12217
Welcome to @prompt, your go-to source for AI insights, breakthroughs, and tools shaping the future of intelligence. Contact: @LightEarendil
⚡️ Google is renting $920M/month of compute from SpaceX
110,000 Nvidia GPUs, CPUs, memory, the whole stack, Oct 2026 through June 2029.
Google is already the world's largest single owner of AI compute, but demand still caught them short. Anthropic just signed a similar deal at $1.25B/month for the Colossus 1 cluster.
xAI built those data centers to run Grok. Now it's renting them to its biggest rivals. Honestly, not a bad pivot.
🤖 Canada wants its own supercomputer. No AWS required.
Canada's "AI for All" strategy calls for a public AI supercomputer and investment in sovereign, Canadian-owned compute and cloud infrastructure. That's the government treating compute as national infrastructure, not a vendor contract.
Canada's current AI data centre and cloud setup is "largely foreign-owned," and sovereign compute capacity is "nascent" with "significant investment" needed to cut reliance on foreign providers. So yeah, the stakes are real.
Carney launched the plan Thursday. Big ambitions, thin details so far.
⚡️ Beijing's Booster Robotics is quietly becoming the platform that runs robot soccer
Each team runs fully autonomously, driven by AI, with zero human intervention. The slapstick hides real tech.
Booster's T1 robots trot, tackle, shoot, and defend entirely on their own. Teams bolt their own AI models on top of Booster's hardware, making it the Android of humanoid robotics.
China isn't just building robots. It's building the platform everyone else runs on.
⚡️ Bigger quant beats aggressive compression on Qwen3.6 35B
ByteShape benchmarked NTP vs MTP GGUFs across 4090, 5090, 4080, 5060 Ti, plus Intel, Ryzen, and Raspberry Pi 5.
MTP gives GPUs a 20-40% gen-speed boost. CPUs? It just adds overhead. Stick to NTP on CPU.
Bottom line: lower bpw is not automatically better. Largest quant that fits often won.
🇨🇳⚡️ Ex-Samsung chip chief says China will pop the DDR5 price bubble by H2 2027
Kye-hyun Kyung, former head of Samsung's DS Group, says Chinese firms investing aggressively in memory production could flood supply and push prices down in the second half of next year.
CXMT leads the charge. Kyung cites data projecting global production capacity at 6 million wafers/month by H2 2027.
Catch: he warns that if Big Tech cuts AI budgets, the industry could flip straight into severe oversupply.
Source
Avatar V just achieved a Face Similarity score of 0.840 and a Sync-C score of 8.97, leading its category. For reference, the highest possible “ground truth” is measured at 0.861. With just one more update, HeyGen’s avatars could become almost impossible to distinguish from real video.
We’re now introducing Avatar V, which finally solves the challenge of character consistency. Just record yourself for 15 seconds and, from that moment on, the system can generate videos in any style while perfectly capturing your likeness, expressions, and quirks.
This model is the new standard for AI avatars—a bold statement, but the data backs it up. 👀
Anthropic just submitted patches to FFmpeg, reportedly using its latest model, Claude Mythos. The code quality is indistinguishable from that of top-tier engineers.
If these were truly discovered and written by AI, things are about to get really strange, very fast.
The latest Grok Imagine update is truly impressive. It's fascinating to see how quickly this tool is evolving and pushing the boundaries of what we thought was possible. Every new release seems to bring unexpected features and improvements, making it an exciting time to follow its progress. 🚀
Читать полностью…
OpenAI researchers have made significant progress by solving five more Erdős problems with their internal AI model. This achievement highlights how artificial intelligence is advancing in serious mathematical reasoning. 🔍
Personally, I found the solution to Erdős Problem 1091 especially fascinating. The question considers whether a graph with chromatic number 4, but all its smaller subgraphs having chromatic number at most 3, must contain an odd cycle rich in diagonals. OpenAI's model delivered a clear counterexample, shedding light on the issue.
For those interested, it’s a great challenge to try and reconstruct the proof from Figure 5—it's an enjoyable exercise for math enthusiasts.
China is pushing AI into classrooms as a major nationwide experiment, aiming to use the tech for things like grading, lesson design, supporting rural educators, and monitoring student progress.
Local schools and counties are now free to choose commercial AI systems and integrate them into daily education for four key areas: assisting teachers, streaming better lessons to remote areas, tracking how students do, and keeping an eye on student behavior or mental health.
Overall, this rollout isn’t heavily funded—Beijing sets the direction and goals, tech companies provide solutions, and it’s up to local officials to innovate with their existing resources.
A new version of GlassWorm malware is spreading through npm, PyPI, GitHub and developer tools, stealing browser data, crypto wallets and even installing a remote access trojan. It hides its command servers inside Solana blockchain transactions, making it harder to detect or shut down. 🤢
Once inside, it logs keystrokes, steals cookies, captures screenshots and even installs a fake Google Docs Offline extension to monitor sessions and steal crypto credentials. It can also show fake Ledger or Trezor recovery screens to capture seed phrases.
