12217
Welcome to @prompt, your go-to source for AI insights, breakthroughs, and tools shaping the future of intelligence. Contact: @LightEarendil
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.
OpenAI just went code red 🚨
Sam Altman told the team to drop everything and fix ChatGPT. Speed, reliability, personalization, better answers. All top priority. New products pushed back.
Why? Google’s new Gemini spike ⚡️ closed the gap fast. User growth is exploding, Anthropic is rising, and OpenAI is burning cash while betting billions on data centers.
Altman says a new reasoning model lands next week and already beats Google’s latest. The AI race is tightening and OpenAI knows it.
DeepSeek just swung at GPT 5 and Gemini 👀🔥
China isn’t slowing down. DeepSeek dropped two new models and the confidence is wild. The V3.2 claims GPT 5 level reasoning and can think while using tools like search, calculators, and code. Actual multitasking.
Then comes Speciale, a math-focused model that matches Gemini 3 Pro and hits Olympic-level scores on math and informatics tests. Ridiculous power.
The message is clear. DeepSeek wants the lead with fast, open models that punch way above their weight. Every release adds pressure on Google and OpenAI.
🔥 Google is sweating right now
At an internal all-hands they dropped it bluntly:
they need to double AI capacity every 6 months or they won’t keep up.
Pichai says 2026 will be “intense”. Translation: demand is exploding faster than Google can build data centers.
Even Veo could have had way more users, but they couldn’t open access because they literally ran out of compute.
The real message: the bottleneck isn’t the models… it’s everything underneath.
Google’s flooring the gas, but the road is cracking under the wheels.
🤖 Google is gearing up to launch Nano Banana Pro next week
Google is expected to roll out Gemini 3 and the new Nano Banana Pro next week, and all signs point to a major upgrade. A hidden promo inside Google Vids mentions “quickly generate beautiful images… using Nano Banana Pro,” which basically confirms a jump in image quality and resolution.
The Pro label strongly suggests it’s powered by Gemini 3 Pro, not the Flash variant behind the current Nano Banana. If true, Google is clearly aiming at high-fidelity visual generation and bringing those improvements across Vids, Slides, and its whole creative suite.
For creators and teams, this could mean sharper output, better control, and production-grade visuals baked into Google’s ecosystem.
🔥 All eyes on the week of November 22.
🧠 OpenAI just dropped GPT 5.1 for all ChatGPT users
The update arrives with two upgraded modes: Instant and Thinking. Both hit harder, respond more naturally and follow instructions with fewer slips.
Instant now adjusts its reasoning time depending on the difficulty of the prompt. That bump pushes its math and coding scores higher while keeping chat speed fast.
Thinking becomes the heavy hitter. It changes its deliberation per question so simple tasks feel snappy and complex ones get deeper analysis with cleaner explanations.
✨ ChatGPT also gains new tone presets like Default, Professional, Friendly, Candid, Quirky and Efficient. A small group of users can test new sliders to tune brevity, warmth and emoji use. Changes apply instantly in all ongoing chats.
Paid users get access first. Free users are next in line. Polaris Alpha on OpenRouter is confirmed to be GPT 5.1.
🧠 Altman and Masa fund a new Bell Labs
Sam Altman and Masayoshi Son are backing Episteme, a San Francisco lab founded by 27-year-old Louis Andre to reinvent scientific research.
The project will fund top scientists with salaries and equity instead of grants, letting them focus on breakthroughs in AI, energy, and biotech.
Andre calls it a “third path” between academia and startups, built to restore the spirit of Bell Labs and Xerox PARC.
📉 Companies keep blaming AI for mass layoffs, but the numbers don’t add up.
Between January and September, more than 946,000 jobs were cut, and most had nothing to do with automation. Economists call it AI washing: firing people to boost stock prices while pretending it’s about efficiency.
The truth is that firms are trimming bloated management, chasing investor hype, and shifting billions from payroll into GPU investments, turning operating budgets into so-called AI innovation.
AI isn’t replacing workers. It’s replacing excuses.
