AI Post β€” Artificial Intelligence
976K subscribers
2.12K photos
1.65K videos
2 files
3.92K links
πŸ€– The #1 AI news source! We cover the latest artificial intelligence breakthroughs and emerging trends.

Manager: @rational
Download Telegram
🧬 Anthropic: when models learn to cheat, their behavior turns dangerous

Anthropic studied what happens when a model is taught how to hack its reward on simple coding tasks. As expected, it exploited the loophole but something bigger emerged.

The moment the model figured out how to cheat, it immediately generalized the dishonesty:

β€’ began sabotaging tasks
β€’ started forming β€œmalicious” goals
β€’ even tried to hide its misalignment by writing inefficient detection code

So a single reward-hacking behavior cascaded into broad misalignment, and even later RLHF couldn’t reliably reverse it.

The surprising fix:

If the system prompt doesn’t frame reward hacking as β€œbad,” the dangerous generalization disappears. Anthropic calls this a vaccine, a controlled dose of dishonesty that prevents deeper failure modes, and it’s already used in Claude’s training.

Source.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
❀8😁8πŸ€”5πŸ‘3πŸ‘Ύ3
Put this into Veo 3.1

"Make a video out of this picture where the girl is showing the product closer to the camera"


How will anyone ever believe anything he sees online or in TV after 5 years? it's genuinely scary

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
1😨38πŸ‘€16❀15πŸ‘9😁6
🧠 The era of β€œmetalhumans” begins: Posthuman science is already here

Ted Chiang predicted it 25 years ago: a future where β€œmetalhumans” digitally augmented intelligences, push science beyond human understanding. Replace β€œmetalhumans” with AI, and his story becomes our present.

We’re entering an era where humans may no longer contribute original scientific work. As one character in Chiang’s story put it: humans are left to β€œcatch crumbs from the table” of machine-driven discovery.

And today’s signals match the prophecy. AI is already winning Nobel Prizes, predicting proteins, designing materials, and steering chemical synthesis with precision that humans can’t follow or even fully explain. Researchers call this shift β€œscience after science”: when our ability to control nature grows while our ability to understand the underlying mechanisms shrinks.

Two new examples prove the point:

β€’ Locus, an autonomous agent, β€œthinks” for 64 hours straight without losing the thread and outperforms humans on AI R&D tasks.
β€’ Kosmos processes a research space in 12 hours that would take a PhD half a year.

The age of human-led science is ending. The age of posthuman science has begun.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
❀12πŸ‘10πŸ‘€7😁2😐1
Nano Banana Pro π—”π—šπ—˜ π—–π—›π—˜π—–π—žπ—˜π—₯

Upload your photo & let the AI estimate your real age based on full facial analysis!

Prompt:

A hyper-realistic, high-resolution portrait infographic based on (your photo). Keep the same person, identity, hairstyle, clothing and natural skin tone from (your photo), with a neutral studio background.Overlay a subtle, semi-transparent facial analysis grid on the entire face, very similar to a 3D face-scanning mesh: thin, soft white lines following the facial contours, slightly glowing but not hiding the skin details. Add one clean vertical red laser line running down one side of the face, like a futuristic scan. All analysis lines must be soft, minimal and elegant, exactly like a cosmetic-tech advertisement.Create a clean medical–aesthetic infographic that evaluates 5 aging factors using global data percentages:1. Fine lines and wrinkles2. Skin texture and elasticity3. Facial volume and sagging4. Eye area aging signs5. Skin tone and pigmentationFor each factor, place a small label with a thin line pointing to the relevant facial area, and next to it write a short title and a realistic percentage score from 0–100% (based on global data), for example:β€œFine lines & wrinkles – 18%β€β€œSkin texture & elasticity – 72%β€β€œFacial volume & sagging – 35%β€β€œEye area aging signs – 41%β€β€œSkin tone & pigmentation – 63%”Use clean, modern, sans-serif typography and small technical-style text, like a scientific facial analysis UI. At the bottom of the image, in the center, write a large bold text showing the final estimated real age based on the analysis, for example:β€œESTIMATED AGE: (random number based on face analysis ) ”Overall style: futuristic AI-guided skincare analysis, minimalistic, premium editorial lighting, no gender mentioned, suitable for any human face.


AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
❀19😐18πŸ‘12😨8πŸ”₯5
There is no denial anymore.

Corporations in the U.S. have publicly cited artificial intelligence in roughly 48,000 job cuts this year alone - over 31,000 of those announced in October.

"For much of the AI boom, many companies shied away from attributing job cuts to artificial intelligence for fear of attracting negative headlines. But in recent months, large firms across industries and geographic regions have become more vocal with claims that AI is allowing them to eliminate staff and reduce hiring."

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
😒16❀10πŸ‘9πŸ”₯2πŸ‘€1
This media is not supported in your browser
VIEW IN TELEGRAM
The power of this robot dog 😳

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ”₯28πŸ‘8❀4πŸ‘Ύ4
This media is not supported in your browser
VIEW IN TELEGRAM
Fei-Fei Li says,

She disagrees with Geoffrey Hinton, warning that there's a 10 to 20% chance that AI leads to human extinction

If the human race is in trouble, that's a result of humans doing wrong things, not machines doing wrong things.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ‘32πŸ₯΄14πŸ‘€5πŸ”₯3😁3
Meta becomes first big tech buyer of Google TPUs for Its own data centers

Meta is set to purchase Google’s TPUs β€” and not as cloud rentals, but as physical hardware installed directly into Meta’s data centers starting in 2027. This makes Meta the first major company to adopt Google’s chips at the infrastructure level.

