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.
825
published research events
Latest work
A significant claim from a single source is published only after further confirmation.
Researchers at TU Delft developed a system that uses an LLM to convert natural language commands (e.g. “I'm in a hurry, drive faster”) into adjustments to autonomous vehicle parameters. The system adjusts the motion planning algorithm with safety in mind, keeps the user involved in the decision-making process, and in simulations…
An investigation by AlgorithmWatch found that ChatGPT, Gemini, Grok and Claude refer users to anti-abortion organizations in at least 25 % of queries without warning about their ideological position. Profemina appeared in 17 % of responses. In German conversations, chatbots recommended Caritas, which cannot issue…
A research paper on arXiv introduces Weighted Memory Tree, a hierarchical memory technique with dynamic retention scores for LLM agents. On the GAIA-Text benchmark with Qwen3-8B, Gemma 4 E4B and Llama-3.1-8B, it achieved an accuracy improvement of 9.97 percentage points and a reduction in token usage of 32.8 % compared with…
The study presents the KREL framework, which combines LLM with external guidelines for ICD coding. The system automatically assigns standard diagnosis and procedure codes to medical reports, reduces LLM hallucinations, and outperforms existing methods on benchmarks.
The research framework AgentMercury synthesizes 4 783 executable environments from business scenarios across 14 industries and 50 countries. Qwen3.5-4B improved performance from 12.3 to 15.7 on EnterpriseOps-GYM and from 45.9 to 56.0 on AIME26. Training the model to construct environments increases the success rate from 3.3% to 83.3%.
The research team introduced TriPLU, a technique using a direct trilinear product of three streams instead of a gated FFN layer. On a 1M-byte prefix of TinyStories, it achieved a validation loss of 1.0637 compared with 1.1017 for SwiGLU. The paper acknowledges that the results are specific to a low learning rate and do not demonstrate general…
The research defines canonical predictive models for three channels: modeling the environment, the agent, and their interactions. It uses computational mechanics and epsilon-machines. It demonstrates how coupling the agent and the environment reduces model complexity. Using a POMDP example, it shows that a constrained model can be finite rather than…
A team at Apple Machine Learning Research proposes the IVT framework for video analysis, which enables models to think visually during training without generating intermediate images. It achieves comparable or better performance than Visual CoT with 5× lower end-to-end latency.
The research framework WIEN-INR for implicit neural representations distributes modeling across resolution levels and improves the capacity to reconstruct details with a new Enhancement Network. Smaller networks retain the full spatial frequency content with lower requirements for training and scientific data storage.
Researchers have developed a method called cross-model confusion mapping that combines microphones and an AI model to detect the last remaining bruks in New Zealand's forests. Instead of training the model directly, they first use BirdNET (a model for identifying 6000+ bird species) to find sounds that AI confuses with bruks, then these…
Research: the new Mental World Modeling (MWM) framework extends world models (Sora, Genie 3, JEPA, Marble) to include modeling human beliefs, emotions and intentions. The authors published the MENTIS implementation, which models both physical and mental aspects of human behavior without additional training in a social…
The new RecPFN model brings in-context learning to recommendation. It is trained on synthetic data with a causal prior and predicts the next items from a few examples without updating its weights. It achieves the best zero-shot performance on eight benchmarks and is efficient in deployment.
A scientific paper on arXiv analyzes three transformer models (BERT, RoBERTa, BART) in the context of automatic text summarization, including their architectures and pretraining.
The research paper presents SNAIL, a hybrid system for automatically recognizing the names of bioinformatics software and databases in scientific texts. It combines lexical modeling with transformers (SciBERT) and achieves better results than both bioNerDS2 and general-purpose LLMs (ChatGPT, Gemini, Grok, Claude).
The research team proposes the TUP method, which distills Best-of-N selection (one model instead of selecting from N samples) by removing low-rated generations and increasing the weights of better ones. The method enables offline training with binary cross-entropy without dependence on a specific prompt. The paper includes theoretical justification and…
OpenAI published solutions to long-standing mathematical problems, dramatically improving AI capabilities in abstract mathematics. This sparked a discussion in the mathematical community about the future of mathematics as an academic discipline and the role of mathematicians.
A research team proposed MAVEN, a hierarchical framework for evaluating multimodal content against macro-societal values. The framework organizes values into 6 primary dimensions and 72 indicators and is grounded in international human rights instruments. The authors created the benchmark MacroValue-Bench and a compact…
Researchers introduced a method for adapting language models to the audio modality by training only a lightweight projector, while the encoder and LLM remain frozen. It achieves competitive performance on the MMAU, MMAR, MMSU and MMAU-Pro benchmarks with substantially less training data.
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.
