Google’s Gemini 1.5 Pro is a new, more efficient AI model

Posted on 15/02/2024 by Will Shanklin

On Thursday, Google unveiled Gemini 1.5 Pro, which the company describes as delivering “dramatically enhanced performance” over the previous model. The company’s AI trajectory — viewed internally as increasingly critical for its future — follows the unveiling of Gemini 1.0 Ultra last week, alongside the rebranding of the Bard chatbot (to Gemini) to align with the new model’s more powerful and versatile capabilities.

In an announcement blog post, Google CEO Sundar Pichai and Google DeepMind CEO Demis Hassabis try to balance assuring their audience about ethical AI safety while touting their models’ rapidly advancing capabilities. “Our teams continue pushing the frontiers of our latest models with safety at the core,” Pichai summarized.

The company needs to emphasize safety for AI skeptics (including one former Google CEO) and government regulators. But it also needs to stress its models’ accelerating performance for AI developers, potential customers and investors concerned the company was too slow to react to OpenAI’s breakout success with ChatGPT.

Pichai and Hassabis say Gemini 1.5 Pro delivers comparable results to Gemini 1.0 Ultra. However, Gemini 1.5 performs at that level more efficiently, with reduced computational requirements. The multimodal capabilities include processing text, images, videos, audio or code. As AI models advance, they’ll continue to offer a more versatile array of capabilities in one prompt box (another recent example was OpenAI integrating DALL-E 3 image generation into ChatGPT).

Alphabet Inc. and Google CEO Sundar Pichai attends the inauguration of a Google Artificial Intelligence (AI) hub in Paris on February 15, 2024. (Photo by ALAIN JOCARD / AFP) (Photo by ALAIN JOCARD/AFP via Getty Images) — *Google CEO Sundar Pichai*

Gemini 1.5 Pro can also handle up to one million tokens, or the units of data AI models can process in a single request. Google says Gemini 1.5 Pro can process over 700,000 words, an hour of video, 11 hours of audio and codebases with over 30,000 lines of code. The company says it’s even “successfully tested” a version that supports up to 10 million tokens.

The company says Gemini 1.5 Pro maintains high accuracy in queries with larger token counts when it has more new data to learn. It says the model impressed in the Needle In a Haystack evaluation. In this test, developers insert a small piece of information inside a long text block to see if the AI model can pick it out. Google said Gemini 1.5 Pro could find the embedded text 99 percent of the time in data blocks as long as one million tokens.

Google says Gemini 1.5 Pro can reason about various details from the 402-page Apollo 11 moon mission transcripts. In addition, it can analyze plot points and events from an uploaded 44-minute silent film starring Buster Keaton. “As 1.5 Pro’s long context window is the first of its kind among large-scale models, we’re continuously developing new evaluations and benchmarks for testing its novel capabilities,” Hassabis wrote.

Google is launching Gemini 1.5 Pro with 128,000-token capabilities, the same number at which OpenAI’s (publicly announced) GPT-4 models max out. Hassabis says Google will eventually introduce new pricing tiers that support up to one million-token queries.

NEW YORK, NEW YORK - MAY 02: Demis Hassabis attends 2023 WSJ's Future Of Everything Festival at Spring Studios on May 02, 2023 in New York City. (Photo by Joy Malone/Getty Images) — *Google DeepMind CEO Demis Hassabis*

Gemini 1.5 Pro is also adept at learning new skills from information in long prompts — without additional fine-tuning (“in-context learning”). In a benchmark called Machine Translation from One Book, the model learned a grammar manual for Kalamang, a language with fewer than 200 speakers globally that it hadn’t previously been trained on. The company says Gemini 1.5 Pro learned to perform at a similar level as a human learning the same content when translating English to Kalamang.

In a piece of the announcement that will catch developers’ attention, Google says Gemini 1.5 Pro can perform problem-solving tasks across longer code blocks. “When given a prompt with more than 100,000 lines of code, it can better reason across examples, suggest helpful modifications and give explanations about how different parts of the code works,” Hassabis wrote.

On the ethics and safety front, Google says it’s taking “the same approach to responsible deployment” it took with Gemini 1.0 models. That includes developing and applying red-teaming techniques, where a group of ethical developers essentially serve as devil’s advocate, testing for “a range of potential harms.” In addition, the company says it heavily scrutinizes areas like content safety and representational harms. The company says it continues to develop new ethical and safety tests for its AI tools.

Google is launching Gemini 1.5 in early access for developers and enterprise customers. The company plans to make it more widely available eventually. Gemini 1.0 is currently available for consumers, alongside a Pro variant that costs $20 monthly.

This article originally appeared on Engadget at https://www.engadget.com/googles-gemini-15-pro-is-a-new-more-efficient-ai-model-181909354.html?src=rss

Google’s Gemini 1.5 Pro is a new, more efficient AI model

Posted on 15/02/2024 by Will Shanklin

This article originally appeared on Engadget at https://www.engadget.com/googles-gemini-15-pro-is-a-new-more-efficient-ai-model-181909354.html?src=rss

Russian and North Korean hackers used OpenAI tools to hone cyberattacks

Posted on 14/02/2024 by Kris Holt

Microsoft and OpenAI say that several state-backed hacking groups are using the latter’s generative AI (GAI) tools to bolster cyberattacks. The pair suggests that new research details for the first time how hackers linked to foreign governments are making use of GAI. The groups in question have ties to China, Russia, North Korea and Iran.

