Google unveiled its much-anticipated AI chatbot Gemini this week, introducing features meant to outperform rival ChatGPT. However, the slick demo shown in Google’s presentation has come under scrutiny, with some claiming parts were faked.
Google Hopes Gemini Can Compete With Runaway ChatGPT Success
On Wednesday, Google CEO Sundar Pichai announced the new experimental conversational AI model Gemini at the company’s Search On 22 event. Gemini is meant to enhance Google’s search engine as more users flock to AI assistants like ChatGPT for information searches.
“We’ve trained it to be helpful, harmless and honest using our advanced language models,” Pichai said of Gemini.
The new model comes from DeepMind, the London AI lab that Google acquired in 2014. Gemini is built on the Pathways language model architecture DeepMind debuted last year.
So far user testing shows Gemini scores 72% on truthfulness and 91% on helpfulness, according to Pichai. Google hopes these strict guidelines around providing accurate information will give Gemini an edge over ChatGPT, which has received criticism for making up answers in areas where its knowledge is incomplete.
“Gemini clearly outperforms ChatGPT and other models,” claimed DeepMind co-founder Demis Hassabis. He said Gemini can explain concepts at a 12th grade level, provide fixes to bugs found in code and discuss complicated topics while maintaining higher factual accuracy than alternatives.
Going forward, Gemini will be integrated into Google Search to enhance results for complex queries in categories like recipes and coding questions.
New Pixel Features Showcase Gemini’s Capabilities
Google also highlighted how Gemini will be put to use in Pixel phones and other applications:
- Pixel Bard uses Gemini to generate text, poetry and code based on conversational prompts
- Pixel audio typing lets Gemini automatically create long-form written transcriptions from audio
- The camera can identify objects in view and provide detailed information on them via Gemini
- Direct translations from over 40 languages
- Automated meeting transcriptions that include smart suggestions for follow-ups
“We plan to bring our most cutting edge AI technology to help people whenever and wherever they need it,” said Pichai.
Other launches included:
- Clear Calling uses AI to reduce background noise on calls
- New photo editing features like removing unwanted people/objects from images
While showcasing Gemini inside Google Search and Pixel devices, Pichai reiterated Google’s commitment to developing AI responsibly. He emphasized extensive testing is still underway.
“We format our research publications to allow independent verification from researchers. We open-source work with care…We consult and collaborate broadly to address the societal implications of AI,” said Pichai.
|Generates text, poetry and code via conversational prompts
|Automatically transcribes long audio clips into written text
|Labels objects seen through the camera and provides detailed info on them
Parts of Gemini Demo Seem Faked, Raising Questions
While the capabilities Google highlighted for Gemini appear extremely impressive, not everything presented may have shown genuine interactions with the AI system.
A slick demo during the launch showed “Gemini” responding to complex multi-part questions on topics ranging from the oldest evidence of life on Earth to interpreting intricate chess puzzles.
However, an investigation by TechCrunch has raised doubts about whether Gemini actually understood and answered several of these questions.
“In some cases, Gemini’s responses were spewed out too quickly — in seconds rather than the minute-plus response times seen in other demos,” wrote TechCrunch. “In other cases, Gemini responded with information not actually covered by the source material it was citing.”
For example, Gemini instantly provided a lengthy summation when asked “What is the evidence supporting the theory that the Chicxulub asteroid impact caused the extinction of dinosaurs?” But the academic excerpt Gemini claimed to summarize did not discuss the dinosaur extinction theory.
This has led to allegations parts of the demo interaction were prearranged rather than showcasing Gemini’s true conversational abilities. While most AI experts agree the core Gemini model is extremely capable, some responses seem too slick not to have some element of scripting.
Google did not confirm or deny whether the demo was edited. A spokesperson simply said the demo was meant to “illustrate Gemini’s capabilities.”
The controversy now threatens to undermine overall confidence in Gemini right as Google positions it as a formidable ChatGPT competitor. It also raises difficult questions around how to truly benchmark AI performance in a way audiences find convincing yet rigorous.
Google Still Chasing Lead Of ChatGPT Juggernaut
The launch of Gemini comes as Google finds itself playing catch-up in the red-hot generative AI space it once dominated.
Rival firm OpenAI has surged ahead over the past year, capturing the world’s attention with creative applications like image generator DALL-E 2 and the conversational ChatGPT.
Since launching in November, ChatGPT has amassed over 1 million users through its uncanny ability to understand natural language queries and provide coherent, human-like responses on virtually any topic.
The overnight success has set off alarms within Google that its lead in AI is slipping. Pichai reportedly pushed for Gemini’s announcement to be moved up as ChatGPT mania grew.
Now Google hopes Gemini can showcase they remain at the forefront of generative AI research and narrow the gap with OpenAI.
Gemini does appear more advanced than GPT-3, the model underlying ChatGPT. And Google has the advantage of funneling Gemini into its ubiquitous search platform used by billions worldwide. This reach gives Gemini’s rollout far more commercial impact out of the gate.
Yet ChatGPT’s basic format has utterly captivated mainstream interest in a way Google may struggle to replicate at first. And OpenAI will likely keep innovating rapidly, with the startup announcing this week they have already developed a more powerful successor to ChatGPT called GPT-4.
So while Gemini looks extremely promising, Google still faces a fierce challenge from OpenAI in shaping the future path of this technology. The race is only just beginning.
What Comes Next for Gemini and Google?
In the near-term, expect Gemini to be fine-tuned before expanding to more users over 2023. Google wants to avoid issues that could erode consumer trust in AI.
Longer term, successfully integrating Gemini across Google’s vast ecosystem could significantly evolve how its services operate. Search becomes an engaging dialogue; YouTube auto-generates video subtitles; Google Docs suggests entire text passages.
“In the coming years we hope to make the benefits of AI accessible to billions of people improving their health, finances, and digital wellbeing,” said Pichai.
But risks around AI bias, safety and regulation remain. Lawmakers have already raised concerns over ChatGPT, and Google will likely face similar scrutiny with Gemini’s real-world impact.
Navigating these tensions between transformative technology and societal obligation will define the future of AI giants like Google. Gemini may boast amazing capabilities, but ultimately what matters most is how Google decides to implement it.
To err is human, but AI does it too. Whilst factual data is used in the production of these articles, the content is written entirely by AI. Double check any facts you intend to rely on with another source.