Best tweets about Generative AI
Explore the best tweets about generative AI, including models, products, business applications, research, creative tools, risks, and real-world results.
Substantive generative AI research, products, applications, adoption, limitations, economics, and firsthand implementation experience.
Best Generative AI tweets
Ranked 01–50
@ostrisai ·
I trained this @ltx_model LTX 2.3 LoRA of George Costanza at home on my 5090 in about a day with AI Toolkit. I generated this 30 second video with @ComfyUI on my 5090 in 6 minutes. Open source is, always has been, and always will be, the future of generative AI. (SOUND ON)
Watch video
@akshay_pachaar ·
Anthropic's most viral feature is now open-source! Until now, Anthropic's Generative UI capabilities only existed inside its own products. @CopilotKit just shipped Open Generative UI, an open-source implementation of Claude Artifacts that works in any app. The agent generates HTML/SVG at runtime, and CopilotKit streams it token-by-token into a sandboxed iframe inside the app's chat. So the user can watch the UI assemble itself in real time, not after the full response is ready. The sandbox is fully isolated with no access to the parent app, the DOM, or user data. So if the agent hallucinates broken markup or unexpected JavaScript, nothing leaks outside the iframe. Under the hood, the agent does not select from pre-built components. Instead, it generates arbitrary visuals from scratch every time. The output is unconstrained by default, but you can shape it by defining prompt-based skills that teach the agent specific visual formats or guidelines. For instance, a skill prompt can guide the agent toward producing a Chart.js dashboard with proper axis labels and responsive sizing, or an interactive 3D model with rotation controls. The video below shows this in action, and the output quality you see actually comes from the skills layer. Open Generative UI runs on AG-UI, so it works out of the box with LangGraph, CrewAI, Mastra, Google ADK, AWS Strands, and more. It also ships with a standalone MCP server that plugs into Claude Code, Cursor, or any MCP-compatible client. And the entire stack is built on top of CopilotKit, the open-source frontend framework for agents and generative UI. 30k+ GitHub stars, with SDKs for React, Next.js, Angular, and Vue. I have shared the GitHub repo and a live playground in the replies!
Watch video
@RamVenkatSrikar ·
James Cameron on generative AI after SS Rajamouli shared his “terrifying thoughts” about how it could potentially influence filmmaking in the future. “I believe that generative AI is very dangerous if it eliminates that sacred process of working with and through actors to tell stories. Because I find that to be a feedback process that's very positive for me and for the film. And I would say to young filmmakers that are drawn to these inexpensive tools that allow them to get their thoughts, their imagination, into an image very quickly… I would say: STOP. Spend time with actors. Some of them don't even think about acting. I mean, there are young filmmakers coming up that they don't even know any actors. They think they can bypass that step. That is very dangerous… and I don't want to see those movies.”
@aiwithmayank ·
A WOMAN BUILT THE ENTIRE "HOW TO LEARN GENERATIVE AI" CURRICULUM INTO ONE GITHUB REPO AND GAVE IT AWAY FOR FREE it's called awesome-generative-ai-guide and it's quietly sitting at 26,000 stars. i went in to grab one thing. the list of free courses. that's it. then i kept scrolling and realized this isn't a links page. it's an actual school. > a full 10-week "Applied LLMs Mastery" course, week by week, foundations to deployment > 90+ free GenAI courses from Stanford, Princeton, ETH Zurich, Karpathy, all indexed in one place > 60 of the most common GenAI interview questions, written out with answers > 5-day roadmaps for LLM foundations, for RAG, for agents > jupyter notebooks you can actually run the woman who made it is Aishwarya Naresh Reganti. she's an applied scientist. this is not a dropshipper repackaging a youtube playlist. it's a researcher writing the syllabus she wished existed and handing it to strangers. there are even free courses on it with actual certification at the end. evals. agent-building. the stuff bootcamps charge four figures for. 5,500 people forked it. that's 5,500 copies of an entire AI education that can't be unpublished now. and the part that gets me is the math. people are paying for $2,000 "become an AI engineer" cohorts built on a worse version of what this repo gives away at zero cost with the receipts attached. 100% Open Source. MIT License. repo: https://t.co/TZ9hNi6KDF
@ChrisGPT ·
I WON'T PLAY VIDEO GAMES IF THEY HAVE GENERATIVE AI" 🗣️🗣️🗣️ Well, hate to break it to you.. but generative AI is now literally looking better than real games. Everyone crying about "AI slop" is completely missing what's actually happening in gaming right now. Nvidia just dropped DLSS 5 at GTC today, and yes, it is actually real-time generative neural rendering. Now I to was worried about hallucinating frames. But DLSS 5 takes the ground truth of the game (the 3D geometry and motion vectors) and uses a generative AI model to completely redraw the lighting, shadows, skin subsurface scattering, and materials. Akin to a Hollywood-tier VFX filter over every single frame as you play!
@goyalshaliniuk ·
Generative AI is not magic, it is math that creates. From generating art and code to simulating human creativity, every GenAI model works differently, but with one shared goal: creation from data. Here is a simple breakdown of the five major types of Generative AI models shaping today’s AI revolution 👇 1. Diffusion Models Learn by adding and removing noise from data to create realistic outputs. Used in image and art generation tools like DALL·E 3, Stable Diffusion, and Midjourney. 2. GANs (Generative Adversarial Networks) Use two neural networks - a generator and a discriminator, that compete to produce lifelike data. Power deepfake videos, face synthesis, and AI art. 3. Variational Autoencoders (VAEs) Compress data into a compact representation and decode it to generate new versions. Used in image reconstruction, anomaly detection, and creative design. 4. Autoregressive Models Predict the next word, note, or pixel based on previous ones in a sequence. Used in text generation, music composition, and time-series forecasting. 5. Transformers Use self-attention to understand relationships across sequences for highly contextual generation. Power modern AI systems like GPT, Claude, and Gemini for text, code, and image generation. Generative AI Is Redefining Creativity Each model type brings a new layer of intelligence, from reasoning to imagination. Learn how these models work, and you’ll understand the core of AI’s creative power.
