This week has been filled with surprises from AI underdogs. Google’s Gemini is becoming the public’ favorite while Grok topped the chatbot performance leaderboard. A new medical AI model is helping doctors detect over 1,000 diseases, including cancer.
Meanwhile, ChatGPT is finally taking strides to increase safety among its youngest users by rolling out guardrails for users under-18. Here’s the biggest AI news you need to know this week.
Gemini is public favorite among 27 different AI models
In one of the largest public surveys of its kind, more than 21,000 participants across the U.S. and U.K. ranked 27 different AI models — and Google’s Gemini 2.5 Pro came out on top. The study, conducted via Prolific’s Humaine ranking system, evaluated models on categories like communication style, reasoning, trust and overall user experience.
Gemini beat out competitors including ChatGPT and Claude in most subcategories, underscoring its reputation as Google’s most polished reasoning model yet. ChatGPT ranked eighth overall.
Key points:
- Gemini 2.5 Pro ranked #1 in a 27-model public poll across the U.S. and U.K.
- 21,352 participants from multiple demographics contributed to the results
- Evaluation criteria: communication, fluidity, reasoning, trust and overall user experience
- ChatGPT placed 8th overall; Claude’s versions ranked 11th & 12th
ChatGPT creates ChatGPT features for teens
OpenAI is finally rolling out new safety rules for teen users. The updates include an age-prediction system to detect users under-18 and erring on the side of caution by implementing the guardrails if age is unknown. Stricter filters are promised on sensitive topics (sexual content and self-harm, etc.) in addition to parental controls such as blackout hours and the ability to link accounts.
These changes are a start, but bring up tough questions about how age will be determined, how often mistakes will affect trust and whether teens feel safe opening up if they fear being monitored or judged.
What to know
- New age-detection rule: OpenAI will route users it determines are under 18 to a version of ChatGPT with stricter safety rules.
- Default under-age mode: If age can’t be confirmed, the system errs on the side of caution by applying under-age restrictions anyway.
- Parental controls introduced: Parents can link to their teen’s account, disable or limit features (like memory or history), define safe hours (blackout hours) and get alerted if the AI detects signs of distress.
- Conversation limits for teens: Graphic sexual content, flirtatious chat, self-harm discussions are restricted. In extreme, imminent-harm cases, law enforcement contact is a possibility if parents are unreachable.
Custom Gemini Gems are shareable now
Google is expanding the reach of Gemini, by giving users the ability to now share their custom assistants, Gems. For those more familiar with ChatGPT, these are essentially custom GPTs. Before now, Gemini Gems were locked into private, personal use, but now they can be shared as easily as a Google Doc, letting coworkers, classmates and anyone else collaborate with a tailored AI, without building it from scratch.
This move underscores Google’s push to make Gemini a far more flexible and user-friendly platform.
What to know:
- Gems are customizable AI assistants you can design for specific tasks like coding, editing or brainstorming.
- Sharing Gems works in the same was as everything else on Google Drive. When collaborating, others can view or edit a Gem.
- Now available to Gemini Advanced, Business and Enterprise subscribers in over 150 countries.
Grok tops the ARC-AGI leaderboard
Elon Musk’s Grok 4 has surged to the front of the pack in the AI arms race, topping the ARC-AGI leaderboard — a benchmark that measures how many problems an AI can solve as well as the efficiency of the model. In recent testing, Grok 4 outperformed rivals like ChatGPT and Gemini on this tough metric, highlighting it’s superior real-world problem solving and speed.
Despite its speed, use of live web search and impressive handling of complex engineering prompts, users still have concerns around Grok’s accuracy, content moderation and bias.
What to know:
- ARC-AGI leaderboard leader: Grok 4 claims the top spot, meaning it solves more AI benchmark challenges per resource unit than competitors.
- Strengths: near-instant web search, improved handling of engineering and complex reasoning tasks and performance gains compared to previous Grok versions.
- Weaknesses & concerns: issues with content accuracy (some false claims), questions of bias and the challenge of maintaining moderation and reliability under heavier usage.
- Implication: Demonstrates that efficiency and quality are increasingly rewarded in AI benchmarks.
New AI model can predict your risk of 1,000+ diseases, including cancer
A new AI model called Delphi-2M represents a leap forward in medical AI. Trained on anonymized health data from nearly 2.3 million people in the UK and Denmark, it can forecast your risk of over 1,000 diseases and estimate when they might occur.
Unlike other AI models, Delphi-2M gives you a risk score for 1,0000 conditions, using disease histories, lifestyle factors, age and sex to simulate health trajectories over decades. The model can predict not just what disease might come, but when.
In trials, it achieved an average AUC of 0.76 across hundreds of diseases in UK datasets, although accuracy dropped in Danish populations. The model is not a diagnostic tool and not a replacement for human doctors.
But the tool is promising as forecasting tool, useful for spotting general risk trends, planning prevention and illuminating the potential for AI in personalized health.
Key points:
- Massive dataset: Trained on ~2.3 million people from UK Biobank + Danish registries.
- Holistic + temporal modeling: Uses detailed inputs (age, sex, lifestyle, past diagnoses) and predicts both the next disease and when it will occur.
- Forecasting, not diagnosing: Intended to inform prevention, not replace medical advice.
The takeaway
In just a week, these updates highlight a clear shift in the AI landscape towards real-world integration and bigger breakthroughs.
From OpenAI’s teen protections to Google’s shareable Gems, from Grok topping performance benchmarks to Delphi-2M predicting disease risk, AI is becoming my personal, practical and accelerating fast. At Tom’s Guide, we’re tracking every shift so you don’t miss a beat.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button!
More from Tom's Guide
- I’m planning on switching to Gemini after using ChatGPT as my daily chatbot for years — here’s why
- Chatbots don’t care about punctuation — here’s what happened when I tested it
- ChatGPT's Voice mode is 'back' yet still incredibly cringe — here's why
![HIDevolution [2025] ASUS ROG...](https://images.fie.futurecdn.net/products/60cbb9059616003fe7353df7bc0e56e02dd01b3c-100-80.jpg.webp "HIDevolution [2025] ASUS ROG...")
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
