Microsoft’s ‘Bingest’ Mistake
In less than a week, Microsoft’s updated Bing, which is run on a customized version of OpenAI’s ChatGPT, went from being the “next great thing” to a “brand-sinking albatross.” And Microsoft is entirely to blame.
ChatGPT is a truly fascinating example of an emerging industry that is also enjoyable to use. So, it is not strange that its novelty would cause its capabilities to be overstated by everyone from high-powered tech types to folks who are typically uninterested in the domain. This is true of every subsequent AI-adjacent construct that comes down the line.
What are the benefits and drawbacks of generative AI’s perspective on art, literature, or philosophy? This topic is at the ideal “tech readiness level” for conversation over tea or a beer. How can we tell what is authentic, imitational, or hallucinatory? What does this mean for designers, programmers, and customer care agents? After two years of cryptocurrency, there is finally a topic worth discussing!
The excitement appears exaggerated in part because it is a technology that is essentially intended to spark debate and in part because it pulls from the debate that surrounds every advancement in artificial intelligence. It almost has the same effect as “The Dress” in that it calls for a response, and that response calls for other responses. To some extent, the hype is created by itself.
Large language models like ChatGPT are particularly suited for experiments with minimal stakes that go beyond simple conversation, like the never-ending Mario game. In fact, that’s really the core of OpenAI’s development methodology: release models first privately to smooth out any rough edges, then publicly to see how they perform when a million users are concurrently giving them a test drive. People will eventually give you money.
The Bing Mistake
Microsoft wasted no time in making claims about how it has worked to make its customized BingGPT (not what they named it, but I’ll use it as a disambiguation in the absence of logical official names) safer, smarter, and more capable in last week’s big announcement. It really had a whole unique wrapping system it named Prometheus that was designed to lessen the likelihood of improper replies.
Regrettably, we appear to have skipped right to the part where Prometheus has his liver pulled out in front of everyone, as anyone familiar with hubris and Greek myth could have foreseen.
Welp!
In the first instance, by linking its brand too closely to OpenAI’s, Microsoft committed a strategic misstep. It was at a distance and exempt from responsibility for whatever mischief GPT gets up to because it is an investor and a party interested in the study the organization is performing. But, someone took the colossal mistake to fully embrace Microsoft’s already ludicrous Bing branding, turning the worst traits of the conversational AI from a curiosity to a burden.
ChatGPT is largely forgivable since it is a research software. Few people would have trusted it as a product before, and no one will now, given the claims on the box that it can help you write a report, plan a trip, or summarize current events.
Neither OpenAI nor ChatGPT will take responsibility for those mistakes. Everything that goes wrong will be a Bing issue as a result of Microsoft’s desire to control the marketing, branding, and user interface. Also, Microsoft’s search engine, which has always been outmatched, will suddenly be like the person in the old joke who said, “I constructed that wall, should they call me Bing the bricklayer? They don’t, though. One setback results in unwavering doubt.
No one will ever trust Bing to organize their vacation after one disastrous trip upstate. No one will believe a news piece can be objective if its summary is inaccurate (or defensive). Once vaccine misinformation is spread, no one will believe it when it comes to determining what’s real or not.
Nobody will believe Microsoft when it claims that “we solved it,” because it already pinky-swore that this wouldn’t be a problem because of Prometheus and the “next-generation” AI it controls.
The well that Microsoft just dumped Bing into has been contaminated. The unpredictable nature of consumer behavior today makes it difficult to predict the effects of this. Because to the surge in activity and interest, some users may decide to stick around, and even if Microsoft delays the entire deployment (which I believe they will do), the overall result will be an increase in Bing users. A win nonetheless, albeit a hollow one.
I’m more concerned about Microsoft’s tactical mistake of ostensibly misunderstanding the technology it chose to commercialize and promote.
They didn’t see this coming, did they?
It is obvious that any huge AI model has a fractal attack surface, cunningly creating new vulnerabilities where old ones are strengthened. Humans will always find a way to exploit that, and in recent years, determined quick hackers have shown how to circumvent security measures for the good of society.
If Microsoft had made the decision that it was okay with the thought that someone else’s AI model with a Bing sticker would be assaulted from all sides and probably say some pretty strange things, that would be sort of alarming. Honest but risky. Say it’s a beta, just like the rest of them.
Yet it really seems as though they were unaware that this would occur. They actually don’t appear to comprehend the nature or intricacy of the issue at all. And all of this comes after Tay’s famed corruption! Microsoft ought to be the one company that is the most reluctant to introduce a simple model that picks up on discussions.
Given that Bing is Microsoft’s lone defense against Google in search, one would expect that some testing would be required before risking such a valuable brand. The fact that BingGPT has experienced so many problematic problems in just one week seems to be irrefutable evidence that Microsoft did not thoroughly test it beforehand. We can skip the specifics because there are many possible reasons why it may have gone wrong, but the final effect is undeniable: the new Bing just wasn’t ready for mass use.
Now that everyone in the world seems to understand this, why didn’t Microsoft? It’s likely that it was deceived by the buzz surrounding ChatGPT and, like Google, chose to “rethink search” quickly.
Yes, people are currently rethinking search! They’re questioning whether Microsoft or Google can be relied upon to deliver search results that are even minimally factually accurate, whether or not they’re AI-generated. The few other businesses working on the task have not yet done it at scale, and neither company (not Meta) has ever proved this competence.
I fail to see how Microsoft can turn things around. They committed to the new Bing and the promise of AI-powered search in an effort to leverage their partnership with OpenAI and surpass a sluggish Google. The cake cannot be unbaked.
They probably won’t make a complete retreat. That would result in humiliation on a much larger scale than it is already experiencing. However, it could not even benefit Bing because the harm has already been done.
In a similar vein, it is difficult to picture Microsoft continuing on as usual. Its Intelligence is quite strange! Well, it is forced into doing a lot of this stuff, but it also threatens people, assumes various identities, humiliates its users, and has random hallucinations. They must acknowledge that, if not lies, but at the very least untrue statements about how poor Prometheus controlled inappropriate behavior were made. As it is obvious from what we have seen that they did not thoroughly test this system.
The only sensible course of action for Microsoft is one I assume they already chose: throttle invites to the “new Bing” and push back the release of certain features, a few at a time. Give the present version a deadline or a finite number of tokens, if possible, to make the train ultimately stop and slow down.
This is the result of implementing a technology that you did not develop, do not completely comprehend, and are unable to adequately analyze. Major AI deployments in consumer applications may have been delayed by a large amount as a result of this fiasco, but it certainly suits OpenAI and others who are developing the newest models just fine.
Though AI may play a significant role in search in the future, it is by no means prevalent now. Microsoft decided to learn that in a really unpleasant way. (They binged around and found out!)