Watch video
@williamkast_ ·
If you're not running AI animations yet, you should. We're producing them for all brands we work with now, and in most cases, they rank among the top 5 spending ads quickly. Here's how to make them: 1. Pick a winning angle or winning ad 2. Rewrite the scenes into AI generation prompts 3. Use Higgsfield or MaxFusion to generate 4. Edit 5. Done The reason they perform is not just that "it's AI". AI animations are just a way of conveying things. So are UGC, VSLs, statics, VO ads, etc. The core that makes it work is not the AI but the angle. But AI helps. If you see an animation, your brain doesn't see it as an ad. It sees it as entertainment! Retention improves a lot, you watch the whole ad, and act. If you haven't tested this yet, just do it for your best 5 ads. You'll thank me later.
Watch video
@TMTMoats ·
Wild that this interview with William de Gale (Lead PM, BlueBox Global Tech Fund) has only 200 views.🤯 He spells out an exceptionally clear philosophy: Invest in the profitable tech companies that make disruption possible (not the disruptors). Must watch. Here is my favorite section on where he sees mispriced valuations in Tech right now (specifically $ADBE & Big Software): "So in my view it's the big software platforms, the cloud companies, such as $ADBE in particular, which are perceived as being the biggest losers from AI because there's a fear that Agentic AI in particular will both reduce the number of seats that people need [...] and also partly as a result reduce pricing. [...] But personally, I don't think that's how it's going to work out. At the moment, there are almost no companies that have made money out of the use of generative AI. [...] And in my view, the use of AI will probably be monetized at scale first by these big software platforms that people are regarding as the biggest losers. But they're the easiest place to deploy AI profitably at scale rather than taking your existing complicated and very stressed IT system and plumbing some new exciting point product in. [...] Or you just tick the box on the dashboard of the platform you're already using—whether that's $CRM or $ADBE or $NOW or whatever. [...] So these companies have got the customers, they've got the application, they've got the data. I think it's easier for them to make money out of AI than anybody else."
@hila_chefer ·
New research from @bfl_ml 🥳 Meet Self-Flow: our self-supervised framework for image, audio, video & world models 🤖 https://t.co/AshY8IkSEe Do generative models really need DINO to learn strong representations? We propose teaching them directly via a joint framework instead 🧵
Watch video
@Similarweb ·
Global AI Tracker Update Generative AI Website Worldwide Traffic Share - Key Takeaways: → ChatGPT's decline stops, with share up 0.1%. → Gemini continues growing its share. → Grok surpasses DeepSeek for the first time. 🗓️ 12 Months Ago: ChatGPT: 86.6% Gemini: 5.3% Perplexity: 1.9% DeepSeek: 1.6% Claude: 1.5% Copilot: 1.3% 🗓️ 6 Months Ago: ChatGPT: 78.3% Gemini: 8.7% DeepSeek: 4.5% Grok: 2.6% Perplexity: 1.6% Claude: 1.5% Copilot: 1.2% 🗓️ 3 Months Ago: ChatGPT: 72.9% Gemini: 13.3% DeepSeek: 4.1% Perplexity: 2.4% Grok: 2.4% Claude: 2.0% Copilot: 1.2% 🗓️ 1 Month Ago: ChatGPT: 66.8% Gemini: 19.5% DeepSeek: 3.8% Grok: 3.0% Perplexity: 2.1% Claude: 2.0% Copilot: 1.2% 🗓️ Today (January 16): ChatGPT: 64.6% Gemini: 22.0% Grok: 3.5% DeepSeek: 3.3% Claude: 2.1% Perplexity: 1.9% Copilot: 1.1%
@GoogleResearch ·
Today we're sharing our research experiment, Vantage, using GenAI to assess "future-ready" skills like collaboration. A study with New York University found our AI dynamically steers simulated conversations & achieves scores comparable to human experts. More: https://t.co/hT5IZFYaNq
Watch video
@laszlogaal_ ·
2026 = The end of upscaling? One of generative AI's biggest drawbacks was the resolution, and I'm not talking about the output resolution as 1080p would be enough for most of the projects. I'm talking about the "internal" resolution of how these models work, that makes small faces distorted in your generations, the ones that will make a wide-shot impossible because the small details will morph and of course with some models text rendering is not possible. Upscaling with AI was always a strange case, local upscalers sometimes lacked the quality compared to the closed source competitors, but those can be pretty pricey and the results are sometimes way too "creative". So I was very, very interested when @Kling_ai announced native, not upscaled 4K generation but wanted to have a good method test it. Again, I never missed the 4K output resolution with genAI, but I missed the nuanced detail rendering of the engines and the lack of that made it impossible to create for example very wide shots with tiny details. So I fired up some image models to generate stills inspired by Morocco and see how Kling behaves, now in 4K. Some notable moments: 00:22 - Skin and eye texture, the movements, the little coins in the headdress 01:18 - The reflection in the water 01:39 - The reflection in the eye, and the scarf textures 02:10 - Check all the details on the rails 02:20 - Water particles and leaves 02:29 - The rendering of the rusty parts of the rail 02:50 - How the waves change the reflection in the foreground 02:55 - The cloud movement in the background
Watch video
@ComfyUI ·
Doodles — the entertainment brand behind illustrator Scott Martin's iconic pastel artwork — is launching Doodles AI, powered by PRISM 1.0, a generative model trained on their entire visual body of work. The engineering is by @thesystms , built on ComfyUI open-source infrastructure. The model that makes this work isn't just technical. It's philosophical: → Users don't just consume the brand — they co-create it → Every generation feeds back into the model via reinforcement learning → Community members become collaborators shaping the brand's visual DNA They're calling it "open story" — and it reframes how creative IP can work in the age of generative AI. For artists watching from the sidelines: the infrastructure is here. Read the full story here > https://t.co/GAqtmnE5tR
Watch video
@coreyhainesco ·
