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.
224
published research events4
new research papers today
Latest work
A significant claim from a single source is published only after further confirmation.
Researchers have developed an approach for predicting the topology of transmembrane proteins using the SchNet graph neural network. The model was trained on DeepTMHMM data with 5-fold cross-validation, uses atom-level embeddings, and demonstrated potential for topology predictions without pretrained weights.
Researchers have developed a framework for training navigation models for microbots in under 10 minutes instead of hours. A vectorized simulator with 10,000+ environments achieves 190,000 transitions per second. A 33.7% reduction in action variance and a +2.1% increase in obstacle clearance were achieved. Code and data are public.
Research published in Nature Machine Intelligence shows that multitask learning leads RNNs to a modular organization similar to brain networks. Functional demands, not just spatial constraints, shape network topology. Code and data are publicly available.
A research team writing in Nature Machine Intelligence presents GenFocal, an AI framework for generating detailed regional weather forecasts from coarse climate models. The tool does not require paired training data and increases the accuracy of forecasts for complex phenomena such as heat waves and tropical cyclones.
The paper improves theoretical guarantees for federated optimization of stochastic variational inequalities. The authors analyze the Local Extra SGD algorithm and propose a new algorithm, LIPPAX, with better convergence in regimes with a bounded Hessian and operator. No practical released product.
A meta-analysis of 19 studies and a proprietary experiment with 411 Australian participants show: 75% of people correctly identify deepfakes, but 47% fail on authenticity — 29% label a real video as fake. Deepfakes reduce trust in newspapers and democratic institutions.
Fudan University studied the development of the Atria Dawn Preview model (744B parameters, MoE). AI agents were involved in 96.5% of tasks, but humans made the decisions in 85.5% of cases regarding method and 93.4% regarding goals. A third of AI-assisted tasks would not have been attempted at all without agents.
Artificial intelligence analyzed thousands of Reddit posts and revealed previously overlooked side effects of weight-loss drugs: fatigue, body temperature fluctuations, and menstrual problems. The findings require verification on a representative sample of patients.
A study with 3 132 participants shows that access to an AI model reduces willingness to say "I don't know" from 36–44% to 6–3%. Confidence increases 2.5 times, but the correctness of answers drops from 27.6% to 10.0%; financial incentives partially mitigate the problem.
Nvidia SoL-Pi automates optimization of the harness (control layer) of coding agents and reduces token usage by 50 percent versus Codex and 54 percent versus Claude Code, without loss of performance. A search over 152 directions across 535 environments generated 3000 runs; the method separates search from evaluation in order to prevent…
Epoch AI published the results of the FAB (Furniture Assembly Benchmark) benchmark testing the ability of AI models to identify errors in IKEA furniture assembly. GPT-6 Astra achieved 80% accuracy, Claude Fable 5.1 70%, Claude Opus 5 61%. The benchmark from November 2025 shows dramatic progress…
CESifo research (Fairlie, Wu) finds no evidence of a decline in employment of new graduates due to AI, in contrast to the Stanford study. The authors focused on recent graduates because the AI impact may first show up as a reduction in hiring. They mention an increase in AI spending and the use of the ChatGPT Enterprise product for…
The authors propose a framework for learning cross-task relationships in multi-task models. Application to YouTube recommendation systems (Notifications, Homepage, Watch Next) shows improvements in accuracy and user satisfaction metrics.
A research paper on arXiv presents the CataOPD method, combining reinforcement learning and real-time distillation, which improves LLMs' ability to solve complex tasks. The method achieves better results on out-of-distribution tests, and the code is publicly available.
A research preprint describes a pipeline for automated response to compliance questions using smaller language models. A LegalBERT-based retriever increased Recall@10 on the ObliQA benchmark from 0.256 to 0.774. RAFT-LoRA fine-tuning improved answer quality, but the models transfer poorly to other legal…
The technical report introduces the Pistis family of models (27B and 9B parameters) built on Qwen3.6/3.5. The new IDRL (Interleaved Distillation and Reinforcement Learning) approach combines distillation and reinforcement learning in a single training loop. Two special variants: Pistis-Thinking for multimodal reasoning…
A study comparing Knowledge Tracing models XGBoost+TreeSHAP (AUC 0.777–0.786) with deep models DKT, SAKT, AKT, SimpleKT. On identical data, they achieved the same accuracy (0.697–0.720); the difference in accuracy stemmed from the available information, not from the model. TreeSHAP explanations were stable (Spearman…
The study benchmarks the ability of LLMs to respond to ad hominem attacks in dialogues. An analysis of a corpus of presidential debates shows that LLMs focus on logical defenses and cannot strategically employ ethical counterattacks. Safety fine-tuning limits their argumentative flexibility.
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 briefingThe “What is worth attention” selection sits beside Live · AI Flash, followed by research and links to other Radar sections. On mobile, these 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.
