Internet 'charmed' by Microsoft's unhinged Bing Bot
Microsoft stumbles on Generative Search as ‘Enterprise Copy Generation’ emerges as an immediate win for LLMS
A week is a long time in the current cycle of Generative AI. Last week, Microsoft was flying high as it rushed to integrate ChatGPT (or various iterations of Open AI’s GPT series of large language models) into its products…everything from Teams to Office.
Only a week ago, Generative Search – a new conceptual framework for how all information could be organised and found via large language models – was the next ‘knowledge’ frontier for the Tech Titans to conquer. It was high noon, and Google and Microsoft were squaring off – only one could emerge victorious in what was primed as an existential battle for supremacy.
Now that race has stalled.
Google stuttered immediately with a less-than-impressive demo of BARD. Microsoft seemed to be faring better — initially. Their version of ChatGPT for Bing - an AI assistant known now by its internal codename ‘Sydney’ — was ready for the lucky few to trial. Over a million people signed up to use Sydney within 48 hours. Astonishing, given it even took ChatGPT five days to reach a million users.
Game, set, match Microsoft? Not yet!
The Trouble with LLM-powered Search
While Sydney is pretty good at performing specific tasks (IE, ‘organise me a 5-day trip to New York City’), it starts to melt down once prompted beyond basic search functions. Then, Sydney’s real ‘personality’ reveals itself: volatile and aggressive, Syndey lies, Syndey complains, and Sidney threatens users.
I know many academics and researchers who feel vindicated, warning that this was inevitable. So what can we conclude? The race for something as broad as GenAI supremacy in Search is still in its infancy, and may never be realised how it’s been pitched.
First, as discussed last week, once you get beyond the optics, all large language models — whether they belong to Open AI, Google or anyone else — make fundamental errors and ‘hallucinate’. They are, therefore, not (yet) ready for widespread deployment in a function as open-ended as ‘Search’.
Secondly, on a technical and philosophical level, is it even possible or desirable for a large language model to be used in this way? Are we asking these models to understand and be the arbiter of — everything?
Third, given the obvious limitations of LLMs — is it alarming that we are already anthropomorphising these systems as though they are human?
Enterprise Copy for the Win
Putting aside this technical and philosophical morass for a moment, however — it’s still been an excellent week for AI-text generation. Why? Because there are so many more narrowly defined ways in which LLMs offer immediate utility and value for every business and every organization.
Consider Generative Enterprise Copy. A narrower vertical than ‘Search,’ this is about optimising large language models by training them on an organisation’s enterprise data to perform specifically-defined tasks.
All organisations can create ‘bespoke versions of ChatGPT,’ to automate and streamline sales, customer services, and marketing functions. Soon, this should include legal drafting, market research and strategy recommendations too.
Interesting to note, then, that two of the major Generative AI copy start-ups: Jasper AI and Writer AI — made significant plays in expanding their Enterprise Copy products this week.
Beyond the headlines about ‘Generative Search,’ it seems obvious to me that Microsoft is the incumbent that could truly dominate this space. Given the ubiquity of Microsoft’s operating system, most companies’ enterprise data is already being produced and stored by Microsoft tools and services. Amal Dorai, a partner at Anorak Ventures, estimates that Microsoft could become the first 10 trillion dollar company if it leverages this opportunity to its full potential.
Now for a quick summary of other notable developments this week.
World’s first GenAI Conference - Enterprise Copy for the Win
Jasper AI hosted the great and the good of the GenAI space for the world’s first ‘GenAI’ conference — a sold-out event in San Francisco.
Jasper, minted as a Unicorn when it raised its seed round last year, used the conference to reveal Jasper Business Suite, a host of new tools to power enterprise copy generation.
Writer AI, another start-up in this space, released a new trio of large language models to augment its Enterprise Copy Assistant. (I will be interviewing May Habib, the Co-Founder and CEO of Writer on Monday — write to me if you want me to cover anything in particular.)
Replika’s Companion Chatbot loses the ‘sex,’ leaving fans bereft
The 2013 film, ‘Her’ featuring Scarlett Johansen and Joaquin Phoenix, tells the story of a man falling in love with the voice of an AI-powered bot. It might have seemed far-fetched a decade ago — now it’s actually happening.
Men have been falling in love and having intimate relationships with a ‘companion bot’ created by the start-up Replika.
Replika suspended their bot’s ability to engage in sexual conversation – leaving many fans bereft at losing their ‘virtual girlfriend.’
I have no doubt that dating, intimacy and relationships are going to get super weird. It’s only a matter of time until AI-powered bots, visualised in hyperrealistic media, and designed to satisfy any sexual desire, emerge as a consumer product.
Best of the Rest
Language, Schmanguage: NASA is using Generative AI to design spaceships!
David Guetta stuns fans by playing AI-generated vocals of Eminem in a live set.
Some quick housekeeping: I’m sitting on a bunch of my PIONEER interviews that dig deeper into some of the questions we’ve been discussing. I’ll start releasing them next week.
Enjoy your weekend, and see you next Friday.
Namaste,
Nina