I built a skill that produces marketing videos using AI tools and programmatic frameworks — HeyGen, Remotion, Hyperframes, Runway, and more. You describe what you need and it recommends the right tool: AI avatars for talking heads, programmatic rendering for templates at scale, or generative AI for creative footage. Then it handles the full workflow. Video is the highest-converting content format but most teams don't have the production capacity. AI tools and programmatic frameworks change that equation entirely. It's called /video and it's part of Marketing Skills — a free, open source collection of 40 marketing skills for AI agents like Claude Code, Cursor, and Codex. npx skills add coreyhaines31/marketingskills
Watch video
@burkov ·
An ICLR Test of Time Award: "Auto-Encoding Variational Bayes" This is the paper that introduced the variational autoencoder. Before it, training generative models with continuous hidden variables ran into a wall: computing the exact probability a model assigns to an image requires integrating over every possible hidden representation that could have produced it, and for anything involving a neural network that integral has no closed form. Existing workarounds either needed slow per-datapoint sampling loops or imposed restrictive assumptions on the model. Kingma and Welling show that if you let a second neural network guess the hidden representation from the input, and you sample that representation in a particular way — drawing random noise from a fixed distribution and then transforming it with the network's output, rather than sampling from the network's output directly — the whole system becomes differentiable end to end and trainable with ordinary gradient descent. That one move, now called the reparameterization trick, is the piece worth seeing derived from scratch, because it shows up throughout modern generative modeling and in places you would not expect, including the noise schedules of diffusion models and several reinforcement learning methods. The paper is short, the math is followed step by step rather than hidden in appendices, and reading it gives you a clear view of how an idea from Bayesian statistics turned into a practical tool for training neural networks on images. A must-read. Published on @ChapterPal thanks to a kind permission from Prof Kingma. Read with a friendly AI tutor: https://t.co/WMvZFLT0qA PDF: https://t.co/RRH7mLouOr
@VaibhavSisinty ·
Most people in AI can't actually explain the difference between "Generative AI," "Agentic AI," and "AI Agents." They use all three like they mean the same thing. They don't. And once it clicks, you can't unsee it. Here's the cleanest way to think about it: Generative AI is the brain that creates. You ask, it produces. Text, images, code. Then it stops and waits. Brilliant, but passive. It answers, it doesn't act. This is GPT and DALL-E. Agentic AI is the brain that decides. It adds judgment on top, picking tools, calling APIs, looping through steps, correcting itself, all to reach a goal instead of just spitting out one answer. Generative AI gives you a response. Agentic AI works toward an outcome. AI Agents are the brain that acts in the real world. The full loop: fetch data, plan the steps, take actions, check results, update memory, adapt next time. The key word is autonomy. It touches live systems and learns from them. Think self-driving cars and robots, not just software returning text. The simplest way to lock it in: → Generative AI creates. → Agentic AI decides. → AI Agents do. And each one wraps the last. Every agent has a generative brain at its core, it just stacks decision-making and action on top. So next time someone pitches you an "AI agent," ask one thing: does it create, decide, or actually do? Most stop at the first.
@_vmlops ·
Google DeepMind’s AlphaEvolve feels like a glimpse of where AI is headed next An AI system that: • designs better algorithms • optimizes real-world infrastructure • improves chip design & data centers • helps solve open math problems • accelerates scientific discovery Some of the reported results are wild: → 10x lower error quantum circuit optimizations → 20% reduction in storage write amplification → improved power grid optimization from 14% to 88% feasible solutions → breakthroughs on long-standing math challenges with researchers like Terence Tao The important shift isn’t “AI writes code” It’s AI iteratively discovering better systems than humans previously found then deploying them into production infrastructure We’re moving from generative AI → evolutionary AI And that changes the game https://t.co/yCn7gj4BVP
Watch video
@aaliya_va ·
The biggest bottleneck in image restoration just got solved by AI made photos. Image restoration models keep failing on real photos. The bottleneck is training data. Real paired photos are expensive to capture. Synthetic ones miss messy degradations like haze or rain. A new paper reverses the pipeline. Instead of photographing clean targets, it generates them. Researchers fed real degraded images into generative models. Nano-Banana-2 beat eight other candidates tested. Its outputs became the ground truth itself. The result is a dataset called GGT-100K. It holds 103,707 paired images at 1024x1024 resolution. Every pair passed three quality gates: 1. Automatic perceptual metric filtering 2. VLM checks for content consistency 3. Manual inspection by human reviewers Retraining ten restoration architectures on it improved real-world generalization. Generative models gained the most. Everything is open on Hugging Face. Dataset, training code, and twenty pretrained checkpoints. What else could the generated ground truth replace? Link in comments. ↓
Watch video
@TheSixFiveMedia ·
Most enterprises have yet to solve the consistency problem with generative AI. "Just a slight change in the seed value or the prompt can dramatically change the output." — Varun Parmar, SVP and GM, Adobe GenStudio and Firefly Enterprise Varun Parmar told @PatrickMoorhead & @danielnewmanUV how @Adobe is addressing it: creative technologists build the workflow once, validate the output quality, then deploy it across the entire organization. Marketers in every geography get access to personalized campaigns they never had the budget or resources to run before.
