METR’s Post

OpenAI’s commitment to exploring democratic inputs, as outlined in its “Democratic Inputs to AI” program, signals a willingness to involve the public in shaping the behavior of powerful AI systems. The company’s newly established “collective alignment” team aims to collect democratic public input on AI behavior, raising anticipation about the potential influence the public may have on shaping AI values. https://lnkd.in/ebn2gqqf FuturistSpeaker.com #foresight #predictions #futuretrends #futureofwork #futurejobs #keynotespeaker #futuriststrategy #mostquoted

Going to be interesting to watch. He carries just a tad more #credibility safety-wise than Sam Altman. Jan Leike carries more credibility safety-wise too, and he's off at yet another company. #Competition is always a good thing in the marketplace. https://lnkd.in/g9XziiRh #AI #AIsafety #GenerativeAI

More high-profile exits from OpenAI. Recent departures include top figures like John Schulman, Peter Deng, and Greg Brockman (sabbatical), who have been instrumental in AI advancements. How will this affect the future of AI research at OpenAI? Details ⬇️ #OpenAI #AIleadership #AI #TechNews #Innovation #AIresearch #FutureOfAI #TechLeadership #DailyAI https://lnkd.in/d-himm8H

Wow! If you are wondering what Sora is really about-read this article. Sharon Goldman hits the nail on the head. Not only does it give insight into Open AI's past actions but it will probably be predictive of what's to come. According to the author, we need to look at Open AI's present actions alongside aspirations toward AGI. She quotes a fascinating interview with an insider who already commented on the power of video before its release: “You can imagine how a model that would be capable of generating a video could plan across long-time horizons, think about cause and effect, and then reason about things that have happened in the past.” What a read. https://lnkd.in/deX-q4ab

OpenAI reveals there are 5 'levels' of AI evolution towards human intelligence, and we’re already nearing level 2! 🤖✨ This insight sheds light on the rapid advancements in AI capabilities and their potential implications. Stay tuned as we explore the journey toward achieving human-like understanding. https://lnkd.in/ecjAftv2 #OpenAI #AILevels #ArtificialIntelligence #FutureTech

I really like this letter. Read it if you want to understand the AGI hype.

Always a great summary of current AI topics from my ol' pal Connie Guglielmo. If one were to read only one AI-related column a week, my pick would be Connie's. #ArtificialIntelligence #Insight #StraightTalk

✦ In a TECHCRUNCH DISRUPT interview, TONY FADELL criticized Sam Altman, highlighting his own EXTENSIVE AI EXPERIENCE and stressing the need for improved TRANSPARENCY in AI systems. ✦ FADELL advocated for GOVERNMENT INVOLVEMENT in AI development to mitigate potential RISKS associated with lack of oversight. ✦ Citing examples of flawed AI in CRITICAL CONTEXTS, he expressed concerns about the EFFECTIVENESS of LLMs and the implications for safety. #TonyFadell #OpenAI #AITransparency #TechCrunchDisrupt #AIRegulation #SamAltman

Earlier today, a group of former and current employees of OpenAI and Google DeepMind published an open letter to warn about advanced AI (https://righttowarn.ai/). In my view, the companies building frontier AI models have a responsibility to ensure the safety of their products and analyze their impacts. Google DeepMind recently published a +270 page report on the dangers of AI agents - which is great! OpenAI on the other hand is keen on applying a social media-like business model to AI, chasing profits without thinking twice about current and potential harms to society. Check out my post 👇

AI Landscape is getting interesting day by day, started by former OpenAI employees and backed by billions in funding from Amazon, Anthropic is one of OpenAI’s most formidable rivals. They’ve just released a new family of AI models called Claude 3, one of which they claim beats OpenAI’s GPT-4. Claude 3 is multimodal, which means it can understand text, images, PDFs, and other formats. The model will be able to process more data than GPT-4 — about 150,000 words at a time (200k context window), with larger windows for select users. It also has improved memory recall, surpassing 99% accuracy. How to access and use Anthropic’s Claude 3 Go to the Claude website and sign up. You can enter your prompt in the text box and press enter to get a response. Here are some things you can try with Claude: - Summarize Files like PDFs, research papers, essays, and more. Understand Images: Upload an image, like a complicated diagram, and ask Claude to explain it to you. - Understand Complex Calculations: You can upload images of a calculation and ask it to explain what the equations are for and how to use them. - Suggest Design Improvements: Upload an image of a website — or even your living room — and ask how you can improve the design. - Transcribe Handwritten Notes: Upload pictures of handwritten notes and ask Claude to transcribe it for you. https://claude.ai/ Courtesy: Superhuman Newsletter