This is especially dangerous for developers and crypto users, since the infection often comes from poisoned packages or compromised dependencies.
RSA just hit 59.31% on ARC-AGI-2 and the method is dumb simple
Generate reasoning chains in parallel. Split into random subsets. Ask the model to merge each subset into one better chain. Repeat.
No code scaffolding. No multi-agent systems. Just the model arguing with itself.
With Gemini 3 Flash it beats Gemini DeepThink at 1/10th the cost 💸. Nearly matches Poetiq with a fraction of the complexity.
LLMs are inconsistent but weirdly good at picking the best parts from multiple attempts. RSA just turns that into a loop.
github.com/HyperPotatoNeo/RSA-ARC
🛌🤖 Imagine if a single night of sleep could reveal your future health.
That’s what SleepFM is claiming: a model trained on 585,000 hours of polysomnography from 65,000 people. The paper says it can predict 130 diseases, with some big numbers: all-cause mortality (0.84), dementia (0.85), heart attack (0.81), stroke (0.78) (metrics like C-Index/AUROC).
It’s still not as simple as wearing a smartwatch. This is lab-grade data: brain, heart, and breathing signals captured overnight.
🩺 OpenAI is quietly moving ChatGPT closer to your real life
With ChatGPT Health, users can link medical records and wellness apps so answers are grounded in their own health data. It’s not a doctor and it won’t diagnose you, but it aims to help you understand your information better and ask smarter questions.
Health chats live in a separate space, stay private, and aren’t used to train models. ChatGPT wants to be a personal assistant that follows you beyond search and work, while trying to earn trust where it matters most.
🎧 OpenAI is betting that the future of tech won’t be on a screen, but in your ears.
They’re rebuilding their audio AI and working on an “audio first” device that talks and reacts like a real conversation partner. At the same time, Meta, Google, Tesla and a wave of startups are all pushing toward a world where we don’t tap or swipe. We just talk.
With Jony Ive shaping the hardware, the goal is clear: less screen time, more human interaction.
The future won’t be seen. It will be heard.
⚡️ Anthropic's Mythos hits 52x on training code. Humans top out at 4x.
Anthropic runs the same test every release: optimize training code. Skilled human, 4-8 hours, gets 4x. Claude Opus 4 averaged 3x in May 2025. Mythos Preview hit 52x by April 2026.
Anthropic frames it as a possible path to recursive self-improvement, but admits Claude still hasn't shown the research taste to pick which problems matter.
Narrow task, lab conditions, unreleased model. Still a brutal trendline.
🤖 The "Cannes AI film" didn't actually screen at Cannes
Higgsfield AI's Hell Grind, a 95-minute generative sci-fi feature built by 15 people in 14 days for under $500k, was pitched as a Cannes debut. The festival itself went out of its way to say otherwise.
Higgsfield held screenings at a private industry event and a commercial cinema in the town of Cannes, not a Festival de Cannes venue. The CEO still called it a Cannes premiere on LinkedIn. So.
80% of that $500k went straight into AI compute costs, which is the one genuinely interesting data point here.
🔐 1Password keeps your secrets out of Codex's brain entirely
1Password released an MCP server for OpenAI Codex that pulls credentials from vaults at runtime, then discards them. Mounted, used, gone.
Credentials never appear in code, terminals, or the model's context window. That last part matters most.
Just-in-time credentials are the new seatbelt for AI coding agents.
⚡️ AMD's local AI box hits $3,999. It was $1,700 last summer.
AMD's Ryzen AI Halo workstation opens pre-orders next month at $3,999 with 128GB LPDDR5X. Same chip, same memory, same story. Memory prices exploding since launch are now fully baked into the consumer price.
For $4k you're one click away from an NVIDIA DGX Spark. Tough sell.
Source
⚖️🤖 Musk sues OpenAI for "stealing a charity." Jury: you waited too long.
The AP reports that a federal jury in Oakland unanimously tossed Elon Musk's lawsuit against Sam Altman and OpenAI. The jury found Musk had three years to sue and simply did not file on time. Judge Gonzalez Rogers immediately adopted the verdict and dismissed all claims.
90 minutes of deliberation. Three weeks of trial, private emails, Brockman's diaries, Satya Nadella on the stand. Didn't matter. The merits never got touched.
By ruling on timing, the jury dodged the actual question: whether donors to nonprofit AI labs have standing to challenge for-profit conversions under charitable trust law. That question is still live. Musk's team says they're appealing.
Bottom line: OpenAI heads toward its trillion-dollar IPO with the biggest legal threat dead on arrival, and accountability for the nonprofit-to-profit pivot remains untested.
OpenAI is preparing to introduce a new model with enhanced cybersecurity features, but like Anthropic’s Mythos Preview, it will only be available to a select group of trusted organizations for now. This cautious, phased release reflects a growing trend: as AI models get better at independent hacking tasks, companies are starting to handle launches much like responsible vulnerability disclosures, emphasizing control and oversight over speed. 🔒
Читать полностью…
Claude Mythos Preview is now in private testing at $25/$125 per million tokens, but access is strictly limited to approved partners.
