Google’s Gemini chases a one-year-old

A video that popped up on Instagram recently showed a young family getting excited when their toddler takes its first steps. The mom jumps up and down in excitement, and the dad scoops up the baby in joy. While not a perfect parallel, it reminded me of the excitement that ChatGPT generated when it made its debut last November.

Compare it to the unveiling of Gemini by Google. It was more like a grown-up adult trying to impress someone by standing on one leg. And you have to admit any one-year-old toddler is more adorable than a grown-up. 

Since the arrival of ChatGPT, the chatbot's capabilities have continued to impress everyone, including the people who worked on it. According to a New York Times report, insiders at OpenAI had expected a few thousand users to try the chatbot, but it attracted a million users in just five days.

The rapid progression of artificial intelligence (AI) after that left Google flat-footed, despite the tech giant having worked on such a platform for years. As ChatGPT was perceived as an existential threat to the internet search engine giant, Google hurriedly changed tack and declared an internal code red. 

After a few initial stumbles, they finally unveiled what was dubbed the ‘ChatGPT killer’ last week, showcasing impressive feats and stats in a video. However, the tech world appeared more disappointed than impressed, and many reviewers quickly posted unflattering responses. 

The company seemed to have shot itself in the foot when it admitted that its demo video was not one hundred percent accurate. This gave the impression that, in their rush to catch up with OpenAI, Google was more focused on speed than content.

More disappointment was in store as Google said the powerful version of its chatbot, Gemini Ultra, was slated for rollout only next year. The two versions it introduced, Gemini Nano and Gemini Ultra, failed to capture the imagination of many users.

One reason for this may be the constant stream of new AI capabilities that has been hitting us almost daily.  This had sent expectations soaring, especially for Google's products, given the company’s substantial resources. But the rollout simply lacked a “WOW” moment, despite showcasing some impressive features.

However, tech analyst Alberto Romero suggests that critics of Google may not be entirely objective in their assessments. “Whether you think Gemini is a success or not depends mainly on whether you like Google or not,” he wrote.

His sentiment was echoed by Trivandrum-based Brijesh Madhavan, who founded CurveLogics, a firm specialising in AI solutions. 

“Open AI also faced some criticism when ChatGPT was released. The backlash primarily came from analysts who nitpicked minor flaws and exaggerated them. However, those familiar with the field were appreciative,” he says.

One noteworthy feature in the demo showcased Gemini's ability to sift through 20,000 research papers, identify the most relevant ones, and generate a summary within one hour. The researchers who watched this demo probably did jump with excitement, like the mom on that Instagram video. 

The potential scientific advancements achievable by harnessing such capabilities are almost limitless. To provide context, just days before the unveiling of Gemini, the AI research firm DeepMind, recently acquired by Google, announced that its researchers had successfully created 700 new materials in the lab and outlined structures for another 2.2 million new materials. 

These capabilities hold the promise of bringing elusive goals, such as achieving superconductivity or finding a cure for cancer, closer to reality.

While the potential impacts are significant, the reality on the ground is that the race is focused on securing a major share of the business AI is poised to generate. Google is striving to catch up with OpenAI, and Elon Musk's ChatGPT rival, Grok, has been released to X Premium users. Meanwhile, Meta has added its own picture-generating feature.

In China, Alibaba unveiled a feature that enabled users of its chatbot to animate any photos. (One of the early comments in GitHub about this was “can’t wait to use this for porn".) While the big names grab the headlines with their moves, less famous firms like Mistral AI and Extropic AI have also been releasing powerful programs without much fanfare.

The breakneck speed at which AI developments are occurring poses the risk of going off the rails. It also highlights the inherent biases and flaws in AI programs, which are trained on available digital data. This has resulted in AI chatbots drawing inaccurate and questionable conclusions, a phenomenon already present on the internet. Experts warn that this could significantly impact future research.

Furthermore, as the available data is predominantly in English, it fails to present a comprehensive picture of reality. The dominance of one language in data is not only detrimental to research but could also lead to the marginalisation of diverse perspectives.

This concern is not only pertinent from a global perspective but also in diverse societies like India, where powerful entities are shaping new narratives across domains from science to history. The lack of verified and digitised data has been a drawback in India, although efforts are underway to overcome this disadvantage.

Recent reports revealed that Bangalore-based Sarvam AI has successfully raised funding worth 41 million US dollars. Co-founded by Vivek Raghavan and Pratyush Kumar, the firm 'aims to focus on India’s unique needs, which include training AI models to support various Indian languages and voice-first interfaces,' reports Your Story.

The project, backed by influential tech financiers like Khosla Ventures, is a welcome step forward, and more such efforts are needed as very little data in regional languages is digitised. 

Niyas Mohammed, an expert in machine learning and co-founder of Trivandrum-based NeuralCraft, says that is a big challenge. “It's impossible to amass the volume of info used to train ChatGPT for any of our Indian languages – even eBooks are still not widespread for most publishers.”

“Without enough data to capture nuances, up-to-date facts, and a diversity of information, we need to explore creative, new ways to make this happen. But I am really excited about  what companies like Sarvam are going to build.”

A missed bus to UAE

Despite the significant presence of Keralites in the Gulf countries, it seems that Kerala startups missed an opportunity to fully capitalise on the opportunities during the 28th annual United Nations climate meeting (COP28). Reports suggest the conclave was dominated by firms from European countries and Asian nations like Japan and Singapore.

Despite the thin presence, two startups with links to Kerala  managed to attract some attention. Cochin-based Zara Biotech was the only Indian firm among 22 companies shortlisted for product pitching for their algae-based decarbonisation product. Singapore-headquartered mistEO was one of the five finalists for the IOT stream of the COP28 UAE Techsprint. The company has a presence in Trivandrum and its CEO Samuel John grew up in the city.

Startup investor strikes it rich

Investing in startups is a high-risk, high-return game. Nothing illustrates this more than what Fireside Ventures pulled off with their investment in Mamaearth. The startup successfully grew as a seller of organic products for babies and moms and its IPO in early November was received warmly by the market.  Your Story reports that the Bengaluru-based Fireside became an early investor in the startup in 2017 and bought shares at 7.33 rupees. It sold 60.88 lakh shares at 378 rupees per share recently, booking a profit of more than 4,500 percent.

GTA trailer sets YouTube on fire

The first trailer for the new Grand Theft Auto (GTA) version set a world record with 93 million views on YouTube within 24 hours. Rockstar Games released the trailer a day in advance after a low-quality version was leaked online. The previous GTA version was released in 2013, and fans have eagerly anticipated a newer version for years. The multiplayer game has sold 190 million units since then, making it the third-best-selling video game of all time, behind Tetris and Minecraft, according to Forbes.

Time for all dogs to upskill 

Teaching old dogs new tricks is difficult, even when the dogs are robots. But Swiss robotics firm Swiss-Mile has trained their ANYmal to stand on two wheels and perform useful tasks like opening doors and sorting boxes, as reported by Interesting Engineering. It even mimicked humans when while tossing boxes away, as this demo video shows. Here’s a friendly warning to man’s best friend: you better update your skills, or your jobs might be gone too.