Tesla is being investigated by the US government after reports the firm's self-driving cars had broken traffic laws, including driving on the wrong side of the road and not stopping for red lights. It said it was aware of 58 reports where the electric cars had committed such violations, according to a filing from the National Highway Traffic Safety Administration (NHTSA). An estimated 2.9 million cars equipped with full self-driving tech will fall under the investigation. Tesla, whose boss Elon Musk recently became the world's first half-trillionaire, has been approached for comment. The NHTSA's preliminary evaluation will assess the scope, frequency, and potential safety consequences of the Full Self-Driving (Supervised) mode.

We introduce EarthPT -- an Earth Observation (EO) pretrained transformer. EarthPT is a 700 million parameter decoding transformer foundation model trained in an autoregressive self-supervised manner and developed specifically with EO use-cases in mind. We demonstrate that EarthPT is an effective forecaster that can accurately predict future pixel-level surface reflectances across the 400-2300 nm range well into the future. For example, forecasts of the evolution of the Normalised Difference Vegetation Index (NDVI) have a typical error of approximately 0.05 (over a natural range of -1 -> 1) at the pixel level over a five month test set horizon, out-performing simple phase-folded models based on historical averaging. We also demonstrate that embeddings learnt by EarthPT hold semantically meaningful information and could be exploited for downstream tasks such as highly granular, dynamic land use classification. Excitingly, we note that the abundance of EO data provides us with -- in theory -- quadrillions of training tokens. Therefore, if we assume that EarthPT follows neural scaling laws akin to those derived for Large Language Models (LLMs), there is currently no data-imposed limit to scaling EarthPT and other similar `Large Observation Models.'

A key goal of current mechanistic interpretability research in NLP is to find linear features (also called "feature vectors") for transformers: directions in activation space corresponding to concepts that are used by a given model in its computation. Present state-of-the-art methods for finding linear features require large amounts of labelled data -- both laborious to acquire and computationally expensive to utilize. In this work, we introduce a novel method, called "observable propagation" (in short: ObsProp), for finding linear features used by transformer language models in computing a given task -- using almost no data. Our paradigm centers on the concept of observables, linear functionals corresponding to given tasks. We then introduce a mathematical theory for the analysis of feature vectors: we provide theoretical motivation for why LayerNorm nonlinearities do not affect the direction of feature vectors; we also introduce a similarity metric between feature vectors called the coupling coefficient which estimates the degree to which one feature's output correlates with another's. We use ObsProp to perform extensive qualitative investigations into several tasks, including gendered occupational bias, political party prediction, and programming language detection. Our results suggest that ObsProp surpasses traditional approaches for finding feature vectors in the low-data regime, and that ObsProp can be used to better understand the mechanisms responsible for bias in large language models. Code for experiments can be found at github.com/jacobdunefsky/ObservablePropagation.

For many islanders, delays to the postal service are an inescapable part of life. But that should no longer be the case for those living in Orkney, after it became the first place in Britain to have mail delivered by a drone. The new Royal Mail service will see post transported from the Kirkwall delivery office to the village of Stromness, where drones will then transfer items to posties on the islands of Hoy and Graemsay for their regular routes. Currently, mail arrives at Kirkwall Airport before being sent by plane or ferry to Orkney's 19 inhabited islands. But the challenging geography and weather conditions often result in delivery disruptions.

Royal Mail has begun using drones to deliver post in the Orkney islands, helping pave the way for drone deliveries to islands around the UK and on the mainland during emergencies. The service between the village of Stromness on Orkney's main island to the nearby islands of Hoy and Graemsay, using aircraft able to carry up to 6kg, is Royal Mail's first permanent drone delivery service. Using drones allows Royal Mail to provide a faster and more secure delivery service to islands such as the Orkneys, avoiding ferries or scheduled air services subject to weather cancellations, tides and timetables that do not suit the postal service. Royal Mail has been testing and evaluating drone services on Scottish islands for some time, as has the NHS, which has trialled their use for flying urgent medical samples from the small Hebridean islands of Coll and Tiree. Chris Paxton, the head of drone trials at Royal Mail, said these flights were far faster and more efficient, and helped cut carbon emissions.

Text generation is of great importance to many natural language processing applications. However, maximization-based decoding methods (e.g., beam search) of neural language models often lead to degenerate solutions--the generated text is unnatural and contains undesirable repetitions. Existing approaches introduce stochasticity via sampling or modify training objectives to decrease the probabilities of certain tokens (e.g., unlikelihood training). However, they often lead to solutions that lack coherence. In this work, we show that an underlying reason for model degeneration is the anisotropic distribution of token representations. We present a contrastive solution: (i) SimCTG, a contrastive training objective to calibrate the model's representation space, and (ii) a decoding method--contrastive search--to encourage diversity while maintaining coherence in the generated text. Extensive experiments and analyses on three benchmarks from two languages demonstrate that our proposed approach significantly outperforms current state-of-the-art text generation methods as evaluated by both human and automatic metrics.

The UK's Royal Mail wants to set up as many as 50 drone routes over the next three years to make deliveries to remote communities. The plan, which requires approval from the Civil Aviation Authority, would see the service secure up to 200 of the autonomous devices from logistics drone company Windracers. The Royal Mail said the first communities to benefit would be the Isles of Scilly (off the coast of Cornwall in south-west England) and the Scottish islands of Shetland, Orkney and the Hebrides. Test flights started last year. In the most recent one, held in April, the service was able to use a UAV to deliver mail to Unst, Britain's most northerly inhabited island, from Tingwall Airport on Shetland's largest island.

Royal Mail is building a fleet of 500 drones to carry mail to remote communities all over the UK, including the Isles of Scilly and the Hebrides. The postal service, which has already conducted successful trials over Scotland and Cornwall, will create more than 50 new postal drone routes over the next three years as part of a new partnership with London company Windracers. Drones, or UAVs (uncrewed aerial vehicles), can help reduce carbon emissions and improve the reliability of island mail services, Royal Mail claims. They offer an alternative to currently-used delivery methods that can be affected by bad weather – ferries, conventional aircraft and land-based deliveries. They can also take off from any flat surface (sand, grass or tarmac) providing it is long enough.

Earlier this year, a 71-year-old man in the Swedish city of Trollhattan was shovelling snow outside his house when he suffered a cardiac arrest -- his heart suddenly stopped beating. A passing doctor rushed to help and began performing cardiopulmonary resuscitation (CPR) to keep blood and oxygen flowing to his brain and other vital organs, while a bystander called for an ambulance. It was a race against time. With no pulse, the man's heart needed to be shocked back into life using a defibrillator. The drone took just three minutes to arrive, reaching the patient well ahead of paramedics.

As readers recall, I'd previously reviewed Richard Abbott's debut sci-fi novel, Far from the Spaceports, later returning for more Mitnash and Slate in its sequel, Timing. It was rather exciting listening to it, and I am so pleased to have the opportunity to share it here, along with some author comments as to the linguistics involved in setting up the pieces. First, for those unfamiliar with the novels and their plots, I've linked the book covers to their respective Amazon blurbs. Abbott's world-building opens a new type of sci-fi, one accessible even to those not typically enamored of the genre (such as myself), and the above-mentioned duo will capture your imagination as they seek to solve the mysteries of high-tech crime in space. Today you'll hear--and can read along--a bit of discussion between Mitnash and Slate, along with another pair, Rydal and Capstone, as the group talks about oddities in the data they are studying.