Google's Gemini chatbot may have a Pokemon game 'problem'
Google's Gemini and other AI chatbots may have a
"problem". A new research indicates that these AI models can exhibit irregular behaviour, like "panic," when confronted with challenges in Pokemon games.
Tired of too many ads? go ad free now
According to a report by
DeepMind, Gemini 2.5 Pro experiences "qualitatively observable degradation in the model's reasoning capability" when its Pokemon are close to defeat. This observation comes as AI companies like Google and Anthropic are studying how their latest AI models navigate early Pokemon games. Researchers believe that observing AI models playing video games can provide useful insights into their capabilities.
How Google's Gemini and other chatbots reacted to older Pokemon games
In recent months, independent developers have launched
Twitch streams
like 'Gemini Plays Pokemon' and 'Claude Plays Pokemon,' showcasing AI models playing the classic game in real time, the report mentions. This offers an alternative, more contextual way to benchmark AI performance beyond traditional testing methods.
Each stream reveals how the AI reasons through problems, offering insight into its decision-making process.
While these models have advanced quickly, they still struggle with tasks like playing Pokemon efficiently, often taking hundreds of hours to finish. The real intrigue lies in observing the AI's behaviour and choices during gameplay, rather than its speed.
'Over the course of the playthrough, Gemini 2.5 Pro gets into various situations which cause the model to simulate 'panic,''
the report noted.
This means, during gameplay, the AI can enter a state of 'panic,' where its performance declines and it stops using some available tools.
Tired of too many ads? go ad free now
While the AI doesn't experience emotions, this behaviour resembles how humans may make poor decisions under stress, making it both intriguing and unsettling.
'This behaviour has occurred in enough separate instances that the members of the Twitch chat have actively noticed when it is occurring,'
the report added.
Apart from Gemini, Claude has also shown unusual behaviour during the gameplay. At one point, it wrongly assumed that fainting all its Pokemon would transport it forward in the game, leading it to intentionally lose battles, which is a strategy that backfired, as it was sent to the last used Pokemon Center instead.
Despite such errors, the AI has excelled at solving in-game puzzles. With some human help, it used task-specific AI tools to navigate boulder puzzles and plan efficient routes.
'With only a prompt describing boulder physics and a description of how to verify a valid path, Gemini 2.5 Pro is able to one-shot some of these complex boulder puzzles, which are required to progress through Victory Road,' t
he report highlighted.
The report also suggests that since Gemini 2.5 Pro independently created many of its tools, upcoming versions may be able to do so without human help, potentially even developing a 'don't panic' module on its own.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Hans India
an hour ago
- Hans India
Amazon CEO Andy Jassy warns AI will shrink corporate jobs
Amazon CEO Andy Jassy has told employees that artificial intelligence (AI) will significantly reduce the company's corporate workforce in the coming years, emphasizing the need for staff to embrace the technology. In a memo to employees on Tuesday, Jassy urged teams to "be curious about AI" as the company integrates the technology "in virtually every corner of the business." He acknowledged that while AI will eliminate certain roles, it will also create new opportunities requiring different skills. 'We will need fewer people doing some of the jobs that are being done today, and more people doing other types of jobs,' he wrote. 'It's hard to know exactly where this nets out over time, but in the next few years, we expect that this will reduce our total corporate workforce.' Amazon, which employed over 1.5 million people globally at the end of last year—including approximately 350,000 in corporate roles—has joined a growing list of tech giants ramping up AI use. Jassy highlighted that half a million sellers on Amazon's platform already use AI tools to manage product listings, and advertisers are increasingly leveraging its AI offerings. As AI becomes more capable of handling routine tasks, Jassy expects new kinds of digital agents to emerge that can perform shopping, scheduling, and other day-to-day responsibilities. 'Many of these agents have yet to be built, but make no mistake, they're coming and coming fast,' he added. The statement comes amid widespread industry concern that generative AI—now able to write code, produce content, and analyze data—could displace millions of entry-level and white-collar jobs. Prominent figures like Anthropic CEO Dario Amodei and AI pioneer Geoffrey Hinton have warned that AI may permanently eliminate large swaths of traditional employment. Jassy concluded by stating that those who adapt to AI will be "well-positioned" at Amazon, but left no doubt that the corporate workforce is set to become leaner as AI adoption accelerates.