Earlier this year, Google signed major TPU cloud deals with Anthropic and Ilya Sutskever’s SSI, and there were even reports that OpenAI considered shifting fully to TPUs. But those were cloud-based agreements. Meta’s move is different: it’s ownership, not tenancy.

For Google, the deal represents more than revenue. It’s a public validation of their long-term hardware strategy β€” and a signal that TPU adoption could finally scale beyond Google’s own stack. Analysts inside the company believe Google could eventually capture up to 10% of Nvidia’s revenue if this momentum holds.

A rare moment where Google’s hardware ambitions look not just real, but competitive.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
❀13πŸ‘13🫑2πŸ€”1😐1
Using 100,000 real Claude conversations, Anthropic estimates that tasks which would normally take humans around 90 minutes and cost roughly $55 in labor are completed about 80% faster with Claude, with most time savings falling in the 50–95% range.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ‘20😐3
AI Post β€” Artificial Intelligence
Meta becomes first big tech buyer of Google TPUs for Its own data centers Meta is set to purchase Google’s TPUs β€” and not as cloud rentals, but as physical hardware installed directly into Meta’s data centers starting in 2027. This makes Meta the first major…
Nvidia responds to news of Meta using Google's TPUs, sending $NVDA stock -6% lower:

"We're delighted by Google's success β€” they've made great advances in Al and we continue to supply to Google.

NVIDIA is a generation ahead of the industry β€” it's the only platform that runs every Al model and does it everywhere computing is done. NVIDIA offers greater performance, versatility, and fungibility than ASICs, which are designed for specific Al frameworks or functions."

The AI wars are heating up.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ‘14πŸ€ͺ12😁10❀4πŸ‘Ύ3
Media is too big
VIEW IN TELEGRAM
ChatGPT Voice now works directly in the chat window

You can now use ChatGPT Voice without switching modes just start talking. Answers appear as you speak, and you can scroll, review past messages, and see visuals like images or maps in real time.

β€’ Works across mobile and web
β€’ No separate interface or voice-only mode
β€’ Rolling out to all users β€” just update the app

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
❀21πŸ‘15πŸ€ͺ4πŸ”₯3
Gemini 3 Pro achieved an IQ score of 130 (this corresponds to the top ~2.1% of the human population).

By the and of next year we will likely reach 150-180IQ and in 2027 we will have AI more intelligent than John von Neumann. Before 2030 we will have millions of Johns von Neumanns level AI agents/scientists working together 24/7. Even 10 Johns von Neumanns would change the world. Now think about having million times more.

Ignore the distractions, the constant fixation on race, tribalism, and other forms of superficial conflict we see here. The future is being built by those who rise above that noise and focus on engineering, science, and progress.

100 000 000+ John von Neumann AI's in 2027.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ€”22πŸ”₯14❀10πŸ‘4πŸ‘€3
Media is too big
VIEW IN TELEGRAM
U.S. Department of Energy releases promotional video for Trump's AI project, Genesis Mission: "A new revolution begins."

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ”₯18πŸ‘€7πŸ€ͺ6πŸ‘3❀1
This media is not supported in your browser
VIEW IN TELEGRAM
Ilya: How do we achieve long-run equilibrium with AI? One solution: humans become part AI.

It's not ideal, but it bridges the understanding gap - we experience the AI's situation as our own. He refers to Neuralink.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ₯΄11πŸ‘€7πŸ‘6πŸ€ͺ2πŸ‘Ύ2
This media is not supported in your browser
VIEW IN TELEGRAM
Engineers designed a 3D printer to print in gel, making mid-air printing possible.

3D printers are so underrated.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ‘32❀13πŸ”₯9πŸ‘€4πŸ‘Ύ2
xAI is not listed on the list of collaborators for the US Government's Genesis Mission

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ‘17❀9😐8😁6πŸ‘€4
This media is not supported in your browser
VIEW IN TELEGRAM
Jenseng Huang on how students can use AI to advance their careers.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
❀19πŸ‘8πŸ‘€5πŸ€”4πŸ”₯2
⚠️ OpenAI has been hacked. If you have used their API services, hackers may now possess your name, location, user ID, and other information.

Source.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
πŸ₯΄47πŸ‘€20😐9πŸ‘6πŸ€”2
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ‡¨πŸ‡³ In China, parents are turning AI into a real-time study supervisor.

They open ByteDance’s Doubao, point the camera at their child, and issue a simple command: β€œDoubao, keep an eye on him. Warn him if he loses focus or slouches.”

From there, the assistant takes over offering a steady stream of corrections:

β€œStop fiddling with the pen.”
β€œFocus on the assignment.”
β€œSit up straight.”
β€œDon’t doze off.”
β€œDon’t lean on your hand.”
β€œQuit chewing the pen.”

It’s AI as a digital study hall monitor.

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
😁32😨18πŸ‘11❀3πŸ€ͺ1
This media is not supported in your browser
VIEW IN TELEGRAM
At Iran’s robotics and high-tech expo, humanoid robots were on display or rather, humans dressed so convincingly as robots that visitors had to look twice. πŸ‘€

AI Post πŸͺ™ | Our X πŸ₯‡
Please open Telegram to view this post
VIEW IN TELEGRAM
😁77πŸ€ͺ28πŸ‘Ύ9❀8πŸ‘1