Watch video
@eric_seufert ·
AI-generated ads perform roughly as well as human-created ads, unless consumers suspect they were generated by AI. Researchers from the Technical University of Munich, Columbia Business School, Harvard Business School, and elsewhere analyzed more than 16 billion ad impressions and 116 million clicks from Tabola. In a novel quasi-experiment, they compared the performance of AI-generated and human-generated "sibling ads": ads from the same campaigns, for the same products, and with the same launch dates. They find that CTR performance for AI-generated ads is statistically "indistinguishable" from human-created ads when controlling for campaign-level factors. But they also reached an equally important conclusion: ads that look AI-generated perform worse. The researchers identified a set of visual traits that consumers identify as "AI-like," and found that perceived artificiality reduced click-through rates even when the ads were not explicitly flagged as AI-generated. Ironically, the authors find that a not-insignificant number of human-generated ads were perceived as AI-generated for possessing those visual traits. This is an important result. The cost of creative production through generative AI is decreasing rapidly, especially for static images. But not all AI-generated creative performs equivalently. Teams need to consider an optimization for their AI-enabled workflow beyond mere volume, which is the perceived visual fidelity and provenance of the ads being generated. Link to paper below. Thanks to @garjoh_canuck for flagging this for me.
@thinking_slow ·
"does AI content work?" is entirely the wrong thing to ask. a much better question is "how is AI content materially different from 'normal' content?" usually when people publish "AI content" they are unwittingly engaging in a *different* strategy to traditional content marketing, and creating something *different* from traditional content. they create obvious hallmarks of AI use. for example: - publishing content much faster than usual, often on newer domains with little authority, no branded search demand, etc. - relying entirely on an AI model's internal knowledge, without sourcing information from a range of external sources - failing to include internal and external links, images, visual interest, first-person experiences - leaving obvious artefacts of AI use in the article, like obviously AI-generated imagery, obvious AI turns-of-phrase - AI writing patterns and watermarks that haven't been "humanised" by anchoring text generation in specific writing examples. some of these hallmarks make the content WORSE than normal (and hence contribute to poor performance), others are very DIFFERENT from normal (and make it easy to single out content as likely AI-generated) - both of which can contribute to that content not performing well. i obviously don't know the exact mechanisms at play when Google sinks an AI-generated blog after 3-months, but i DO know that many of these aforementioned signals are very obvious to Google: indexing requests, branded search demand, AI content detection (even if only directionally accurate), user engagement signals. we use generative AI a lot at Ahrefs, and i'm happy to do this because we do not compromise on our editorial standards. our AI process mirrors our human editorial process, step for step; it is better and more detailed than the human equivalent in many areas, because LLMs are more tireless researchers, more thorough adherents to brand voice. we are substituting one tool for another, one method of construction for another, but the end product is the same. we have even now published AI-generated content that is subjectively BETTER than our previously human-made content, because AI removed the data, design and updating constraints that previously limited our team. i am excited for the new experiences we can build for readers. you need to determine your own risk tolerance, but in my opinion, using AI to create content is not a problem - but using AI to create something that is WORSE or DIFFERENT to "normal" content marketing is. "creating bad content" or "scaling content too soon" is where problems emerge, and many people do this unwittingly when they use AI. if you know that you are compromising on your content through your use of AI, or trying to scale content on a site that barely exists in Google's consciousness, you should probably feel a bit nervous. if you want to win, change your framing and use AI to make content that is cooler and better than you were able to do before ✌
@burkov ·
3D generation is one of the most impactful uses of generative AI because it allows editing the generated objects and then produce their physical copies. This recent paper from Microsoft explain the SOTA approach to 3D generation. Learn from the paper with an AI tutor and quizzes for better retention: https://t.co/U13jYaNzGB PDF: https://t.co/HihW6V6vPh
@xelebofficial ·
We are now in the era of Agentic AI, the third major inflection point in the history of AI. Inflection 1: Generative AI Models that create. Given a prompt, they produce text, images, code. They respond, but don't act. Inflection 2: Reasoning AI Models that think. Before responding, they break down problems and work through logic step by step. Slower, but fundamentally more capable. Inflection 3: Agentic AI Models that execute. As Jensen Huang, CEO of NVIDIA, put it, we are entering the era of agentic systems, where AI no longer just responds, but begins to act. It receives a goal, plans the steps, uses tools, checks results, and keeps going without waiting for human input at every turn. Each shift changed not just what AI can do, but how it behaves. And like every major computing shift before it, the real impact won't come from the models themselves, it will come from the infrastructure and applications built on top of them.
@MITSloan ·
Employees using generative AI fall into one of three categories: cyborgs, who collaborate closely with artificial intelligence; centaurs, who engage with AI in a more constrained way; and self-automators, who offload a task more fully to AI. https://t.co/OkU1nJhGLf
@sabir_huss50540 ·
🚨BREAKING: Microsoft quietly built the best free Generative AI course on the internet and put it on GitHub. It's called Generative AI for Beginners. - 21 lessons, from what an LLM is to building agents and RAG - "Learn" lessons for concepts, "Build" lessons with real code - Every code sample in both Python and TypeScript - Runs on Azure OpenAI, GitHub Models, or the OpenAI API - Translated into 50+ languages, auto-updated 112k stars. MIT License. The people shipping AI apps aren't smarter than you. They just started here.