Mint
an hour ago
- Mint
Google Pixel 10 series may feature dual speakers and SIM tray shift, leak hints
Google's next-generation flagship smartphone lineup, expected to include the Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, and the foldable Pixel 10 Pro Fold, has been the subject of numerous leaks in recent weeks. Now, a newly surfaced report offers fresh insights into the design of the Pixel 10 and Pixel 10 Pro—this time via an alleged protective case leak. According to Android Authority, the protective case comes from Thinborne, a well-known accessory maker. The company has reportedly developed cases for the Pixel 10 and Pixel 10 Pro ahead of their official unveiling. The leaked case design suggests that while the upcoming phones may appear similar to last year's Pixel 9 and Pixel 9 Pro, there are subtle shifts in dimensions and hardware placements that indicate noteworthy updates. The report claims that the Pixel 9 Pro can fit snugly into the case made for the Pixel 10, hinting at comparable sizing between the generations. However, the alignment of key elements such as port cutouts and the camera bar tells a different story. Visuals of the leaked case suggest that the USB-C port and other bottom-edge components have been repositioned. Unlike the single-speaker cutout seen in previous models, the Pixel 10 and Pixel 10 Pro are expected to feature dual-speaker cutouts on the underside of the device. Additionally, the SIM tray, traditionally located on the side or bottom, appears to have been moved to the top edge in the new design. There are further indications that the rear camera system has been slightly revamped. The case does not align perfectly with the camera bar of the Pixel 9 Pro, leading to speculation that the Pixel 10's camera module will be marginally larger. The case also partially obscures the 5G antenna on the older model, reinforcing the theory that external hardware elements have shifted enough to warrant a redesign. These changes, though subtle, could mean existing Pixel 9 cases will not be compatible with the Pixel 10, despite the superficial resemblance between the two generations.
The Hindu
2 hours ago
- The Hindu
Google Veo 3: Creative Breakthrough or Crisis for Journalism?
Published : Jun 18, 2025 18:51 IST - 5 MINS READ Launched in May 2025 at Google's annual I/O developer conference, Google Veo 3 is the tech giant's direct challenge to Microsoft-backed OpenAI's video generation model, Sora. Developed by Google DeepMind, the advanced model marks a major leap in generative AI, promising high-quality, realistic video creation from text or image prompts. But in an age flooded with misinformation and deepfakes, a tool like Veo 3—with its ability to produce lifelike video and synchronised audio—raises pressing questions for journalism. It opens new creative possibilities, yes, but also invites serious challenges around credibility, misuse, and editorial control. What is Google Veo? Veo 3 touts itself as a 'cutting-edge tool' offering 'unmatched realism, audio integration, and creative control'. It comes at a high price—$249.99/month under the AI Ultra plan—and is currently available in the US and 71 other countries, excluding India, the EU, and the UK. Ethical concerns loom, but Google pitches Veo as a powerful resource for filmmakers, marketers, and developers. According to Google, Veo 3 can generate 4K videos with realistic physics, human expressions, and cinematic style. Unlike many competitors, it also produces synchronised audio—dialogue, ambient noise, background music—adding to the illusion of realism. Also Read | When AI breaks the law, who gets arrested—the bot or its maker? The model is designed to follow complex prompts with precision, capturing detailed scenes, moods, and camera movements. Users can specify cinematic techniques like drone shots or close-ups, and control framing, transitions, and object movement. A feature called 'Ingredients' allows users to generate individual elements—like characters or props—and combine them into coherent scenes. Veo can also extend scenes beyond the frame, modify objects, and maintain visual consistency with shadows and spatial logic. Google's website features examples of Veo in action, including projects in marketing, social media, and enterprise applications. The Oscar-nominated filmmaker Darren Aronofsky used it to create a short film, Primordial Soup. On social media, AI artists have released viral Veo clips like Influenders, a satire featuring influencers at the end of the world. Veo 3 is integrated into