Crucible moves back into ‘closed beta’ — here’s what that means

http://feedproxy.google.com/~r/venturebeat/SZYF/~3/E3ZgYmoWRo8/

Amazon isn’t quite pulling the plug on Crucible, but it is making a drastic move on its path toward reviving the team-based shooter. Starting tomorrow, July 1, at 9 a.m. Pacific time, Crucible will move back into “closed beta.” This is coming after the game already released on Steam. Put as simply as possible, this means Amazon will no longer have the game up for new players to download (it’s a free-to-play game). But if you already added it to your library, you can continue to access it without interruption.

In a blog post on the Crucible website, Amazon says that most things won’t change if you’re an active player. But if you want to play with anyone else, they’ll need to add it to their library now.

“If you’ve got friends who want to be part of the Crucible beta, you can encourage them to get the game before tomorrow morning,” reads the blog post. “In the near future, newly interested folks will be able to sign up through playcrucible.com.”

But one of the major problems with Crucible is that it doesn’t exactly have a lot of active players.

VB Transform 2020 Online – July 15-17. Join leading AI executives:
Register for the free livestream.

Above: Crucible’s audience has vanished on Steam.

Image Credit: SteamDB

Crucible hit a peak of only 159 concurrent players over the last 24 hours. That is down from 25,145 concurrents at its peak a month ago.

As it stands today, Crucible seems destined to join Boss Key’s shooter Lawbreakers and Valve’s Artifact as one of gaming’s biggest flops in recent years.

Where Crucible went wrong

If anyone knew exactly why one game succeeds and another doesn’t, they would likely be an extremely wealthy consultant or publishing executive. But you can look to a number of factors for some insight here.

Crucible came out at a tough time. It essentially went head-to-head with Riot’s Valorant, which had a ton of built-in interest as the next big game from the League of Legends studio. But Riot also pumped up the marketing for Valorant by paying streamers to play the game for their massive audiences. Crucible did not have the same level of external support.

Amazon also doesn’t have a history in gaming. The company has spent a lot of money in the space. It has acquired talent and studios. But all it has to show for it — in terms of consumer products — is the disappointing Grand Tour racing game based on the company’s Prime Video show.

Making games is hard. It’s so hard that companies like Google and Amazon cannot just waltz in with stacks of cash and find immediate success. Ubisoft has 15,000 employees all dedicated to making about a dozen or so video games. Amazon hasn’t shown that it’s willing to dedicate those kinds of resources to its gaming products yet. And that likely means that it’s going to continue failing when it comes to this market.

Facebook’s MARGE AI summarizes and translates documents without fine-tuning

http://feedproxy.google.com/~r/venturebeat/SZYF/~3/UrXfvNnvkh4/

In a paper published on the preprint server Arxiv.org, Facebook researchers describe Multilingual Autoencoder that Retrieves and Generates (MARGE). It’s a language model that generates words, sentences, and paragraphs by retrieving related words, sentences, and paragraphs in different languages and identifying patterns within them.

The researchers claim MARGE learns to paraphrase, translate, and summarize text without any fine-tuning, a potential step toward systems that can perform any text task from pretraining alone.

In machine learning, pretraining involves training an AI model on a vast amount of data before it’s fine-tuned on a narrow data set tailored to particular tasks, like summarization. Masked models — which pretrain by removing and then reconstructing parts of an input text — are widely used in the language domain. But by design, they have to memorize a vast amount of encyclopedic knowledge to achieve strong performance.

Facebook MARGE AIFacebook MARGE AI

Above: A demonstration of MARGE’s translation skills.

MARGE, by contrast, emphasizes paraphrasing while reducing the required amount of knowledge. During pretraining, it ingests batches of “evidence” documents and target documents, and it learns to accurately summarize and translate specific snippets of text (conditioned on the evidence documents) as it susses out the relevance of evidence to each target.

VB Transform 2020 Online – July 15-17. Join leading AI executives:
Register for the free livestream.

MARGE first computes a relevance score between every pair of documents, which encourages it to attend more to relevant evidence documents. It then computes the likelihood of reconstructing each target using a modified seq2seq model, a general-purpose encoder-decoder model for language processing. Lastly, MARGE constructs batches so that evidence documents are relevant to the targets, using the relevance model for retrieval.

During experiments, the researchers created a Transformer model with 960 million parameters dubbed MARGE-NEWS, which comprised 2,048 “workers” that processed sub-batches of 4 documents (2 evidence and 2 targets) each for 550,000 steps. They further pretrained it for 100,000 steps on Wikipedia data and rebuilt the index every 10,000 steps, so that MARGE-NEWS took on average 4 monolingual and 4 cross-lingual links per target document. (The documents spanned 26 different languages in total.)