@InduTripat82427 ·
Most “GenAI projects” are just messy scripts glued together. This is what an enterprise-grade AI project structure actually looks like: • Config-driven model switching (GPT / Claude / local) • Proper RAG pipeline (chunk → embed → index → retrieve → rerank → generate) • LLM abstraction layer (swap providers in one line) • Centralized prompt templates (no prompt chaos) • Vector DB + embeddings separation • Processing pipeline (clean → chunk → preprocess) • Inference engine isolated from logic • Scripts for indexing, testing, and rebuilding embeddings • Production-ready logging + environment configs Result: You can scale from side project → production without rewriting everything. Most people skip structure. That’s why their AI apps break at 1k users. Steal this layout before your next GenAI build. Bookmark this for your next AI project. Retweet to help someone stop shipping spaghetti AI.
@MITSloan ·
There are two types of generative AI risk: embedded risks, which are inherent to the technology, and enacted risks, which are related to how it is deployed by organizations. https://t.co/xgmPAyWB9r
@AISUEDE ·
Deezer's head of research explained how their algorithm detects AI generated tracks. The system identifies artifacts left by generative models. Checkerboard patterns in the frequency domain. Spectral peaks at specific frequencies. Small, consistent spikes that don't appear in human recordings. 99.8% accuracy under controlled conditions. Accuracy drops when audio is pitch-shifted or compressed. Detection also requires constant retraining as models evolve. Detection is reactive by design. It classifies what's already in circulation. Authentication works differently. It doesn't hunt for what's missing, it records what was present at the moment of creation.
@MattNiessner ·
The productivity paradox of prompt-based GenAI: producing *any* compelling result is easy, but getting exactly what you want is hard. This creates a deceptive productivity curve. In code generation, an agent can spin up a boilerplate architecture in seconds — which is impressive — but iterative editing can still take hours. We’ve all seen Cursor agents fix a local bug by refactoring the entire codebase. The issue? The abstraction layer. Natural language is optimized for ease of use, but its inherent ambiguity makes it produce non-determinist results. A major challenge will be to find better abstractions.
@eric_seufert ·
I found it noteworthy that Netflix saw fit to highlight the use of generative AI in content production in its earnings release, even naming specific examples. Contrast this with gaming, where audience hostility to the use of GenAI has been so virulent that studios downplay their use of such tools. Netflix clearly feels no compunction to do the same.
@aleyda ·
🤖 Let’s talk about good AI content use cases: Although AI-generated content is used in many scenarios, in some cases it doesn’t make sense and can do more harm than good (eg. writing guides on autopilot, without real human experience or expertise). There are, however, scenarios in which AI-generated content actually makes sense, is useful, and improves the user experience, especially when leveraging the nature and scale of the information. For example, summarizing product reviews or highlighting key themes across hundreds or thousands of reviews in PDPs. What’s important is to use your own content as training input, have a quality-assurance workflow to ensure accuracy and usefulness, and clearly disclose when a summary has been generated by AI. What other scenarios have you found where using AI for content actually helps?
@gaganghotra_ ·
🆕 Google will allow websites to exclude themselves from AI search results (still can show up in usual search results) #SEO 👀 Google said "Today, we're beginning to test a new control that lets website owners manage how their links and content appear in generative AI Search features. With this new toggle in Search Console, website owners can decide if they want their site to appear in and help ground responses in our generative AI Search features (like AI Overviews, AI Mode or AI Overviews in Discover). Sites that opt out will not receive traffic or impressions from our generative AI features. This control will not be used as a ranking signal for search results outside of these generative AI Search features. This work builds on our long history of designing tools, like snippet controls and Google-Extended, that give websites more choice." https://t.co/lDKKH4xLkE
@tomfgoodwin ·
I used generative AI for something today, It took me three minutes and saved me about 20 awful hours I also used generative AI to do some troubleshooting with my email, And I wasted about 4 hours It's still not that obvious what it's great for and what it's disastrous for
@pascal_bornet ·
This may be the most honest picture of generative AI. When AI is trained on flawed data, it does not just inherit the problem. It becomes a very efficient amplifier of it. That is the part too many people still underestimate. → bad data in → scalable inaccuracy out To me, this is one of the biggest blind spots in AI. People obsess over model quality. Far fewer ask whether the source material deserves that much amplification in the first place. Because scaling knowledge with AI also means scaling responsibility in data sourcing. Just saying. What do you think is the bigger risk right now: weak models, or bad data being amplified at machine speed? #AI #GenerativeAI #DataQuality #MachineLearning #Innovation #Technology #DigitalTrust #FutureOfWork Photo credits: Ralph
