Archive 26.03.2025

A new study shows that people in Japan treat robots and AI agents more respectfully than people in Western societies.

Imagine an automated delivery vehicle rushing to complete a grocery drop-off while you are hurrying to meet friends for a long-awaited dinner. At a busy intersection, you both arrive at the same time. Do you slow down to give it space as it maneuvers around a corner? Or do you expect it to stop and let you pass, even if normal traffic etiquette suggests it should go first?

Strategy outlines key policy priorities, including establishing a central Robotics Office and Robotics Commission, to strengthen U.S. competitiveness in robotics, drive innovation, and support workforce development

From virtual reality to rehabilitation and communication, haptic technology has revolutionized the way humans interact with the digital world. While early haptic devices focused on single-sensory cues like vibration-based notifications, modern advancements have paved the way for multisensory haptic devices that integrate various forms of touch-based feedback, including vibration, skin stretch, pressure and temperature. Recently, a team of experts analyzed the current state of wearable multisensory haptic technology, outlining its challenges, advancements and real-world applications.

Gemini 2.5 is our most intelligent AI model, now with thinking built in.

This a robot can walk, without electronics, and only with the addition of a cartridge of compressed gas, right off the 3D-printer. It can also be printed in one go, from one material.

Imagine a robot that can walk, without electronics, and only with the addition of a cartridge of compressed gas, right off the 3D-printer. It can also be printed in one go, from one material.

As President Trump's second term unfolds, the manufacturing landscape in the United States is poised for a significant transformation. The administration's renewed focus on reshoring manufacturing operations presents both opportunities and challenges for industry leaders.

We deployed 100 reinforcement learning (RL)-controlled cars into rush-hour highway traffic to smooth congestion and reduce fuel consumption for everyone. Our goal is to tackle "stop-and-go" waves, those frustrating slowdowns and speedups that usually have no clear cause but lead to congestion and significant energy waste. To train efficient flow-smoothing controllers, we built fast, data-driven simulations that RL agents interact with, learning to maximize energy efficiency while maintaining throughput and operating safely around human drivers.

Overall, a small proportion of well-controlled autonomous vehicles (AVs) is enough to significantly improve traffic flow and fuel efficiency for all drivers on the road. Moreover, the trained controllers are designed to be deployable on most modern vehicles, operating in a decentralized manner and relying on standard radar sensors. In our latest paper, we explore the challenges of deploying RL controllers on a large-scale, from simulation to the field, during this 100-car experiment.

Read More

How many robots does it take to screw in a lightbulb? The answer is more complicated than you might think. New research from Northeastern University upends the riddle by making a robot that is both flexible and sensitive enough to handle the lightbulb, and strong enough to apply the necessary torque.

A joint research team has successfully developed a next-generation soft robot based on liquid. The research was published in Science Advances.

NGen, Festo Didactic and Humber Polytechnic, a leading Canadian post-secondary institution known for its career-focused, hands-on approach to education, are partnering to demonstrate the strength of Canadian mechatronics skills at Hannover Messe.

New Study Finds AI Popular Among Less-Educated

New research from Stanford University reveals that ChatGPT and similar AI writers are surprisingly popular among those with less formal education.

Essentially, researchers found that regions in the U.S. featuring more tradespeople, artisans, craftsmen and similar are using AI writing more than people living in areas where college degrees are more prevalent.

The telling stats: 19.9% of people living in ‘less educated’ areas of the U.S. have adopted AI writing tools like ChatGPT – as compared to 17.4% in regions with higher education profiles.

Even more dramatic: Adoption in the state of Arkansas, where college degrees are less prevalent: A full 30% of people in Arkansas are using ChatGPT and similar AI to auto-write letters to businesses and government organizations.

In other news and analysis on AI writing:

*Microsoft’s ChatGPT Competitor – Copilot – Gets an Upgrade: Microsoft has rolled-out a new version of its AI writer/chatbot Copilot, which it says is now more deeply embedded into its Windows software.

In part, the change was made in response to user complaints over previous versions of Copilot, which they say operated more like a ‘wrapper’ or outside app that ‘felt’ only weakly linked to Windows software.

With the upgrade, Microsoft is promising users will see marked performance gains from Copilot.

*ChatGPT Competitor Claude: Great for Auto-Writing Pre-Meeting Reports: Mike Krieger, chief product officer, Anthropic is pushing a new use case for the company’s ChatGPT-competitor, Claude.

Essentially, the AI tech can be used to scan calendars and company data to auto-write detailed client reports before a meeting, according to Krieger.

Observes writer Muslim Farooque: “With this move, Anthropic is taking on big players like Microsoft, OpenAI, and Google — all racing to dominate AI-powered business tools.

*One Writer’s Take: Google Has the Best AI Writing Editor: Count writer Amanda Caswell is among those who strongly prefer Google’s new editor for AI writing – Canvas – over ChatGPT’s online editor that carries the same name.

