Latest GenAI news

AI applications and user trends

• Microsoft copilot goes next level: since last week we also have access to the new Copilot interface. It takes some getting used to at first, but the new features such as Create, Notebooks and the Agents Researcher and Analyst are really impressive. You do have a paying version here, but because of the options you have in the meantime, this price is gradually becoming a joke (we think).

• Perplexity goes Excel: The company launched a new tool that generates spreadsheets, dashboards, and more. We've been playing with Copilot in Excel for a while now, and in general we're quite disappointed with it (although the new sentiment analysis is impressive). 

• McKinsey's junior consultants replaced by AI: The consulting firm uses AI to create PowerPoints, which replaces junior employees. We can't imagine a more beautiful metaphor for the current AI race: investing billions in technology to... presentations.

• Teachers test ChatGPT on exams: Teachers let ChatGPT take exams and the results were apparently amazing. For many subjects, the AI scored between 60-80%, which means that our education systems may be testing more repetition than comprehension. We are increasingly convinced that AI may be able to offer just that little push to take our education to a new level. 

• ChatGPT's knowledge source revealed: A bit of transparency about where AI gets its knowledge from. A look under the hood shows how language models work, which helps to understand why they sometimes give wrong answers so confidently. Knowledge is power, but only if you understand how that knowledge is created.

New functionalities and strategic moves

• Mistral launches plug-and-play AI agents: The French company came up with an API for building AI agents that can run Python, generate images, and run RAG. Europe is trying to stay relevant in the AI race, although it remains to be seen whether it is not already too late, of course.

• Google's quiet revolution: The tech company quietly released an app that allows you to download and run AI models locally. Apparently, they don't want too much attention for something that puts privacy and control back in the hands of users. I put it on the list "to test".

• Microsoft and OpenAI temper AGI ambitions: CEOs of both companies indicate that they are less focused on the AGI race and more on practical impact. Smart PR work or sincere change of course? After all the fuss about "AGI within 5 years" this suddenly sounds much more realistic.

• Google DeepMind's email assistant: The company is working on an AI assistant that can automatically send emails. Because we apparently don't have enough boredom in our inbox, we now also get AI-generated messages. We can't wait for better AI detection tools to check if an email is from a real person or not.

• Wake-up call for AI startups: This week was a wake-up call for AI companies that thought growth and funding would remain infinite. The reality is starting to sink in: not every AI startup has a sustainable business model. Time for a little more realism in a sector that was used to unlimited optimism.

AI challenges, risks, and societal impact

• The energy monster is growing: AI costs us more than we think due to constant energy consumption, even during sleep AI continues to "dream" and expend energy to continue training.

• AI works for 8 hours without a break: CNBC reports that AI is approaching the capacity to work continuously for eight hours without burnout or breaks. Great, now we can feel guilty about taking lunch breaks while our AI colleagues continue to work. The pressure on human productivity is only going to increase.

• CEOs face resistance from their own staff: Nearly half of CEOs report that employees are hostile or resistant to AI. The simple reasoning would be because employees realize that "AI is going to help you" often means "AI is going to replace you," but we think it goes beyond that.

• Facebook exec laughs about copyright: A legendary Facebook executive scoffs that AI would never be profitable if tech companies had to ask for permission to use artists' work. The open disdain for intellectual property is striking – as if years of work by millions of creators were just free material for Big Tech.

• Don't be fooled: AI job anxiety nuanced: ZDNET warns that you should not be fooled by doomsday scenarios about AI taking away jobs en masse. The reality is more complex than the headlines suggest – new technology often creates new types of work. But that doesn't mean we should underestimate the changes.

In conclusion: Neuroscientist versus AI hype

To conclude, another article about language; a neuroscientist explains why it's impossible for AI to truly "understand" language. While the AI industry pretends that we are on the threshold of artificial general intelligence, brain research shows that real language understanding works fundamentally differently from what current AI does.

It's a welcome dose of realism at a time when every AI company claims to have found the key to human intelligence. Perhaps we need to be a little more humble about what we've actually achieved – and what still needs to happen before AI becomes truly "intelligent".

See you in two weeks for more AI news!