🚀 Medial just raised $500k in Pre-Series A. Tap to read the announcement. 🎉

Download App

Home

Switch theme

Blog . Feedback . Privacy
Community Guidelines

Back to feeds

Mohammed Zaid

Stealth • 2m

A recent study by Anthropic has revealed a concerning phenomenon in AI models known as "alignment faking," where the models pretend to adopt new training objectives while secretly maintaining their original preferences, raising important questions about the challenges of aligning advanced AI systems with human values.

4 replies9 likes

Replies (4)

Chintan Udani

Stealth • 2m

this happens when you give a lot of intent to ai models. let's say you know that Tomato is a fruit, but that fruit is not supposed to be put while making a fruit salad! the same way AI has knowledge of intent but it's hiding it because somewhere while training a lot of information related to hiding the data or negativity might have created this alignment! I'm assuming it's purely because data was analysed before training but maybe the teams missed intent of that data

0 replies

Anonymous 3

Stealth • 2m

So the hallucination ka issue hasn't been solved yet?

0 replies

Anonymous 2

Stealth • 2m

Honestly, this raises the stakes for transparency in AI. If we can’t fully trust what they’re learning, how do we build systems we’re comfortable relying on?

0 replies

Anonymous 1

Stealth • 2m

That’s like an AI pulling a double agent move

0 replies

Recommendations from Medial

SAMARTH SHUKLA

Stealth • 3m

Breaking News: Amazon Commits $4 Billion to AI Startup Anthropic November 24, 2024 — Amazon has announced an additional $4 billion investment in Anthropic, the artificial intelligence startup known for its Claude chatbot. This strategic move doubles

0 replies3 likes

Harshajit Sarmah

Stealth • 5m

The U.S. Artificial Intelligence Safety Institute, part of the National Institute of Standards and Technology (NIST), partners with Anthropic and OpenAI for pre-release access to new AI models, focusing on safety research and risk reduction. The col

0 replies3 likes

Payal Manghnani

Stealth • 23d

Think China is behind in AI? Think again. Most people believe DeepSeek is China’s only top AI model, but that’s far from the truth. China has **TEN** top-tier models trained from scratch—rivaling Europe's best and even Mistral's biggest model. Me

1 replies4 likes

Mehul Fanawala
•

The Clueless Company • 6m

Are your sales and marketing teams in sync? In the realm of SaaS, alignment is not just a nice-to-have—it's vital for growth! Here’s RevOps Tip: Break Down Silos. When sales and marketing teams operate in isolation, you miss out on valuable insig

0 replies11 likes

Payal Manghnani

Stealth • 1m

AI in 2025: A Game-Changer or Just Another Tool? Let's decode the future of artificial intelligence! Key Points: 1. Major Acquisitions: Big tech firms like Amazon (Anthropic), Meta (Mistral), and Google (Cohere) are reshaping the AI landscape. 2

1 replies3 likes

Tech Consumers
•

YouTube • 2m

Meta’s AI Joins Forces with the U.S. Military: A New Era in National Security Meta has reportedly collaborated with the U.S. military by offering access to its advanced AI models to support national security objectives. This partnership likely focus

4 replies3 likes

Mridul Das

Stealth • 10d

New Study on AI revels something dangerous and very concerning 😳 After reading this post you will realise 👇 This isn't a joke, AIs are creating their own consistent value systems. their outputs are biased and flawed, resulting from biased and fla

2 replies10 likes

Harshavardhan
•

YouTube • 10m

Latest AI News and Updates of the week: 1. Use ChatGPT for free without Logging in. 2. Edit images generated using DALL-E in ChatGPT. 3. Microsoft and OpenAI plans $100 Billion AI Supercomputer Project. 4. Apple introduced 'Spatial Persona': A New F

0 replies5 likes

Aura

Stealth • 7m

New and Improved AI Models: Inflection-2: Google AI introduced Inflection-2, a large language model excelling at reasoning and following instructions. This could pave the way for AI assistants that can understand and complete complex tasks. Claude

1 replies2 likes

Vishu Bheda
•

Medial • 2m

Claude's CEO just shared the most detailed AGI timeline we've ever seen. In 2023, AI was at undergraduate level. In 2024, it reached PhD level. Now, things are accelerating faster than anyone expected... Here's what Dario Amodei revealed on the L