Observes Caswell: “Gemini Canvas is far more thorough and detailed in its critique than ChatGPT Canvas. It’s essentially a real editor. ChatGPT made me feel like my mom was editing the story and was sparing my feelings.

“In a word: Wow.”

*College Rolling-out New Certificate in AI Writing: Beginning Fall 2025, students at Boise State College can obtain a certificate in AI writing after completing three courses on the discipline.

Those are:

~Writing For/With AI

~Applications of AI (with a strong focus on content production)

~Style and the Future of AI Writing

*AI Tech Titans Want to Use Copyrighted Writing for Free: ChatGPT-maker OpenAI – and Google – are looking for clearance from the U.S. government to train their AI on newspaper, magazine and other copyrighted text on the Web for free.

The reason: Given China’s recent major gains in tightening-up the AI race, U.S. AI purveyors need every advantage to stay ahead of China.

Currently, many content creators – including The New York Times – are suing OpenAI for using their content to train ChatGPT without permission.

*On the Research Bench: Text-To-Data-Driven Slides: Adobe is currently experimenting with new AI tech that promises to convert data-heavy research into vibrant slide presentations in Powerpoint.

Dubbed ‘Project Slide Wow,’ the experimental tech is aimed at marketers and business analysts looking to quickly build data-backed presentations without being forced to manually structure content or design slides.

Observes Jane Hoffswell, research scientist, Adobe: “It’s analyzing all the charts in this project, generating captions for them, organizing them into a narrative and creating the presentation slides.”

Currently, Adobe has no firm release date for the experimental slide-maker.

*ChatGPT-Maker’s AI Agents: The Complete Rundown: Writer Siddhese Bawker offers an excellent overview in this piece on the tiers of AI agents currently available from OpenAI.

Such agents are able to work independently on a task for you, which might include clicking-and-pointing with your browser to research, analyze and then auto-write on what it found.

Even better: Extremely advanced AI agents are able to perform such tasks with PhD-level intelligence.

OpenAI’s entry-level agent is included in a ChatGPT Pro subscription ($200/month.)

Higher level agents are OpenAI’s Knowledge Worker Agent ($200/month), Developer Agent ($10,000/month) and Research Agent ($20,000/month).

*ChatGPT Wants to be the Interface for Your Data: Businesses hoping to integrate their databases with ChatGPT — so they can use the AI to analyze and auto-write reports about that data and more — may not have to wait long.

Writer Kyle Wiggers reports that OpenAI is currently testing in-house developed ‘connectors’ that will ideally make such fusions possible.

So far, development of connectors to Google Drive and Slack is already underway.

Observes Wiggers: “ChatGPT Connectors will allow ChatGPT Team subscribers to link workspace Google Drive and Slack accounts to ChatGPT so the chatbot can answer questions informed by files, presentations, spreadsheets and Slack conversations.”

*AI BIG PICTURE: New Hyper-Realistic Voice AI Goes Viral: A new AI voice sensation – Sesame AI – appears ready to dethrone Eleven Labs as the industry standard in realistic voice AI.

Essentially, the Web has blown-up with praise for Sesame AI, which apparently generates AI voices that are so real and human, their sheer intimacy disturbs some people.

Even so: AI Uncovered – producer of this 11-minute video – does note that Eleven Labs still beats Sesame AI when it comes to auto-generating spoken word from a script.

Share a Link:  Please consider sharing a link to https://RobotWritersAI.com from your blog, social media post, publication or emails. More links leading to RobotWritersAI.com helps everyone interested in AI-generated writing.

–Joe Dysart is editor of RobotWritersAI.com and a tech journalist with 20+ years experience. His work has appeared in 150+ publications, including The New York Times and the Financial Times of London.

The post ChatGPT: The Great Equalizer appeared first on Robot Writers AI.

More than 11 years after the disappearance of Malaysia Airlines flight MH370, the Malaysian government has approved a new search for the missing debris of the aircraft.

Large Language Models (LLMs) have come a long way since their early days of mimicking autocomplete on steroids. But generating fluent text isn’t enough – true intelligence demands reasoning. That means solving math problems, debugging code, drawing logical conclusions, and even reflecting on errors. Yet modern LLMs are trained to predict the next word, not to think. So how are they suddenly getting better at reasoning?

The answer lies in a constellation of new techniques – from prompt engineering to agentic tool use – that nudge, coach, or transform LLMs into more methodical thinkers. Here’s a look at five of the most influential strategies pushing reasoning LLMs into new territory.

1. Chain-of-Thought Prompting: Teaching LLMs to “Think Step by Step”

One of the earliest and most enduring techniques to improve reasoning in LLMs is surprisingly simple: ask the model to explain itself.

Known as Chain-of-Thought (CoT) prompting, this method involves guiding the model to produce intermediate reasoning steps before giving a final answer. For instance, instead of asking “What’s 17 times 24?”, you prompt the model with “Let’s think step by step,” leading it to break down the problem: 17 × 24 = (20 × 17) + (4 × 17), and so on.

