In het kort
Google introduceert een nieuw AI-model, Gemini, om de concurrentie aan te gaan met OpenAI’s GPT.
Gemini werkt niet alleen met tekst, maar ook met afbeeldingen en video, en heeft verschillende versies.
Hoewel Google beweert dat Gemini superieur is aan OpenAI’s GPT-4, stellen experts dat de prestaties waarschijnlijk vergelijkbaar zijn.
Samenvatting gemaakt met behulp van AI .
Google and Facebook were immediately code-red when OpenAI suddenly launched ChatGPT to the market last year. This first version of the AI chatbot worked so well that millions of people started using it. They do homework, request recipes, program websites, or simply look up information.
Other technology companies have also been working on AI applications for years. Google even calls itself “One.” Artificial intelligence first The company, where AI comes first. But the company could not respond to the huge success of ChatGPT.
Gemini as a competitor to GPT
Programs like ChatGPT work on language models. For example, the free version of ChatGPT works with GPT-3.5, while the paid, more powerful version works with GPT-4. These language models interpret what the user says and then provide an answer.
Gemini does it for Google. However, this model works not only with text, but also with images and video. There will be different versions, the most powerful of which is Gemini Ultra. According to Google itself, this model is better than GPT-4 in many areas. There will also be a Pro version similar to GPT-3.5. There will be a Nano version that works on Android phones.
This means that Google finally has a truly worthy competitor to its OpenAI technology, if we can believe the company itself. But experts are cautious. In fact, GPT and Gemini probably perform very closely.
“The differences between GPT-4 and Gemini Ultra are very small in measurements,” says technology expert Jarno Duursma. “What is striking: Google does not compare its most powerful version with the best version of OpenAI, which is GPT-4 Turbo.”
Gemini Nano werk op Android-telefoons.
Foto: Getty Images
OpenAI can handle more
But Google is clearly entering the fray. That’s nice to see, says researcher Peter van der Putten of Leiden University and the AI lab of the software company BIGA. “It’s OpenAI as David versus Google as Goliath.”
According to Van der Putten, OpenAI, as a newcomer, has much less to lose than Google. OpenAI tests its AI publicly, with bugs visible to everyone. “Google can afford much less than that,” he says. “This is to first test whether the AI works well before launching it. Prejudice and toxic behavior must be eliminated as much as possible, otherwise Google will be held accountable.”
Last week, Google was already punished in the media for its Gemini advertising videos. This was manipulated, so that the results on screen appeared more convincing than they were in reality.
But with the Gemini announcement, Google is hitting the table with a hammer, Duursma says. “They want to show that they are back in racing. And that is true. But you can see that there is pressure behind it.”
ChatGPT: de AI-chatbot van OpenAI.
Foto: Getty Images
Google has brainpower, but it’s a stressful company
Van der Putten says it may be uncomfortable for Google now to not be the market leader and leader for a while. “But we shouldn’t underestimate them. They have the brightest minds, but they also have the hardware and the data. Not just from the search engine, but also from YouTube and all those other services. And most importantly: they have access to a lot of user feedback.” People can broadly indicate what works and what doesn’t. OpenAI has less than that.”
On the other hand, Google is a huge machine because it is a very big company. “It can never be as fast as OpenAI,” Duursma says. “You can see this race as a marathon that sometimes has an obstacle. Google started slowly, fell behind in the first kilometers. It stumbled upon an obstacle here and there. But this also happens at OpenAI, because of the defeat in management.”
It seems as if Google is now at least accelerating the pace to catch up with developments in OpenAI. We’ll have to wait until the Gemini Ultra is released early next year to determine exactly where the companies stand.
Duursma believes Gemini’s multimedia approach—in which it processes images, audio, and video in addition to text—could be particularly useful. “A year from now, Google may look at a faltering start to the generative AI race, but it will look at a golden step with the arrival of Gemini.”