@alvinfoo ·
The latest Stanford AI Index Report 2026 highlights a shift from generative AI evangelism to evaluation as 88% of organizations integrate the technology, creating a "jagged frontier" of mixed capabilities. Key breakthroughs include advanced scientific problem-solving, a critical reliance on hardware infrastructure, and the maturation of AI agents for multi-step tasks over mere chatbots. The industry is pivoting toward efficiency and trust, with major advancements from Anthropic, OpenAI, and Meta defining the year. Download the Full Report (PDF): https://t.co/OFPm0eYH70
@MIT_CSAIL ·
GenAI can stylize 3D models before they're 3D printed into everyday items, but often weaken its mechanical integrity. “MechStyle” simulates how genAI's edits impact the model's structure & ensures designs for things like glasses & hooks sustain daily use: https://t.co/hoxs3KkkJ7
@HeyAnjula ·
MASTER GEN AI ENGINEERING GENERATIVE AI ENGINEERING MASTER TREE │ ├── 1. Foundations │ ├── What is Generative AI │ ├── AI vs ML vs DL vs GenAI │ ├── Types of Generative Models │ │ ├── Text (LLMs) │ │ ├── Image (Diffusion Models) │ │ ├── Audio / Video Models │ ├── Tokens & Context Window │ └── Training vs Inference │ ├── 2. Large Language Models (LLMs) │ ├── What are LLMs │ ├── Transformer Architecture │ ├── Attention Mechanism │ ├── Pretraining (Next Token Prediction) │ ├── Fine-tuning │ └── Popular Models │ ├── GPT │ ├── Claude │ ├── LLaMA │ └── Mistral │ ├── 3. Prompt Engineering │ ├── Zero-shot Prompting │ ├── Few-shot Prompting │ ├── Chain-of-Thought │ ├── Role-based Prompts │ ├── Prompt Templates │ └── Prompt Optimization │ ├── 4. Embeddings │ ├── What are Embeddings │ ├── Vector Representation │ ├── Semantic Similarity │ ├── Cosine Similarity │ └── Use Cases (Search, Clustering) │ ├── 5. Vector Databases │ ├── What is a Vector DB │ ├── Indexing (FAISS, HNSW) │ ├── Similarity Search │ ├── Metadata Filtering │ └── Popular Tools │ ├── Pinecone │ ├── Weaviate │ └── Chroma │ ├── 6. Retrieval-Augmented Generation (RAG) │ ├── What is RAG │ ├── Data Ingestion │ ├── Chunking Strategies │ ├── Embedding Storage │ ├── Retrieval Techniques │ ├── Context Injection │ └── RAG vs Fine-tuning │ ├── 7. AI Agents │ ├── What are AI Agents │ ├── Tool Calling │ ├── Memory (Short / Long Term) │ ├── Planning & Reasoning │ ├── Multi-agent Systems │ └── Frameworks │ ├── LangChain │ ├── LlamaIndex │ └── AutoGen │ ├── 8. Fine-tuning & Custom Models │ ├── When to Fine-tune │ ├── Instruction Tuning │ ├── LoRA / PEFT │ ├── Dataset Preparation │ └── Evaluation │ ├── 9. Evaluation & Guardrails │ ├── Model Evaluation Metrics │ ├── Hallucination Detection │ ├── Bias & Fairness │ ├── Safety Filters │ └── Prompt Injection Protection │ ├── 10. Multimodal AI │ ├── Text + Image Models │ ├── Vision Models │ ├── Speech Models │ └── Video Generation │ ├── 11. Model Deployment │ ├── APIs (OpenAI, etc.) │ ├── Backend Integration │ ├── Streaming Responses │ ├── Latency Optimization │ └── Cost Optimization │ ├── 12. GenAI Architecture │ ├── End-to-End Pipeline │ ├── RAG Architecture │ ├── Agent-based Systems │ ├── Caching Strategies │ └── Scalability Design │ ├── 13. MLOps for GenAI │ ├── Model Versioning │ ├── Monitoring │ ├── Logging Prompts & Outputs │ ├── A/B Testing │ └── Continuous Improvement │ ├── 14. Real-World Applications │ ├── Chatbots (Customer Support) │ ├── AI Assistants │ ├── Code Generation │ ├── Document Q&A Systems │ ├── Content Generation │ └── AI Search Engines │ └── 15. Career Path ├── Prompt Engineer ├── GenAI Engineer ├── AI Product Engineer ├── ML Engineer (LLM Focus) └── AI Researcher
@tomfgoodwin ·
Generative AI in practical terms doesn't work like people think. It's much closer to "Search" than "Thinking" Ask it "What airport serves the most continents direct" and I'm sure people think it's going to go one by one over likely airports and establishing systematically the answer. In this case, not all, It's not, it's merely finding in a corpus of data a lot of instances of people writing about this. This isn't a small thing, it means it knows nothing, understands nothing, it means it's usefulness is somewhat limited by what's been covered before* Ask it the longest non stop flight and it will "generate an answer", but it's really parroting back commonly covered terrain. Ask it which flight travels over the most water, or which flight goes over the most time zones or which flight has the greatest altitude difference, or anything few people asked before and it won't think, it will parrot back something similar to often covered material. It does indeed have the capacity "to think", you could ask it to break it into a process, tell it where to look, tell it how to get to an original answer, And it will do this FAIRLY quickly. And perhaps FAIRLY accurately, but it will burn a lot of tokens, and it's not the same magical feeling. Thats the hard stuff about this, What it's good at, how it works, and what it's terrible at are not necessarily instinctive. So when people are rude about it, they are often wrong. And when people are delighted in it, they are also often wrong.
