A preprint is a signal, not a finished product or an independently confirmed result. We therefore track research papers separately and show the actual state of supporting evidence for each one.
1381
published research events
Latest work
A significant claim from a single source is published only after further confirmation.
The study introduces Progress-Aware Distillation, a method that penalizes agents for stagnation during training. The authors report higher success rates and greater efficiency across eight models with 0.6 to 8 billion parameters; the available abstract does not specify the size of the improvement.
The MemDuet framework pairs a small model for selecting source material with a large model for verifying and updating memory. Across three datasets, the authors report an average of 9.74× faster processing and 73.91 % fewer output tokens compared with RL-MemAgent.
The authors present FACES, a method for training multimodal emotional support systems. With response length controlled, they report an increase in automated evaluation scores from 3.61 to 4.04 and a preference rate of 75.3 % in human evaluation over the comparison systems.
In an interview about the new book “The Collision: What AI Does to Us”, Professor Eric So argues that relying too heavily on AI may hinder the development of thinking skills. He cites research on learning; however, the available excerpt does not establish the extent of this risk or demonstrate causation.
The study introduces a public benchmark for analyzing hate speech in Persian. According to the authors, LLM performance declines on more complex semantic reasoning; intermediate reasoning steps improve severity estimates compared with a single-pass assessment.
The authors compared 138 studies on sparse autoencoders in NLP. They propose a task-based taxonomy and assess the evidence on interpretability, faithfulness to the internal mechanisms of models, and practical applications. They highlight differences in evaluation methods.
The study presents new guarantees for the IR-PRM+ algorithm using previous-utility prediction. For finite two-player zero-sum matrix games, it establishes constant external regret and an average duality gap of order O(T⁻¹) under both simultaneous and alternating updates.
Selection-based Structured Reasoning replaces free-form reasoning generation with selection from predefined options. According to the authors, across seven tests with models containing 2 and 4 billion parameters, the method reduced reasoning latency by more than 90 % and overall inference latency by 28–54 % while maintaining the success rate.
A study on ToolBench with 46 984 APIs shows that retrieval before planning leads to better tool selection for tasks involving a single API, while the reverse order works better for tasks involving multiple APIs. However, according to the study, planning before retrieval roughly doubles inference costs.
The Anthropic Institute presented the questions shaping its research agenda. It plans to study the impacts of AI using information available within a lab developing advanced models and to share its findings with the public. The available excerpt does not include the specific questions.
Part of an overview of multiple AI topics; only this event has been covered.
According to Haydn Belfield, internal measurements at Google Deepmind do not show that AI is speeding up research. AI writes over 75 % of production code, but a human reviews every change before it is submitted. The success rate of the Argon system declines as tasks become more complex.
According to the study described, AI models flagged the risk of intimate partner violence an average of more than three years before female patients entered a hospital support program. The commentary highlights the study’s limitations and the risk that the assessment could interfere with patients’ decisions about sharing their experiences.
According to a study by Neil Thompson's team, the success rate of the best models tested on selected theoretical problems rose from approximately 2 % to 35 % between mid-2024 and mid-2026. The tests were conducted without internet access and used solutions published after the training data cutoff.
Researchers used AI to design enzymes, preserving the catalytic core of two types of proteases and replacing the rest of the structure. They demonstrated the enzymes' ability to cleave selected targets on proteins associated with neurological diseases.
The authors introduced HiRO-ACEv2 for global climate simulations at 3 km resolution. According to their measurements, the computational cost is approximately 300 times lower than for X-SHiELD; when processing coarsened data, the statistics for even very rare extremes differ from the reference simulation by no more than 3 %.
Researchers at Anthropic ran the Gemma 4 model with 26 billion parameters on 15 Android TV devices. In a separate simulation, the GLM-5.2 model managed to spread the Llama 3.1-8B model across the entire network. The excerpt highlights the distinction between the experiments and the predicted threat.
Researchers developed a machine learning model that corrects errors in existing forecasting models. According to the article, when applied to an established model, it doubled the accuracy of forecasts two to six weeks ahead. The study was published in the journal Nature.
According to the research described, consumers who have experienced discrimination are more likely to choose AI recommendations over human advice, even in a different area. The authors link this to feelings of shame and the perceived impartiality of AI. The available excerpt does not provide the methodology or effect size.
AI Radar monitors Czech and international sources every day, looking for changes that truly deserve attention.
MonitorsOfficial AI company blogs, specialist media, and research sources.
Selects and combinesFilters out information noise and combines articles about the same change into a single event.
Summarizes and explainsExplains significant events in English: what happened, why it matters and where the information comes from.
The result is a quick overview of what has actually changed in the AI world, rather than another stream of articles.
Use the CS/EN switch to read the same Radar in Czech or English. English content is published after its translation has been checked, so new and older items may appear later.
Everything you need to navigate the AI world
Today’s briefingAn overview of news from the past 24 hours and a main story when available. Beside them is Live · AI Flash, followed by research and links to other Radar sections. On mobile, the blocks appear one below another.
AI FlashAn ongoing feed of brief updates with an evidence status. Links lead to a Radar detail page when one is ready, otherwise to the original source. You can also find reset and outage histories here.
Practical applicationsWhat new tools and features can do, what you can try and what their actual impact could be.
Model selectionModel comparison by type of work, capabilities, price and speed.
Research and archiveA separate research overview, topic search and older events by date.
One event, everything that matters
Each row represents one event — not one article. At a glance, you can see its significance, credibility and main point.
Illustrative example, not a current news item.
Importance: ▮▮▮ majorOpenAIModels✓ 6
Agent mode is available to all paying users
Until now, the mode was available only on the highest plan; it is now available on all paid tiers without a waitlist.
▮▮▮ major · ▮▮ important · ▮ we're tracking = how significant the change is✓ 6 = six independent publishers, not the number of articles or feeds✓ official = a clear release, law or incident is substantiated by the relevant authority1 source = no independent confirmation yetbold = who is behind the changegray text = a brief summary of what happened
The detail page contains a fuller summary, its significance and original sources. Practical impact appears in the detail and the For individuals and For businesses views. An AI Flash item reaches the main selection only after it has been expanded and meets the publication rules.
The same news, two practical uses
We first summarize each event in the same way for everyone. Based on those same facts, we then explain what the change means for your own use and what it could mean for how a company operates.
For individualsWhat you can use or try, how the change can help you at work and what to watch out for.
For businessesWhat impact the change could have on processes, costs, risks and other business decisions.
Today’s briefing is the same for everyone. Pages
For individuals and For businesses
can be found in the main navigation — they select only events relevant to the given use case.