The researchers report that on the task of cross-lingual sentence retrieval, MARGE outperformed all other unsupervised models (i.e., models that look for patterns in unlabeled data sets) according to one benchmark (BUCC), and performed comparably to Facebook’s leading XLM-R model against another benchmark (Tatoeba). And on BLEU, a metric that measures language translation quality, MARGE achieved 3.58 for German to English — among the highest scores for a system without fine-tuning.

MARGE also edged out state-of-the-art models when tasked with determining whether two sentences are paraphrases and answering questions about documents in Chinese. It struggled in some cases to generate non-English languages, particularly those with non-Latin alphabets, but the researchers report that English-to-French worked well.

“MARGE exhibits strong performance on a range of discriminative and generative tasks in many languages, both with and without fine-tuning … We show that fine-tuning gives strong performance on a range of discriminative and generative tasks in many languages, making MARGE the most generally applicable pre-training method to date,” the coauthors wrote. “Future work should scale MARGE to more domains and languages, and study how to more closely align pre-training objectives with different end tasks.”

It should be noted that the researchers don’t appear to have tested MARGE on data sets designed to uncover gender, racial, ethnic, and other biases, like StereoSet. This is somewhat concerning considering Facebook’s poor ethical track record as of late. A spokesperson recently told VentureBeat the company doesn’t tally diversity statistics by teams like Facebook AI Research, the group that produced this work. And in a recent Twitter exchange, Facebook chief AI scientist Yann LeCun suggested data alone leads to prejudicial AI systems, a position with which scholars like Google ethical AI co-lead Timnit Gebru took issue.

Google’s G Suite finalizes Connected Sheets and introduces AI-driven data cleanup tools

Last April during its Cloud Next conference, Google unveiled Connected Sheets, a type of Google Sheets spreadsheet that works with the full data set from BigQuery, up to 10 billion rows. After just over a year in preview and beta, Connected Sheets is generally available as of today. And in the coming months, it’ll be joined by new capabilities — Smart Fill and Smart Cleanup — that leverage AI to learn patterns between columns to autocomplete data and surface suggestions in Sheets’ side panel.

Connected Sheets, along with Smart Fill and Smart Cleanup, are intended to make it easier for G Suite customers to take informed actions and produce better results. According to Gartner, 87% of organizations have low business intelligence and analytics maturity, meaning they’re largely relying on spreadsheet-based management systems while lacking data guidance and support.

“At Google Cloud, we believe everyone — not just those who specialize in writing complex queries — should be able to harness the power of data,” G Suite product manager Ryan Weber wrote in a blog post. “We continue to build Google AI natively into Sheets, so it’s easy for everyone — not just specialized analysts — to quickly make data-backed decisions.”

Google Sheets Smart FillGoogle Sheets Smart Fill

Above: Smart Fill in Google Sheets.

As previously detailed, Connected Sheets enables users to analyze petabytes of data in Sheets without having to use programming languages like structured query language (SQL). Analyses in Connected Sheets can be performed with tools like formulas, pivot tables, and charts, and can be visualized as dashboards and shared with anyone within an organization.

VB Transform 2020 Online – July 15-17. Join leading AI executives:
Register for the free livestream.

Smart Fill tackles a different problem. Given a column of full names, for example, it automatically detects the pattern, generates the corresponding formula, and autocompletes the rest of the column. Weber compares it to Smart Compose, a Gmail tool that taps AI to autofill emails with fewer mistakes. Smart Fill relies on common patterns of data mappings (e.g., combining one column with another to derive an output column) and analyzes data within a user’s spreadsheet to evaluate whether formulas, data from the user’s G Suite people directory, or knowledge available through Google’s Knowledge Graph will assist in data entry.

As for Smart Cleanup, it surfaces algorithmic suggestions in Sheets tailored to imported data. Specifically, it helps identify and fix duplicate rows and number-formatting issues, showing column stats that provide a snapshot of data, including the distribution of values and the most frequent value in a column. Smart Cleanup similarly evaluates whether common data cleanup actions (like removing duplicates post-import) are relevant for a given sheet, and it surfaces the most appropriate suggestions to aid a user in swiftly cleaning up data prior to analysis.

A Google spokesperson told VentureBeat that AI models developed in Google’s TensorFlow framework are deployed when appropriate to make Smart Fill and Smart Cleanup suggestions more relevant and helpful. “As these features become available, users can expect Smart Fill to continue becoming more intelligent in learning patterns to autocomplete data,” the spokesperson said, “and Smart Cleanup to make data cleanup faster and more accurate for a broader and more diverse set of data cleanup operations and data sources in Sheets.”

