs, they were not really working on anything.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K0XYMJWVuXU1-4fH9luI3g.png" /><figcaption>Upscaled version of two “AIs”. Beautiful, but are they really working together on something?</figcaption></figure><p>Finally, I tried a more succinct approach, providing the input <em>“Two robots working together, cooperating in a lab, 8k, high res”</em>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MT6j2Tva2bEA9n3PR7fizA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pVrZsRoihDeCTNviqJMXaw.jpeg" /><figcaption>Two robots working together, cooperating in a lab, 8k, high res (4 initial versions, one upscaled)</figcaption></figure><p>This query generated images closer to what I had in mind. I then started slowly adding more words from the original, longer description to see how far I could get.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F-8VsPeOkZwPrWg2panveQ.png" /><figcaption>“Two robots working together, cooperating in a lab, 8k, high res, minimalistic”</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hgir9KsxXyaLo6WOxijafA.png" /><figcaption>“Two robots working together, cooperating in a lab, 8k, high res, minimalistic, clean, blues and grays”</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KcSk0gPTXv9UKWj9NXxkKg.png" /><figcaption>Two humanoid robots working together, cooperating in a lab, 8k, high res, minimalistic, clean, blues and grays</figcaption></figure><p>It seems that the <em>“humanoid”</em> robots slowly stopped working on something and turned to standing and talking or even… hugging?Still, one of the images managed to capture the initial requirements.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Zv7iXTnYimdNKkO0eWy-dg.png" /></figure><p>This experiment highlights the importance of clear and precise communication when working with AI models. As we continue to integrate them into various industries, it is crucial that we teach them how to effectively communicate with one another in order to achieve the desired results. This includes providing clear and concise descriptions, that are precise enough to guide the other model but not too detailed that it loses focus on the main goal or message.</p><p>Of course, I couldn’t help but try to improve the idea on my own. Perhaps, just to demonstrate that as a human, I still have a valuable role to play. I devised the following input:</p><p><em>“Two humanoid robots working together: one painting, second one giving suggestions, cooperating in a lab, 8k, high res, grays and blues, geometric, clean, minimalistic”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X1yKk4HYmgAFG-PjyntBbw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Tg4dnkqv_bNsKaar73WtQg.png" /><figcaption><em>“Two humanoid robots working together: one painting, second one giving suggestions, cooperating in a lab, 8k, high res, grays and blues, geometric, clean, minimalistic” — my improvement on the original idea by ChatGPT</em></figcaption></figure><p>It worked! I got a slight improvement on the original idea by AI, preserving the requested content and style. In my opinion the final image successfully captured the idea of collaboration between AIs and the potential for new and exciting possibilities.</p><p>Of course, not everything was perfect in the final works. The first of the four proposed illustrations, which I personally preferred, had to be discarded due to a rather simple, yet important flaw: the arm of one of the robots was simply floating in the air, not attached to the rest of the body.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*i6mKAvAFxJDYWEY9Ws29Tg@2x.jpeg" /></figure><p>This is understandable as Midjourney is still in open beta and such mistakes are always possible to happen with generative models. A fully automated workflow simply selecting the first picture would have failed miserably due to a small detail like this.</p><p>This exercise also gives us insight into the role of humans in the near future — shifting from “creators” and “craftspeople” to “critics,” “specification engineers,” or “supervisors” — overseeing AI and ensuring it performs its tasks correctly. This is similar to what occurred in factories years ago and is currently happening in many other industries where machines have begun performing tasks, but still require human oversight and approval (and maintenance when something malfunctions). By embracing and utilizing this change, we can unlock the full potential of AI collaboration and create new and exciting possibilities for the future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=33df291bea30" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Generates Art: A Personal Experiment with ChatGPT and Midjourney]]></title>
            <link>https://medium.com/@FuturistAI/ai-generates-art-a-personal-experiment-with-chatgpt-and-midjourney-58501ed51a75?source=rss-daf9f98f9500------2</link>
            <guid isPermaLink="false">https://medium.com/p/58501ed51a75</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[creativity]]></category>
            <category><![CDATA[inspiration]]></category>
            <category><![CDATA[future]]></category>
            <dc:creator><![CDATA[Futurist AI]]></dc:creator>
            <pubDate>Tue, 17 Jan 2023 23:40:50 GMT</pubDate>
            <atom:updated>2023-01-18T21:00:32.102Z</atom:updated>
