It was a Monday afternoon in June 2023 when Rita Meier, 45, joined us for a video call. Meier told us about the last time she said goodbye to her husband, Stefan, five years earlier. He had been leaving their home near Lake Constance, Germany, heading for a trade fair in Milan. Meier recalled how he hesitated between taking his Tesla Model S or her BMW. He had never driven the Tesla that far before. He checked the route for charging stations along the way and ultimately decided to try it. Rita had a bad feeling. She stayed home with their three children, the youngest less than a year old. At 3.18pm on 10 May 2018, Stefan Meier lost control of his Model S on the A2 highway near the Monte Ceneri tunnel. "The collision with the guardrail launches the vehicle into the air, where it flips several times before landing," investigators would write later. The car came to rest more than 70 metres away, on the opposite side of the road, leaving a trail of wreckage. Several passersby tried to open the doors and rescue the driver, but they couldn't unlock the car. When they heard explosions and saw flames through the windows, they retreated. Even the firefighters, who arrived 20 minutes later, could do nothing but watch the Tesla burn.

In this paper, we propose a novel approach to conformal prediction for generative language models (LMs). Standard conformal prediction produces prediction sets--in place of single predictions--that have rigorous, statistical performance guarantees. LM responses are typically sampled from the model's predicted distribution over the large, combinatorial output space of natural language. Translating this process to conformal prediction, we calibrate a stopping rule for sampling different outputs from the LM that get added to a growing set of candidates until we are confident that the output set is sufficient. Since some samples may be lowquality, we also simultaneously calibrate and apply a rejection rule for removing candidates from the output set to reduce noise. Similar to conformal prediction, we prove that the sampled set returned by our procedure contains at least one acceptable answer with high probability, while still being empirically precise (i.e., small) on average. Furthermore, within this set of candidate responses, we show that we can also accurately identify subsets of individual components--such as phrases or sentences--that are each independently correct (e.g., that are not "hallucinations"), again with statistical guarantees. We demonstrate the promise of our approach on multiple tasks in open-domain question answering, text summarization, and radiology report generation using different LM variants.

This paper is concerned with the automation and simulation of pick and place processes in the domain of CFK aircraft production. We introduce a workflow which starts from a CAD construction, extracts relevant data out of it, assigns grippers to the CFK pieces and schedules the single steps using a PDDL solver. Finally, the result is visualized in Blender where also prior mistakes can be identified.

This repo contains a dataset of Python programming puzzles which can be used to teach and evaluate an AI's programming proficiency. We present code generated by OpenAI's recently released codex 12-billion parameter neural network solving many of these puzzles. We hope this dataset will grow rapidly, and it is already diverse in terms of problem difficulty, domain, and algorithmic tools needed to solve the problems. Please propose a new puzzle or browse newly proposed puzzles or contribute through pull requests. To reproduce the results in the paper, see the solvers folder.

The Technion – Israel Institute of Technology, Israel's top university for science and technology, has been ranked number one in Europe, and number 15 worldwide, in the field of artificial intelligence by CSRankings. The rankings considered data from 2016 to 2021, including metrics like computer vision and natural web processing. As opposed to many university rankings that are survey-based, CSRankings measures each department based on how many publications by faculty appear at the most prestigious computer science conferences. This approach incentivizes faculty to publish at top venues and requires a built-in judgment of the research. The Technion has 46 researchers engaged in core AI fields and more than 100 researchers in related fields, such as health and medicine, autonomous vehicle, cybersecurity, and fintech.

CTech -- The Technion's efforts to advance the field of artificial intelligence have positioned it among the world's leaders in AI research and development. CSRankings, the leading metrics-based ranking of top computer science institutions around the world, has ranked the Technion in first place in the field of artificial intelligence in Europe, and 15th worldwide. In the subfield of machine learning, the Technion is ranked 11th worldwide. The data used to compile the rankings is from 2016 to 2021. One of the innovations that is part of the framework of the Technion's AI prowess is the Machine Learning and Intelligent Systems (MLIS) research center, which aggregates all AI-related activities.

A system created by MIT researchers could be used to automatically update factual inconsistencies in Wikipedia articles, reducing time and effort spent by human editors who now do the task manually. Wikipedia comprises millions of articles that are in constant need of edits to reflect new information. That can involve article expansions, major rewrites, or more routine modifications such as updating numbers, dates, names, and locations. Currently, humans across the globe volunteer their time to make these edits. In a paper being presented at the AAAI Conference on Artificial Intelligence, the researchers describe a text-generating system that pinpoints and replaces specific information in relevant Wikipedia sentences, while keeping the language similar to how humans write and edit.

Conditional mean embeddings (CME) have proven themselves to be a powerful tool in many machine learning applications. They allow the efficient conditioning of probability distributions within the corresponding reproducing kernel Hilbert spaces (RKHSs) by providing a linear-algebraic relation for the kernel mean embeddings of the respective probability distributions. Both centered and uncentered covariance operators have been used to define CMEs in the existing literature. In this paper, we develop a mathematically rigorous theory for both variants, discuss the merits and problems of either, and significantly weaken the conditions for applicability of CMEs. In the course of this, we demonstrate a beautiful connection to Gaussian conditioning in Hilbert spaces.

Detecting fake news is getting more difficult as more false information pours onto the internet ... [ ] every day, and from very influential sources. New papers from MIT explore how current methods are failing, and bring new weapons to the fight against fake news. Now is the time to make facts great again. "Fake news," the 2017 Collins word of the year, poses a serious threat to the values of honesty, truth, and accountability--values that purveyors of falsified information don't seem to hold too closely. Apart from the most obvious dangers of spreading false information (erosion of trust, political or national hostility, widespread uncertainty) the prevalence of AI systems on social media mean that unverified claims and slanderous falsehoods are picked up and distributed at eye-watering speeds.

The 21st century has opened up a boundless mass of headlines, articles, and stories. This information influx, however, is partially contaminated: Alongside factual, truthful content is fallacious, deliberately manipulated material from dubious sources. According to research by the European Research Council, one in four Americans visited at least one fake news article during the 2016 presidential campaign. This problem has recently been exacerbated by something called "automatic text generators." Advanced artificial intelligence software, like OpenAI's GPT-2 language model, is now being used for things like auto-completion, writing assistance, summarization, and more, and it can also be used to produce large amounts of false information -- fast.