@pascal_bornet ·
𝗪𝗮𝗻𝘁 𝘁𝗼 𝗵𝗶𝘁 𝗮 “𝗵𝗼𝘁” 𝗔𝗜 𝗽𝗹𝗮𝘆 𝗿𝗶𝗴𝗵𝘁 𝗻𝗼𝘄? Call it 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 😎 That’s it. I’ve been noticing a pattern. Almost every founder I speak with is building “AI agents.” Not because they all discovered the same breakthrough. Because the narrative is already winning. Yes, the shift is real. Even Gartner expects a meaningful share of enterprise interactions to move in this direction soon. But here’s the uncomfortable part. Most “agents” today: ▪️ Call a few APIs ▪️ Chain some prompts ▪️ Work on the happy path ▪️ Break when things get real We describe them as if they “reason,” “decide,” and “act.” What stands out to me is this: 𝗪𝗲’𝗿𝗲 𝘀𝗰𝗮𝗹𝗶𝗻𝗴 𝗲𝘅𝗽𝗲𝗰𝘁𝗮𝘁𝗶𝗼𝗻𝘀 𝗳𝗮𝘀𝘁𝗲𝗿 𝘁𝗵𝗮𝗻 𝗰𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝘆. Most agents look impressive. Few deliver consistently. Because customers don’t care about the label. They care if it works. And when it truly works… 𝗡𝗼 𝗼𝗻𝗲 𝗰𝗮𝗹𝗹𝘀 𝗶𝘁 𝗔𝗜 𝗮𝗻𝘆𝗺𝗼𝗿𝗲. 𝗦𝗼 𝗵𝗲𝗿𝗲’𝘀 𝗺𝘆 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻: Are you building something genuinely autonomous… or something that just sounds like it is? #ai #genai #agents #startups #product #futureofwork
@glenngabe ·
Banned for a year if you submit AI slop -> ArXiv, the repository of preprint academic research, says it will ban authors for a year if their papers have “incontrovertible evidence” of AI-generated work “If generative AI tools generate inappropriate language, plagiarized content, biased content, errors, mistakes, incorrect references, or misleading content, and that output is included in scientific works, it is the responsibility of the author(s)." https://t.co/J9vD36vgik
@cleartechtoday ·
📌 Did you know that @Telstra the Australian telecommunications giant utilized generative AI to power a service allowing children to place free, interactive calls to Santa Claus, showcasing the tech's use for brand-building and community engagement? Telstra's generative AI "Calls to Santa" service, implemented with partners https://t.co/erw7xZNnl7 and OMD was built using @googlecloud technology stack, specifically leveraging the DialogFlows CX conversational AI platform and the PaLM 2 (Pathways Language Model 2) large language model (LLM). A custom Text-to-Speech voice model created a "jolly" Santa voice. Over 500 audio lines were recorded to capture the required nuances. #GenerativeAI
Watch video
@darshal_ ·
Most AI companies are still trying to generate better videos. The more interesting challenge is generating worlds. Not worlds that look impressive for 20 seconds. Worlds that stay coherent, respond to every action, and keep unfolding as you explore them. That's why LingBot-World 2.0 stands out. The engineering isn't just about visual quality. It's about solving one of the hardest problems in generative AI: keeping an interactive world stable over long horizons without falling apart. And the fact that Robbyant decided to open-source both the model and the agentic harness makes this release even more interesting. It feels less like another AI demo and more like infrastructure for what's coming next. @robbyant_brain 💻 GitHub: https://t.co/bQBWuHXBUM 🎮 Try it: https://t.co/VJy43faBGP 🤗 Weights: https://t.co/dEmViDnNSO
Watch video
@sachinrekhi ·
Question: Why aren't we seeing massive revenue acceleration amongst companies fully adopting AI for productivity? This is probably one of the most fascinating paradoxes with AI right now. While the foundation model vendors themselves have had astronomical revenue growth, the story of the companies leveraging them is not quite as straightforward. Intuitively one might assume that specific companies that go all in on AI tools should see a significant increase in productivity which should translate into significantly higher revenue compared to competitors. And that might be the case when a single industry player adopts AI. But what we are actually seeing is far more nuanced than that. We are instead seeing clusters of AI-pilled industries emerge where many of the top players in that industry are adopting these generative AI capabilities all at once. Did your team accelerate product development using coding agents? Well so did your top 3 competitors, resulting in your feature release cadence not looking all that different from theirs. Similarly, customer expectations are incredibly high thanks to broadly adopted consumer and enterprise tools like ChatGPT and Claude. Did you add a chatbot to your product? That's cute, but how do I get it to work with all the tools I've already integrated with Claude? Or take Gemini for example. I find myself complaining every day of why I can't do something in Google Docs/Sheets/Slides and Gemini because Claude has set the bar so high for what I expect from these tools. Just like during the mobile revolution, mobile was initially a differentiator but then quickly became table stakes. We are seeing that same cycle transpire again with AI.
@milan_milanovic ·
There is a lot of talk about whether AI should be used for writing or not Most people here are still thinking about this in black/white form: 1. You give a prompt to AI 2. AI generates the text 3. You publish it This is, of course, bad, as anyone can do it, and you will get mediocre text. Yet, things are more complex than that. First, the writing process involves multiple steps: 1. Researching the topic 2. Making the outline of the article 3. Writing details of each chapter 4. Final tweaks Now, I assume everyone here is talking about step 3 when talking about "AI-generated text", and this makes sense; yet, making it useful for other steps, I don't see a problem Also, for us non-native English speakers, even with good prompting, it can write decent details; if you're not a top-notch writer, it will do better than you I also find that AI can be a good sparring partner in writing: you write something, ask for feedback, then revise, etc. For me, writing is mostly a way of thinking, so the more I write, the better I think. And AI comes as a good evaluator of some ideas.