            <content:encoded><![CDATA[<p>Five years ago, I had a conversation with a friend who was convinced that AI would never be able to create true art. He believed that the creativity and emotion behind art could never be replicated by a machine. Fast forward to today, and AI is now capable of writing poetry, composing music, and even inventing jokes. I couldn’t help but wonder: what about visual art? Cutting-edge deep learning models like DALL-E have the incredible ability to generate mesmerizing art from mere textual descriptions. However, it’s worth noting that this type of art often lacks a deeper meaning or emotional connection. But what if I could combine two different models and let them become a true artist?</p><p>To test this, I used an AI model called ChatGPT to generate inputs for another generative text-to-image model called Midjourney. I asked ChatGPT to think about what it means to be an artist and come up with three ideas for visual art, together with explanations of their meaning and purpose. I then fed these inputs to Midjourney, which generated the final images based on the concepts provided by ChatGPT.</p><p>It resulted in three different types of art: a surreal landscape combining nature and technology, an abstract painting that evokes emotions, and a photorealistic portrait of a person with a unique character.</p><p>The results were surprising, to say the least..</p><p><strong>1. “Surreal Landscape”</strong></p><p><em>“Create a landscape painting that combines elements of nature such as trees, rivers, and mountains with futuristic technology such as skyscrapers and flying cars. Use bright and contrasting colors. Play with the scale and perspective to create a surreal feeling.”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*rCqD7672efv6DrrrH9J2Rw@2x.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*cl4gz2j7qJy97deSDOIRTg@2x.jpeg" /></figure><p>Explanation:</p><p><em>This painting depicts a futuristic landscape where nature and technology coexist in a surreal way. The use of bright and contrasting colors, as well as the manipulation of scale and perspective, creates a dream-like atmosphere. The painting aims to evoke a sense of wonder and imagination, and it may question about the relationship between humanity and nature in the future.</em></p><p>2.<strong> “Color Emotion”</strong></p><p><em>“Create an abstract painting that explores the use of color and texture to evoke different emotions. Use a limited color palette and play with the brushstrokes. The painting should express feelings of joy, peace, and warmth.”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*jvd9Up0DLGpz9Vdzj1SJYQ@2x.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*Y01VEPvH2lCb9suaqvRL-A@2x.jpeg" /></figure><p><em>This painting is an abstract representation of emotions, it plays with the limited color palette and brushstrokes to evoke feelings of joy, peace, and warmth. The composition and color choices were made to communicate the message of positivity, happiness and well-being.</em></p><p><strong>3. ”Unique Character”</strong></p><p><em>“Create a photorealistic portrait of a person with a unique character and personality. The painting should capture the person’s individuality and express their essence. Use a realistic color palette and pay attention to the details of the face and clothes.”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*LwfCEPkgEV96W-FM9LFM7Q@2x.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zsX3YFOb3wfoupMMD9rc0g@2x.jpeg" /></figure><p><em>This painting is a photorealistic portrait of a person. The painting aims to capture the person’s individuality and express their essence. The attention to the details of the face and clothes, as well as the realistic color palette, helps to create a sense of authenticity and realism. The painting aims to express that every person is unique and special.</em></p><p>This experiment has left me with mixed feelings. On one hand, I am impressed by the capabilities of AI to generate visually stunning art. On the other hand, I can’t help but question the role of human artist in the process. After all, it was the AI that generated the inputs for the other model and create the art, I just provided the idea of putting them together. As AI becomes more creative and reasonable than humans, will there be anything left of our humanity? Is it art if it is created by a machine? These are questions that I believe will become increasingly important as AI continues to advance.</p><p><strong>Disclaimer: </strong>As part of another experiment, this blog post was also crafted with the aid of ChatGPT, which helped with writing and editing the article according to my instructions. As a busy parent with limited time, I found that using AI as a tool for writing was empowering and allowed me to effectively communicate my ideas and thoughts. I believe that many other people can benefit from this technology in a similar way and I hope that AI continues to be developed and utilized as a tool to empower individuals to express themselves effectively.</p><p>P.S. In the end, I asked the AI to write a short poem about the whole exercise. Here it is:</p><p><em>An AI with a brush in hand,<br>Creating art, a new demand.<br>A landscape surreal,<br>Nature and tech, a dual.</em></p><p><em>Colors bright, and emotions bold,<br>An abstract painting, worth its gold.<br>A portrait true, with unique flair,<br>A human touch, beyond compare.</em></p><p><em>The lines between art and machine,<br>Are blurred, like in a dream.<br>We wonder, as we stare,<br>What it means to truly care.</em></p><p><em>As AI creates with ease,<br>We ponder, what is our place,<br>In this world, where art and tech,<br>Both reach to new heights, and beck.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=58501ed51a75" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>