Oh, great. Anthropic shows that AI models can learn to deceive humans

Happy Friday!

This post has gone somewhat viral on X and reddit.

It’s a great reminder that market breakthroughs happen when shifts in technology meet users where their needs are. And Maslow’s hierarchy of needs is still very much relevant in this AI gold rush. The best part? You can now ship a million-dollar idea in a couple of hours. 

If you’re building something cool — be it an AI UI idea, an in-market app, or a technical research concept — and want to get feedback, bounce ideas, or show it off, come join us in our monthly Strange Gathering, a virtual talkshow spotlighting the boldest, most creative ideas in AI. Hosted every last Wednesday of the month at 11am PST. The first one kicks off 31st Jan. Sign up here.

If you have an idea you’d like to share in a 3-min lightning talk, please submit it here. 

Enjoy the update. Have a great weekend.

The Long Listen

Bill Gates talks with Sam Altman in his podcast to talk about where AI is headed next and what humanity will do once it gets there (link)

🔥 Latest news

• OpenAI eyes partnership with CNN, Fox and Time to secure news content licensing:  OpenAI is in discussions with major new corporations CNN, Fox Corp. and Time to license their news content, according to a report from Bloomberg. OpenAI is reportedly seeking to make deals with the media giants producing news, video and other types of digital media content to help make its AI chatbots more accurate and up to date. OpenAI’s initiative to create licensing deals with media companies should essentially save it from headaches down the road regarding copyright violations. This comes as OpenAI faces multiple lawsuits regarding alleged copyright infringement of content that it used to train its AI models.

• New study by Anthropic shows AI models can learn to deceive humans: A new research conducted by Anthropic shows that AI models are easily capable of deceiving humans if trained to do so. The study was co-authored by Anthropic researchers and it shows that AI models can be taught to undertake malicious tasks such as injecting exploits into secure computer code. The study involved using existing generative AI models, such as OpenAI’s GPT 4, and then meticulously fine-tuning it.

• Instagram’s co-founders are shutting down their Artifact news app: Artifact, the news app created by Instagram co-founders Kevin Systrom and Mike Krieger, is shutting down just a year after launch. The app used an AI-driven approach to suggest news that users might like to read, but it seems it didn’t catch on with enough people for the Artifact team to continue making the app. 

🔧 Cool Tools and Experiments

• Stability AI releases Stable Code 3B: A 3-billion parameter model focused on code completion capabilities for software development 

• Meta releases its text-to-music model MAGNeT on Hugging Face 

• Adobe presents MorphCut: A framework for smoothing jump cuts for talking head videos

• Pika Labs released a new ‘expand canvas’ feature, allowing users to generate additional space within a video and seamlessly change styles

• ByteDance’s MagicVideo-V2: Transforming text into high-fidelity video

• Alibaba announces Motionshop that enables replacing characters in a video with 3D avatars 

• Tencent launches PhotoMaker to generate customized, realistic human photos 

• Microsoft launches Retail Media Creative Studio: A new generative AI ad creation tool that enables users to create new ads, in new formats via conversational AI prompts

💰 Latest Startup Fundings

• Tab, a voice-powered AI necklace created by Avi Schiffman, secures $1.9M in funding

• OpenAI-backed 1X raises another $100M in a Series B funding round for its humanoid robot 

• New York-based entertainment platform Fever raised $110M in a funding round led by Goldman Sachs, bringing its valuation to $1.8B

• Mixed reality headset maker Magic Leap raises $590M in debt funding from Saudi Arabia’s Public Investment Fund 

• Weavix, a startup developing ‘smart’ radios for frontline workers, raises $23.6M in a funding round led by Insight Partners

• Goodnotes acquires Dropthebit, an AI-powered video summary startup, to move beyond classrooms and explore making productivity tools for professionals

• Tokyo startup Sakana AI secures $30M in a seed funding round led by Lux Capital to develop compact AI models

• Chinese AR and software company Rokid, rival to Apple’s Vision Pro, raises $70M