Google's AI filmmaking tool Flow, which allows intuitive prompting. Enterprise access is available via Vertex AI, while general users in supported countries can use it through Google's Gemini chatbot. The journalism dilemma Veo's features raise alarms about potential misuse. It could facilitate the creation of deepfakes and false narratives, further eroding trust in online content. There are also broader concerns about its economic impact on creators, legal liabilities, and the need for stronger regulation. The risks are not theoretical. As highlighted in a June 2025 TIME article, titled 'Google's Veo 3 Can Make Deepfakes of Riots, Election Fraud, Conflict', Veo was used to generate realistic footage of fabricated events—like a mob torching a temple or an election official shredding ballots—paired with false captions designed to incite unrest. Such videos could spread rapidly, with real-world consequences. Cybersecurity threats—like impersonating executives to steal data—are also plausible, alongside looming copyright issues. TIME reported that Veo may have been trained on copyrighted material, exposing Google to lawsuits. Meanwhile, Reddit forums cite personal harms, such as a student jailed after AI-generated images were falsely attributed to them. There is also the threat to livelihoods. AI-generated content could displace human creators, particularly YouTubers and freelance editors, accelerating what some call the 'dead internet'—a space overrun by AI-generated junk media. To mitigate risk, Google claims that all Veo content includes an invisible SynthID watermark, with a visible one in most videos (though it can be cropped or altered). A detection tool for SynthID is in testing. Harmful or misleading prompts are blocked, but troubling content has still emerged, highlighting the limits of guardrails. What should newsrooms do? Despite the risks, Veo presents compelling opportunities for journalism—particularly for data visualisation, explainer videos, recreating historical events, or reporting on under-documented stories. It can help small newsrooms produce professional-quality videos quickly and affordably, even for breaking news. Used responsibly, Veo could improve storytelling—turning eyewitness accounts of a disaster into a visual narrative, for instance, or transforming dry data into cinematic sequences. Prototyping ideas before committing to full production becomes more feasible, especially for digital-first outlets. But Veo's strengths are also its dangers. Its ability to produce convincing footage of events that never happened could destabilise the information ecosystem. If deepfakes flood the news cycle, real footage may lose credibility. The visible watermark is easily removed, and Google's SynthID Detector remains limited in scope, giving malicious actors room to operate undetected. To maintain public trust, newsrooms must clearly disclose when content is AI-generated. Yet the temptation to pass off fabricated visuals as real—especially in competitive, high-pressure news environments—will be strong. And because AI outputs reflect their training data, biases could sneak in, requiring rigorous editorial scrutiny. There is also the human cost. Veo's automation could eliminate roles for video editors, animators, and field videographers, especially in resource-strapped newsrooms. Journalists may need to learn prompt engineering and AI verification just to stay afloat. Also Read | AI is changing work, privacy, and power—what comes next? The legal landscape is murky. If an outlet publishes an AI-generated video that causes harm, accountability is unclear. Ownership of Veo-generated content also remains opaque, raising potential copyright disputes. And then there is the burden of verification. Fact-checkers will face a deluge of synthetic content, while reporters may find their own footage treated with suspicion. As the Pew Research Center reported in 2024, three in five American adults were already uneasy about AI in the newsroom. A critical juncture As Veo and tools like it become cheaper and more widely available, their impact on journalism will deepen. The challenge is not simply to resist the tide but to adapt—ethically, strategically, and urgently. According to experts, newsrooms must invest in training, transparency, and detection tools to reap the creative rewards of AI while safeguarding credibility. Innovation and trust must evolve together. If journalism is to survive this next phase of disruption, it must do so with eyes wide open, they say. (Research by Abhinav Chakraborty)