Google Sheets Smart Cleanup

Google Sheets Smart Cleanup

“Before making critical decisions, it’s important to ensure your data is consistent and error-free,” Weber wrote. “[These features] make data entry quicker and less error-prone.”

Connected Sheets is available starting today for G Suite Enterprise, G Suite Enterprise for Education, and G Suite Enterprise Essentials customers. Smart Fill and Smart Cleanup will arrive on G Suite later this year.

The new Sheets capabilities come as Google looks to inject G Suite with more AI-powered functionality. Recently, the company added a feature that lets users ask natural language questions about data in spreadsheets, like “Which person has the top score?” and “What’s the sum of price by salesperson?” Google Meet earlier this year gained adaptive noise cancellation.

Two years ago, Google rolled out Quick Access, a machine learning-powered tool that suggests files relevant to documents users are editing, to Sheets, Docs, and Slides. And more recently, Google brought assistive features like grammar suggestions and spelling autocorrect to Google Docs in Spanish (previously, they were only available in English).

Waymo to expand autonomous truck testing in the American Southwest

http://feedproxy.google.com/~r/venturebeat/SZYF/~3/p_Ih2SUrMc0/

Today during a briefing with members of the media, Waymo head of commercialization for trucking Charlie Jatt outlined the company’s go-to-market plans for Waymo Via, its self-driving delivery division. In the future, Waymo will partner with OEMs and tier 1 suppliers to equip cloud-based trucks manufactured and sold to the market with its autonomous systems. In addition, Waymo will work with fleets to provide its software services and offer support for things like mapping and remote fleet assistance.

As Waymo transitions to this model, Jatt said that Waymo intends to own and offer its own fleet of trucks — at least in the short term. One of the delivery solutions it’s exploring is a transfer-hub model where rather than an automated truck covering an entire journey, it’ll be a mix of an automated portion and a portion involving manually-driven, human-manned trucks. Automated vehicle transfer hubs close to highways would handle the switch-off and minimize surface street driving.

In a first step toward this vision, Waymo says it’ll soon expand testing on roads in New Mexico, Arizona, and Texas along the I-10 corridor between Phoenix and Tuscon, as previously announced. It this year mapped routes between Pheonix, Al Paso, Dallas, and Houston and ramped up testing in California on freeways in Mountain View, but the focus in 2020 will be on the American Southwest.

Waymo Via

Waymo Via

VB Transform 2020 Online – July 15-17. Join leading AI executives:
Register for the free livestream.

Tests will be primarily along Interstates 10, 20, and 45 and through metropolitan areas like El Paso, Dallas, and Houston. Chrysler Pacifica vans retrofitted with Waymo’s technology stack will map roads ahead of driverless Peterbilt trucks as part of a project known as Husky.

Waymo is also engaged with local delivery under the Waymo Via umbrella, the company reiterated. It currently has two partnerships in the Phoenix area — one with AutoNation and one with UPS. On the AutoNation side, Waymo is performing “hot shot” deliveries where Waymo vehicles travel to certain AutoNation locations and delivers car parts. And on the UPS side, the company is ferring packages from stores to UPS sorting centers.

Waymo began piloting dedicated goods delivery with class A trucks — 18-wheelers — in 2017. After completing tests in 2018 with real loads from Google datacenters in Atlanta, Waymo began limited testing on roads in the San Francisco Bay Area, Michigan, Arizona, Georgia, and on Metro Phoenix freeways.

Waymo’s autonomous trucks employ a combination of lidars, radars, and cameras to understand the world around them. They have roughly two times the sensors compared with Waymo’s cars to handle the trucks’ unique shape and the occlusions they cause, and they place a greater emphasis on long-length perception (the perception range is somewhere beyond 300 meters). But they use the same compute platform found in the fifth-generation Waymo Driver.

As the pandemic drives unprecedented growth in the logistics and ground transportation market, Aurora, TuSimple, and other rivals are investing increased resources in fully autonomous solutions. They stand to save the logistics and shipping industry $70 billion annually while boosting productivity by 30%; according to a recent study from the Consumer Technology Association, a quarter (26%) of consumers now view autonomous delivery technologies more favorably than before the health crisis.

Besides cost savings, the growth in autonomous trucking has been driven in part by a shortage of human drivers. In 2018, the American Trucking Associates estimated that 50,000 more truckers were needed to close the gap in the U.S., even despite the sidelining of proposed U.S. Transportation Department screenings for sleep apnea.