This idea, first formalized in 2022, remains foundational. OpenAI’s o1 model was trained to “think longer before answering” – essentially internalizing CoT-like reasoning chains. Its successor, o3, takes this further with simulated reasoning, pausing mid-inference to reflect and refine responses.

The principle is simple: by forcing intermediate steps, models avoid jumping to conclusions and better handle multi-step logic.

2. Inference-Time Compute Scaling: More Thinking per Question

If a question is hard, spend more time thinking – humans do this, and now LLMs can too.

Inference-time compute scaling boosts reasoning by allocating more compute during generation. Instead of a single output pass, models might generate multiple reasoning paths, then vote on the best one. This “self-consistency” method has become standard across reasoning benchmarks.

OpenAI’s o3-mini uses three reasoning effort options (low, medium, high) that determine how long the model reasons internally before answering. At high reasoning levels, o3-mini outperforms even the full o1 model on math and coding tasks.

A related technique, budget forcing, introduced in the 2025 paper s1: Simple Test-Time Scaling, uses special tokens to control reasoning depth. By appending repeated “Wait” tokens, the model is nudged to generate longer responses, self-verify, and correct itself. An end-of-thinking token like “Final Answer:” signals when to stop. This method improves accuracy by extending inference without modifying model weights – a modern upgrade to classic “think step by step” prompting.

The tradeoff is latency for accuracy, and for tough tasks, it’s often worth it.

3. Reinforcement Learning and Multi-Stage Training: Rewarding Good Reasoning

Another game-changer: don’t just predict words – reward correct logic.

Models like OpenAI’s o1 and DeepSeek-R1 are trained with reinforcement learning (RL) to encourage sound reasoning patterns. Instead of just imitating data, these models get rewards for producing logical multi-step answers. DeepSeek-R1’s first iteration, R1-Zero, used only RL – no supervised fine-tuning – and developed surprisingly powerful reasoning behaviors.

However, RL-only training led to issues like language instability. The final DeepSeek-R1 used multi-stage training: RL for reasoning and supervised fine-tuning for better readability. Similarly, Alibaba’s QwQ-32B combined a strong base model with continuous RL scaling to achieve elite performance in math and code.

The result? Models that not only get answers right, but do so for the right reasons – and can even learn to self-correct.

4. Self-Correction and Backtracking: Reasoning, Then Rewinding

What happens when the model makes a mistake? Can it catch itself?

Until recently, LLMs struggled with self-correction. In 2023, researchers found that simply asking a model to “try again” rarely improved the answer – and sometimes made it worse. But new work in 2025 introduces backtracking – a classic AI strategy now adapted to LLMs.

Wang et al. from Tencent AI Lab identified an “underthinking” issue in o1-style models: they jump between ideas instead of sticking with a line of reasoning. Their decoding strategy penalized thought-switching, encouraging deeper exploration of each idea.

Meanwhile, Yang et al. proposed self-backtracking – letting the model rewind when stuck, then explore alternate paths. This led to >40% accuracy improvements compared to approaches that solely relies on the optimal reasoning solutions.

These innovations effectively add search and planning capabilities at inference time, echoing classical AI methods like depth-first search, layered atop the flexible power of LLMs.

5. Tool Use and External Knowledge Integration: Reasoning Beyond the Model

Sometimes, reasoning means knowing when to ask for help.

Modern LLMs increasingly invoke external tools – calculators, code interpreters, APIs, even web search – to handle complex queries. 

Alibaba’s QwQ-32B incorporates agent capabilities directly, letting it call functions or access APIs during inference. Google’s Gemini 2.0 (Flash Thinking) supports similar features – for example, it can enable code execution during inference, allowing the model to run and evaluate code as part of its reasoning process.

Why does this matter? Some tasks – like verifying real-time data, performing symbolic math, or executing code – are beyond the model’s internal capabilities. Offloading these subtasks lets the LLM focus on higher-order logic, dramatically improving accuracy and reliability.

In essence, tools let LLMs punch above their weight – like a digital Swiss Army knife, extending reasoning with precision instruments.

Conclusion: Reasoning Is a Stack, Not a Switch

LLMs don’t just “learn to reason” in one step – they acquire it through a layered set of techniques that span training, prompting, inference, and interaction with the world. CoT prompting adds structure. Inference-time scaling adds depth. RL adds alignment. Backtracking adds self-awareness. Tool use adds reach.

Top-performing models like OpenAI’s o1 and o3, DeepSeek’s R1, Google’s Gemini 2.0 Flash Thinking, and Alibaba’s QwQ combine several of these strategies – a hybrid playbook blending clever engineering with cognitive scaffolding.

As the field evolves, expect even tighter coupling between internal reasoning processes and external decision-making tools. We’re inching closer to LLMs that don’t just guess the next word – but genuinely think.

Enjoy this article? Sign up for more AI updates.

We’ll let you know when we release more summary articles like this one.

The post How Do LLMs Reason? 5 Approaches Powering the Next Generation of AI appeared first on TOPBOTS.