Google DeepMind’s Post

BCG recently conducted the largest pre-registered experiment on the future of professional work to assess the benefits of using ChatGPT-4 in multiple tasks. The results should surprise no one. Are you officially allowed to use ChatGPT at work? Consulting companies like BCG and McKinsey are no doubt using whatever advantage Large Language Models can offer... Check out the full report here: https://lnkd.in/gsXEw5eD #AI #problem-solving #technology #BCG #LLM

The potential for AI to aid in comedic writing was explored by Google DeepMind researchers, who found that while AI models were effective at simple tasks, they struggled to produce original, stimulating, or funny material. The study involved 20 professional comedians using large language models to generate jokes, with mixed results. The comedians appreciated AI for overcoming writer's block and providing a sketch structure, but found the generated material to be bland and generic. Additionally, the AI's limitations in creating surprising and incongruous content, as well as its bias, were highlighted. While AI may assist in faster writing, it may not produce truly original comedic content due to limitations in the data it draws from.

Human psychology and AI - it is this aspect of generative AI that I find incredibly fascinating. "Chatbots do not have egos, but they imitate humans who do." Though, it is rather human to characterize the chatbot outputs as "lies," they simply use statistical models based on human inputs. Perhaps when human "truth" is defined as "conformity to fact or actuality" but governed by human perception, you get what you get? Now, if chatbots imitate something other than human behaviors, we might dislike the output even more. After all, isn't life "squishy" and imperfect, and don't we like it that way? "Every utopia - let's just stick with the literary ones - faces the same problem: What do you do with the people who don't fit in?" --Margaret Atwood

Blog Alert ⚠️ https://lnkd.in/d5_ZVAB9 👈 💬Join us on this journey as we uncover the fascinating world of 🤖AI language models! 🌐📚 From ChatGPT to Google BARD, we'll explore their profound effects on communication, information access, and more. 🗣️ Don't miss this eye-opening discussion as we unlock the secrets behind AI language models! 🗝️  . . . #blog #chatGPT #GoogleBard #Bard #NaturalLanguage #LanguageGeneration #LanguageTranslation #CreativeandAssistiveWriting #NLPResearch #RealWorldUnderstanding #research #development #9series

I had a fun time presenting an AI 101 talk to Tegria this morning with some of my colleagues - Ray Gensinger, MD and Brian Cahill . I was tasked with explaining how generative AI works in a way that would not put a non-technical audience to sleep or send them scrolling to Instagram in despair. I find trying to teach something is one of the best ways to learn and crystallize my own thinking. To that end, I think the most important things a non-technical person needs to know about Generative AI/ Large Language Models (eg. like ChatGPT, etc.) are 1) all they are doing is predicting the next word in a sequence, 2) they can't reason or plan (even if it seems like they are, and 3) they don't know your data unless you give it to them in your prompt. I am really impressed with the level of engagement and interest in AI at Tegria. (And to add a fourth point from one of the comments - keep experimenting and iterating because we are still discovering the full extent of its capabilities.)

🚀🎉 I'm excited to share my latest medium article on "Machine Psychology." As AI advances, it's crucial to decode the 'black box' of these systems. It's not just about their technical performance, but the implications they hold for our society when they become more and more human like. 🔍 Key Takeaways: - How can companies relying on Large Language Models (LLMs) ensure consistent performance, especially when AI behavior remains so elusive? - Do standardized tests like MMLU truly reflect AI's real-world problem-solving? - Drawing parallels between understanding the human mind and AI systems: Can the insights from psychology help us decode the mysteries of machine thinking? 🤔 Reflect On: - Can we genuinely trust AI systems whose processes remain unclear to us? - While AI doesn't possess human-like consciousness or emotions, how can we form a model of machine psychology that resonates with societal values? Combining the two big fields I have been working with in the last 10 years, I explore the upcoming challenges in understanding AI systems like ChatGPT and the compelling need to understand the psychology of machines. 📖 Link: https://lnkd.in/e9ZsRmSx Share your thoughts and experiences on this topic. I always stay curious about new insights and input! 🌐 Working on this with the team Daksha! #MachinePsychology #ArtificialIntelligence #OpenAI #GPT4 #AIethics

Want to know why LLMs sometimes give the wrong answer? Discover hallucinations in large language models, how they come about and what enterprises can do to mitigate this phenomenon. Learn More https://ow.ly/b1cv50RnUz6

These AI hit pieces are super weird and starting to get super annoying. They pose a question, prompt the base LLM models about it, then talk about how much AI sucks, weird and odd. Want humor in your AI models? I can do that. Because I asked the questions, then experimented. I didn't just go off screaming about how much AI sucks (for whatever reasons I would feel the need to do that). Here is my research on it. https://lnkd.in/gsj_cg3Y

Some new research: "Consultants using AI finished 12.2% more tasks on average, completed tasks 25.1% more quickly, and produced 40% higher quality results than those without." There's some some interesting nuance (see the Jagged Edge) and the application is becoming a "Centaur" or a "Cyborg", using AI enough to understand what is best for you to do and what is best for AI to do. I get really excited about this kind of stuff, and I love learning how different people are using these tools. I met a guy yesterday that uses ChatGPT to translate and fix his auto-captions on YouTube. The article also notes the danger of "falling asleep at the wheel" and just using AI without thinking critically. I've seen that, and the solution isn't to stop using AI, it's to think critically and use it as a tool, not a crutch. The last part that I think is important is how AI levels the playing field in industries. Individuals with less experience had significant improvement (43%) and those with more experience had less improvement (17%). So here's my pitch. It's simple. I use AI a lot, I have helped others use AI in new ways, and I read about AI a lot. If you want to talk about AI, maybe you can't figure out why a prompt isn't working or you want to learn something new, or share cool things that you're doing, send me a message and let's talk.👍 https://lnkd.in/gZi8JifV (Not a perfect post but it was written by me)

With artificial intelligence (AI) and ChatGPT on the rise, many are wondering if they will replace editors 🤔 In our latest blog, we take a deep dive into natural language AI tools and discuss if human editors really are superior when it comes to editorial projects. Head over there to discover: 🌟 The parts of editorial projects where AI might struggle 🌟 Why style and tone of voice needs a human touch 🌟 How editors and natural language AI tools could work together You can find it here: https://loom.ly/QhCNKoE #editingservices #writingservices #copyeditor #agencylife #marketingcommunications