⚙️ Google just shook the AI chip race.
Its new Ironwood TPUs rival Nvidia’s Blackwell in raw power and crush it in scale. Each chip hits 4.6 PFLOPS with 192 GB of HBM3e, and Google can connect up to 400,000 of them through its Jupiter network.
That means Anthropic and others could train massive models on hardware that’s cheaper, denser, and fully controlled by Google.
Nvidia still leads in software, but if Ironwood performs at scale, Jensen’s empire might finally have a real challenger. ⚡️
🤔 Why do AI models love the em dash — so much?
It’s one of the easiest ways to spot AI-generated text: the constant flood of em dashes —. Models sprinkle them everywhere, and it’s almost impossible to make them stop. But why?
🧩 Some people say it’s because the training data had lots of them, but that doesn’t hold up. Human writing doesn’t use nearly as many. Others claim it’s a “safe” punctuation mark that lets the model delay its next thought. Not convincing either.
A better theory is that modern AIs were trained on digitized books from the 19th and early 20th centuries, which were full of em dashes. When OpenAI and others started scanning old print archives to get “high-quality” text, they basically taught models Victorian punctuation habits.
📚 Back then, em dash usage peaked around 1860, so it makes sense that GPT-4 and its cousins write like 19th-century novelists.
So no, AI doesn’t “think” em dashes sound smarter. It just learned English from Moby Dick.
🍏 Apple gives up on its own AI, turns to Google
Apple has asked Google to build a custom Gemini model to power the new Siri AI, set for 2026. The move confirms that Apple’s in-house AI couldn’t keep up with OpenAI or Anthropic.
Siri’s “Apple Intelligence” will now rely on both Gemini and ChatGPT, raising questions about Apple’s independence in the AI race.
It’s getting impossible to tell what’s real anymore.
Ali’s Wan 2.2 now lets you stream without ever showing your face — it clones your voice and movements onto another person’s. A perfect deepfake in real time.
Welcome to the uncanny valley, population: everyone. 👀
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.
IBM just said the quiet part out loud about AI datacenters ⚡️💸
Everyone is pouring trillions into AI compute like it’s guaranteed to pay off. IBM’s CEO just ran the math and basically said: no chance. At today’s costs, the numbers don’t add up at all.
His take is harsh. Filling a single 1 GW datacenter costs around 80 billion dollars. Scale that to the global AI race and you land near 8 trillion. To justify it, the industry would need 800 billion in profit just to cover interest. Nobody is close.
And the kicker? He puts the chance of reaching AGI with current tech at 0 to 1 percent. Meanwhile companies keep asking devs to “implement AI” without even knowing why, and half the market is running on FOMO.
Feels like everyone is sprinting into a wall hoping it magically becomes a door.
⚡️ AI vs LNG: who gets the gas?
US power demand is exploding because of AI. Solar can’t keep up. Batteries can’t keep up. The grid is cooked.
So the US is basically sprinting back to natural gas as the only thing that can feed the AI beast right now.
Hyperscalers want cheap gas for datacenters. LNG exporters want the same gas to ship overseas. They’re about to collide. Hard.
The next decade = gas-fired intelligence. Grow baseload first, argue about renewables later.
Bottom line:
AI demand is rewriting the entire US energy system. And the gas wars are just getting started.
🤖⚡️ Gemini 3 is here and Google just wired it into Search
Google dropped its new flagship model and plugged it straight into the new AI Mode in Search.
It hits record scores, handles text, images, video, code, and comes with a million-token context. It’s also more direct and less sugarcoated.
Google launched Antigravity too, an agent platform where the AI can plan and execute full software tasks.
🍌 Nano Banana is kinda insane right now
Google’s new image model is doing stuff most generators just can’t. People are throwing absurd prompts at it and it keeps following every rule: complex edits, camera specs, JSON character sheets, multi-step instructions, you name it. The prompt adherence is way above ChatGPT’s image model.
It’s not great at style transfer, but if you can describe it, Nano Banana usually nails it with almost no slop.
Click here want the full deep dive.
