Their AI can’t screw up that much.

• Errors in artificial intelligence models cause serious reputational damage

• Apple Intelligence was launched precisely with modest and almost multi-layered features to avoid problems.

• Other companies, such as Microsoft, seem to be making decisions to do just that.

At first it didn’t matter that ChatGPT freaked out and screwed up. It didn’t matter that he changed the ending of Game of Thrones when we asked him, or even that he screwed up some basic math. After all, he only had a few months to live. Nothing happened.

Now it matters.

This is precisely what large technology companies have begun to realize as they plunge into a race in which it seemed more appropriate to arrive early than to arrive well. They all rushed to be the first, and the real serious problems began. Microsoft’s chatbot ranted, and let’s not talk about Bard and then Gemini, who screwed up big time (not once, but twice).

And, of course, this made users change their attitude. What was funny isn’t so funny anymore, especially since we started using these generative AI models for relatively serious things.

Even scientists write research using ChatGPT, and using these assistants is already the bread and butter of programmers who use them regularly. The problem for them is that today 52% of ChatGPT programming responses contain erroneous data.

All of this has begun to clarify for many what some already knew all along. ChatGPTs of the whole world don’t know what they are talking about. They release it and it seems like it makes sense. The sentences are well constructed and the tone of the participants, who respond in natural, clear and reasonable language, is completely safe. As if this really was the final and correct answer.

But in many cases this is not the case, and This becomes a big problem for reputation. those who operate these chatbots. Google is probably the hardest hit by this issue. First of all, because for a quarter of a century we have been using their search engine with the confidence that it will show us exactly what we need – of course, with a lot of advertising.

We trusted Google, but we didn’t trust Gemini as much. Not from Gemini, not from ChatGPT, not from Copilot. And we are good at this: it is important to revise their answers, because it is easy for them to give us partially or completely incorrect answers and confuse us.

Given this situation, companies are starting to realize how important it is for generative AI models to make fewer mistakes or behave in a way that makes us trust them (a little more). And there are several approaches.

The most striking thing is also the most recent. Apple unveiling its Apple Intelligence – since they are not allowed to use the term “AI” – left us a little unimpressed with its capabilities. The AI ​​functions running on their operating systems are more or less the same. Actually, It’s “just the same thing”because many of them operate on a limited basis.

The best example is their AI image generator, which they call Image Playground, which can be used to create emojis and images with a finish that is far from photorealistic.

No more oil paintings of Tim Cook or a possible confrontation between Elon Musk and Mark Zuckerberg. No dressing the Pope in Balenciaga and certainly no making deepfakes of explicit images of Taylor Swift.

The solution may be disappointing—Image Playground looks almost like a toy—but What this certainly does is avoid problems for Apple.. Yes, you won’t be able to do much, but what you do do will probably be pretty good. This not only prevents misuse, but also prevents disasters like the one Stable Diffusion 3 recently experienced: its AI image generation model, one of the most famous in the world, generates anomalous human bodies. This won’t happen with the Apple model (as expected).

Microsoft, it seems, has also recently reconsidered its views. Recall’s recent presentation offered the most exciting feature yet, but criticism soon arose over its privacy and cybersecurity implications. What did the company do? Postpone deployment. It was supposed to be part of the new Copilot+ PCs launching next week, but criticism and complaints forced Recall to be delayed: it will launch soon for Windows Insiders, and later it will be widely available. But now I hit the brakes. It wasn’t worth the risk.

Google may also be starting to realize that rushing is a bad idea, especially in a green segment that could have a significant impact on its business. This is where the Mountain View company is in the most precarious situation compared to its competitors: Apple already has its ecosystem offering, including the iPhone, and Microsoft is also quickly building one in Windows.

Meanwhile, they can’t risk that ChatGPT will end up stealing their wallet and we’ll end up searching more with it than with its search engine. But they can’t take another risk: recommending putting glue on pizza. I wouldn’t want to be in Sundar Pichai’s shoes. right now, because finding the balance (we need something fast, but it also works quite well) is the biggest challenge for all these companies today.

We may be at the beginning of a new mini-era of revolution driven by artificial intelligence. One where chatbots don’t screw up so much.

In Hatak | Social networks allow us to work for them for free. Now they want to blame us for it