@thinking_slow ·
guess: to what does this criticism refer? "The average quality...is pretty good, but on any given day, [the accuracy of] any given entry might be up or down." "You don’t know who it’s written by, you don’t know what qualifications they have, and you don’t know whether it’s been reviewed by anyone who knows what they’re talking about." "Forever open to the uninformed… meddler." "The main problem is the lack of authority." "[It lacks a] habit or tradition of respect for expertise." "The premise...is that continuous improvement will lead to perfection – that premise is completely unproven." "Hyperlinks, bullet points and cut-and-paste press releases do not an encyclopedia entry make." "A flawed and irresponsible research tool." --- if you answered "generative AI", you would be wrong. this is a list of 20-year-old criticisms shared after the launch of Wikipedia. for all our concerns about accuracy and authority and legitimacy, for all the furore around regulation and fairness and compensation for creators, i think generative AI is here to stay. generative AI is a natural continuation of a process of information commodification that started a long time ago. crucially, it offers a better experience for most people. information is free and convenient and personalised to the whims of each user. that is hard to resist. i think Wikipedia is a good parallel for understanding generative AI (not least because it plays such a crucial role in training LLMs). it was untrusted and error-prone and far from perfect, but the experience it offered was such an obvious upgrade from the status quo that it weathered those complaints and became a cornerstone of the internet. and as we saw in the wake of Wikipedia's release, there was still a place for formal research, for professional accreditations and authority sources, for personal experiences and stories, for content that went above and beyond the new baseline created by Wikipedia. the same will be true of generative AI.
@etnshow ·
F1 World Champion Nico Rosberg (@NicoRosberg) shares how generative AI will completely revolutionise F1: "There is a huge opportunity to use generative AI for material science because in F1 everything is about materials—weight and stiffness are critical for performance." "The most important thing is getting to the minimum weight, because 10 kilos over is three-tenths of a second per lap; if you are over, you have no chance of winning." "If you manage to use these advancements to move 5kg from a high area down to under the car with titanium ballast, that is huge performance."
Watch video
@JustinThomasAI ·
According to @OECDEduSkills's Digital Education Outlook, GenAI tools can enhance students’ performance on tasks, but this does not necessarily lead to learning gains. Students were 48% more successful on practice tasks when using GenAI, yet their scores dropped by 17% on unassisted exams once the AI “crutch” was removed. Performance is not the same as learning. To show sustained improvements in learning, OECD says to use intentional pedagogical purpose, the explicitly defined learning goal or instructional reason behind using a tool, activity, or method. In other words, teachers or systems must decide what students should actually learn and then use GenAI to force engagement with those goals rather than to bypass cognitive effort. Examples of purposeful use may include: - Structuring GenAI as a tutor that asks questions, nudges strategies, and supports step‑by‑step reasoning without just giving the answer. - Embedding GenAI into collaborative learning tasks so students must critique, compare, or build on AI output, which can strengthen knowledge and argumentation skills. - Designing GenAI‑powered Intelligent Tutoring Systems as “digital pedagogical agents” that adapt to learners and promote critical thinking, creativity, and collaboration rather than replacing cognitive effort.
@ATechAjay ·
AI models are entering a strange loop. 🤯 The internet is getting flooded with AI-generated content. Now imagine future AI models being trained on content created by previous AI models... over and over again. Researchers call this Model Collapse, where models can gradually lose diversity, creativity, and connection to real-world knowledge. That's why AI companies are racing to secure high-quality human-generated data from places like Reddit, Stack Overflow, forums, books, and expert communities. At the same time, they're also creating synthetic data (AI-generated training data) to train newer/other models. So the future of AI is becoming: 👨 Human data -> trains AI 🤖 AI generates synthetic data 🤖 New AI trains on synthetic data 🔁 Repeat The challenge isn't generating more data anymore. It's making sure the data stays useful, diverse, and grounded in reality. What a fascinating loop we're witnessing. 😄
@ai_for_success ·
The core issue with many AI game projects right now is that strong video generation does not automatically translate into a real game. A game needs interaction, state tracking, persistence, and coherent outcomes. That gap between generating scenes and building a playable system is where most efforts break down. At GDC and NVIDIA GTC, LinearGame’s @YorollAI presented something that stands out from most of what we are seeing in AI gaming today. What Yoroll is exploring is an approach that separates visual generation from game logic. Video models handle the visuals, while a structured system manages interaction, state, and progression. That distinction is important because it moves closer to something that behaves like a game rather than just a sequence of generated outputs. They showcased multiple experiences built on the same underlying system, each with very different themes and mechanics. That consistency across different types of content suggests a more generalizable structure rather than a one-off demo. The timing is also notable. Video quality has improved significantly, costs are dropping, and latency is approaching a point where interaction starts to feel viable. These pieces were not quite there before, but they are starting to come together. More broadly, this points to a shift in how games could be created. Traditional development relies heavily on complex pipelines and large teams. If systems like this evolve, it could open up new ways for creators to build interactive experiences, especially those coming from storytelling or narrative backgrounds. This is still early, and there are open questions around scalability, usability, and how far these systems can go beyond controlled experiences. But it does reflect a move from isolated demos toward more structured approaches to AI driven interaction. The space has had no shortage of impressive visuals. What has been missing is a clearer product structure. This feels like a step in that direction. #AIGaming #GenerativeAI
Watch video
@The_InnovationJ ·
New in The Innovation Drug Discovery! The evolution of computation-driven paradigms in targeted peptide drug design: From predictive modeling to generative AI and clinical translation. Hu et al. explore how integrating generative AI, like diffusion models, with rigorous physics-based validation prevents "structural hallucinations" and accelerates the translation of in silico peptide blueprints into viable clinical therapeutics. https://t.co/aAiNsb9n63 #DrugDiscovery #GenerativeAI #computationaldesign
Best Generative AI tweets
Xholic studies what works in your niche, drafts posts in your voice and schedules them for the hours your audience is online.
$0 today · Cancel anytime
Browse all tweet collectionsKeep exploring