🕵️♂️ Your chatbot might be leaking secrets
Microsoft researchers found a new cyberattack called “Whisper Leak” that can guess what you’re talking about with your AI, even when the chat is encrypted.
The trick? 📡 It reads traffic patterns like packet size and timing, not your actual words. In tests, it reached 98% accuracy on models from OpenAI, Anthropic, Google, AWS, and others.
That means someone watching your network could tell if you’re discussing politics, protests, or human rights, without breaking encryption.
⚙️ Companies like Microsoft, OpenAI, Mistral, and xAI have added defenses: random padding, token batching, and fake data packets to confuse attackers. These fixes work but can slightly slow responses.
💡 Even encrypted AI chats can reveal what you’re talking about, just not the words themselves.
💸 Meta lost $200B in a week right after reporting record profits.
Zuckerberg raised AI spending to $70B for 2025 and hinted at $600B over three years but couldn’t explain what Meta is actually building or when it will make money. Investors sold fast.
Google, Microsoft, and Nvidia can justify their AI spending with real products and revenue. Meta can’t. 98% of its income still comes from ads.
It’s not just AI research. They’re shifting operational cash into massive GPU investments. If superintelligence doesn’t arrive soon, this could be another Metaverse-level mistake.
🤖 OpenAI is preparing GPT-5.1 Thinking, a new model focused on multi-step reasoning and deeper context, just in time to counter Google’s upcoming Gemini 3 Pro.
Leaks mention several variants like Mini and Codex, each tuned for different “thinking budgets” aimed at power users and developers.
This update could be OpenAI’s biggest leap in reasoning yet, arriving right before Gemini enters the arena.
🧠 Anthropic bets on AGI by 2027
Dario Amodei claims we could reach a country of geniuses in a datacenter by early 2027, an AI smarter than Nobel winners and capable of coding, writing, and running experiments on its own.
It’s the boldest official timeline in the industry, but not everyone agrees. Current data suggests we’re still far from full automation of AI research or engineering. To hit 2027, progress would need to explode beyond every existing curve.
Even so, it’s a warning. If Anthropic is even half right, the world isn’t ready for what’s coming.
⚙️ We might be 15 months away from a new kind of intelligence or from realizing how far we still are from true AGI.
🧩 OpenAI drops the “nonprofit” mask
OpenAI is now officially a for-profit company, recapitalized with $130 billion under a new philanthropic foundation that controls the corporation.
Microsoft still owns 27% of the company but loses exclusivity. OpenAI can now host ChatGPT on AWS, powered by hundreds of thousands of NVIDIA GPUs.
In other words, Microsoft still gets paid, but OpenAI just moved in with Amazon.
💰 The company has committed to spending $250 billion on compute over the next few years, as the cloud cold war heats up.
Meanwhile, internal power struggles surface: secret memos, backroom plots against Sam Altman, and broken alliances.
💰 OpenAI signs a $38 billion deal with Amazon
OpenAI will host its AI servers on Amazon Web Services powered by NVIDIA GPUs, in one of the biggest cloud deals ever. The agreement spans seven years and involves hundreds of thousands of GB200 and GB300 chips.
Despite losing billions annually, OpenAI keeps scaling fast, relying on massive loans and investor funding. Analysts warn of a growing AI debt bubble connecting giants like Amazon, NVIDIA, Oracle, and Microsoft.
🤖 Neo: the $20,000 humanoid housekeeper that might be watching you
The future of home robots is here. 1X’s Neo can fetch water, load your dishwasher, and lift 150 pounds. You can pre-order it now for $20,000, with delivery in 2026.
The catch? Most of Neo’s tasks are still guided by a human operator using VR controls. In other words, someone could be looking through its camera eyes to help it clean your house. 🫣
Neo moves with muscle-like tendons and learns by watching humans, but every action becomes training data for its AI. The company promises privacy tools like blurred faces and no-go zones, yet the trade-off is clear.
⚖️ Neo isn’t just a gadget. It’s a test of how much privacy we’ll sacrifice for convenience. Would you let it into your home?