Because of its advanced cybersecurity abilities, Mythos isn’t available to the public. Instead, a select group—including major tech companies—are testing it through a $100M+ initiative called Project Glasswing.
What sets Mythos apart? It goes beyond typical bug-detection—this model autonomously chains multiple vulnerabilities to reach root or remote code execution, bypassing defenses that usually require rare human skills. During internal tests, it uncovered zero-days across all major systems, patched a 27-year-old OpenBSD flaw, and exploited old FreeBSD and Linux bugs—all on hardened systems. Even staff without security experience could request powerful exploits.
Nano-Banana Pro makes turning complex Mermaid diagrams into clear images a breeze. Its ability to handle intricate syntax and quickly produce visuals is impressive. This tool is a real asset for anyone who works with diagrams regularly. 💡
Читать полностью…
Morphic introduces 72 unique workflows to their platform, making creativity more accessible for everyone.
Ever had a great idea but got stuck on how to prompt for it? These new workflows are designed to solve that—no more creative roadblocks. Now, tasks like storyboarding, creating UGC ads, or applying a color grade can be done in just a few clicks.
You can easily run workflows across filmmaking, social content, fashion shoots, marketing, animation, and more. Simply choose your assets and configure options—no complicated prompting required. It’s about making repeatable creative steps effortless to set up and reuse.
More workflows are on the way, so there’s always something new to discover. Give it a try and see what you can make! 🚀
Someone just built a Claude skill that clones any website with a single prompt 🤯
It reverse engineers the design, extracts assets, and rebuilds the full layout automatically. It's also open-source.
GitHub:
https://github.com/JCodesMore/ai-website-cloner-template
Google just released Genie, a real time world simulator you can actually walk through 🌍 Text or images turn into explorable worlds with object permanence and environments that keep expanding instead of collapsing.
This isn’t a game engine or a video model. It’s a foundation model for worlds 🧠 Today you move and look around. Tomorrow, those worlds respond.
Learn more: https://labs.google/projectgenie
ChatGPT is still the nº1 AI site, but its lead is shrinking fast. New Similarweb numbers show ChatGPT dropping to 64.6% global traffic share while Google’s Gemini jumped to 22%, and that’s the first time since 2023 ChatGPT has been under 65%.
The story isn’t "ChatGPT is falling apart", it’s Google finally turned distribution into growth. Gemini is everywhere in Google’s ecosystem, and that kind of default placement is brutally effective. If this trend holds, the market won’t be one king anymore, it’ll be two giants and a long tail fighting for scraps.
It sounds like a joke, but Google Research claims a prompting trick that actually works: paste your prompt twice, exactly the same, before sending it 🤖📋
In a recent preprint, they show that turning <QUERY> into <QUERY><QUERY> improves results when the model is not asked to reason. Across many benchmarks and models, repetition wins most comparisons and never clearly loses.
They also report that outputs don’t get longer and latency usually stays the same, since the extra cost is in reading the prompt, not generating the answer. Important caveat: once you ask the model to “think step by step,” the effect is mostly neutral. And padding the prompt with junk doesn’t work, so it’s not just about more context.
If you try it, repeat the entire prompt, not just the question. Just keep in mind you’re doubling input tokens, so you pay more and use up context faster.
💻🧯 PC makers are finally reading the room
At CES 2026, Dell basically admitted what everyone feels but no one in marketing wants to hear: people are not buying laptops because of AI stickers or NPUs. Every new Dell and Alienware machine still has AI hardware inside, but the pitch has shifted back to performance, thermals, screens, actual use cases.
AI hype is staying under the hood where it belongs. Consumers want a good laptop, not a lecture about neural engines.
@Prompt:
Restore and colorize the uploaded historical black-and-white photograph while keeping the overall scene, pose, and subjects consistent with what is visible. Do not change the composition, but you may reconstruct missing or unclear areas when the image is too damaged to determine exact detail.Читать полностью…
Perform a full restoration across the entire image, including:
repairing heavy fading, stains, cracks, scratches, and missing sections
reconstructing faces, clothing, and background features based on the shapes, silhouettes, and visible cues in the original
clarifying all people, objects, fabrics, and surroundings with natural detail
restoring edges and textures as they would realistically appear, without modern stylization
maintaining the historical era feel and keeping proportions natural
preserving identity and expressions as faithfully as the surviving image allows
When details are too obscured to recover directly, infer them in a historically plausible and realistic way, staying consistent with the likely clothing, hair, and environment of the period.
Apply historically accurate colorization: muted colors, natural skin tones, subtle era-appropriate hues. Avoid modern brightness or saturation.
Keep lighting and general scene structure consistent with the original photograph.
Final output should feel like a faithful reconstruction of the moment captured — restored, completed, and colorized while respecting the historical authenticity of the photo.