According to the companies, the state actors are using GAI for code debugging, looking up open-source information to research targets, developing social engineering techniques, drafting phishing emails and translating text. OpenAI (which powers Microsoft GAI products such as Copilot) says it shut down the groups’ access to its GAI systems after finding out they were using its tools.

Notorious Russian group Forest Blizzard (better known as Fancy Bear or APT 12) was one of the state actors said to have used OpenAI's platform. The hackers used OpenAI tools "primarily for open-source research into satellite communication protocols and radar imaging technology, as well as for support with scripting tasks," the company said.

As part of its cybersecurity efforts, Microsoft says it tracks north of 300 hacking groups, including 160 nation-state actors. It shared its knowledge of them with OpenAI to help detect the hackers and shut down their accounts.

OpenAI says it invests in resources to pinpoint and disrupt threat actors' activities on its platforms. Its staff uses a number of methods to look into hackers' use of its systems, such as employing its own models to follow leads, analyzing how they interact with OpenAI tools and determining their broader objectives. Once it detects such illicit users, OpenAI says it disrupts their use of the platform through the likes of shutting down their accounts, terminating services or minimizing their access to resources.

This article originally appeared on Engadget at https://www.engadget.com/russian-and-north-korean-hackers-used-openai-tools-to-hone-cyberattacks-152424393.html?src=rss

Uber, Lyft and DoorDash drivers are striking on February 14

Posted on 13/02/2024 by Mariella Moon

It could be a challenge hailing a ride from certain airports on Valentine's Day this year. Thousands of rideshare and delivery drivers for Uber, Lyft and DoorDash are planning to hold a demonstration on February 14 to demand fair pay and better security measures, according to Reuters. The strike was announced last week by Justice for App Workers, a coalition representing more than 100,000 rideshare and delivery drivers across the US.

Based on the group's page for the rally, workers participating in the demonstration won't be taking rides to and from any airport in Austin, Chicago, Hartford, Miami, Newark, Orlando, Philadelphia, Pittsburgh, Rhode Island and Tampa. The coalition is asking drivers to join the event and "demand changes from Uber, Lyft, DoorDash, and all the app companies profiting off of [their] hard work." Meanwhile, Rideshare Drivers United, an independent union for Uber and Lyft drivers in Los Angeles, also revealed that its members are turning off their apps on February 14 to protest "the significant decrease in pay [they've] all felt this winter."

While the strikes could see the participation of tens of thousands of workers, Uber believes it won't have an impact on its business since only a small portion of its drivers typically take part in demonstrations. The company told The Hill and CBS News that a similar protest last year didn't affect its operations and that its driver earnings remain "strong." In the fourth quarter of 2023, "drivers in the US were making about $33 per utilized hour," the spokesperson said.

The groups announced the strikes just a few days after Lyft promised guaranteed weekly earnings for its drivers in the country, ensuring that they'll make at least 70 percent of what their riders had paid. DoorDash didn't respond to the publications' requests for comment, but it currently pays its drivers $29.93 for every active hour in states with minimum wage requirements for app-based delivery workers. It recently introduced new fees for customers in New York City and Seattle as a response to their new minimum wage regulations.

This article originally appeared on Engadget at https://www.engadget.com/uber-lyft-and-doordash-drivers-are-striking-on-february-14-055949899.html?src=rss

Google rebrands its Bard AI chatbot as Gemini, which now has its own Android app

Posted on 08/02/2024 by Kris Holt

Just as Microsoft renamed Bing Chat to Copilot to unify its generative AI branding, Google is doing the same thing with Bard and Duet AI. The services now bear the name Gemini, after Google's multimodal AI model. The name change leaked earlier this month. Google has also debuted a dedicated Gemini Android app alongside a paid version of the chatbot that has more enhanced capabilities.

"Bard has been the best way for people to directly experience our most capable models," Google CEO Sundar Pichai wrote in a blog post. "To reflect the advanced tech at its core, Bard will now simply be called Gemini. It’s available in 40 languages on the web and is coming to a new Gemini app on Android and on the Google app on iOS."

Those who download the Gemini Android app can actually replace Google Assistant as the default assistant on their device. So, when you long press the home button or utter "Hey Google," your phone or tablet can fire up Gemini instead of Assistant. You can also make this switch by opting in through Assistant.

Doing so will enable a new conversational overlay on your display. Along with swift access to Gemini, the overlay will offer contextual suggestions, such as the ability to generate a description for a photo you just took or ask for more information about an article that's on your screen.

You'll also be able to access commonly used Assistant features through the Gemini app, from making calls and setting timers to controlling smart home devices. Google said it will bring more Assistant functions to Gemini in the future. That certainly makes it sound as though Google is phasing out Assistant in favor of Gemini. The app also includes access to Gemini Advanced (more on that in a moment).

As for iOS, there won't be a separate Gemini app for now. Instead, you can access it through the Google app by tapping the Gemini toggle.

Gemini is available on Android and iOS in English in the US starting today. Next week, Google will start offering access to the chatbot in more locales in English, as well as in Japanese and Korean. As you might expect, Gemini is coming to more countries and languages down the line.

In addition, Google is opening up access to what it says is its largest and most capable AI model, Ultra 1.0, through Gemini Advanced. The company claims this is able to have longer and more in-depth conversations with the ability to recall context from previous chats. It says Gemini Advanced "is far more capable at highly complex tasks like coding, logical reasoning, following nuanced instructions and collaborating on creative projects."

Gemini Advanced is available now in English in 150 countries and territories. To access it, you'll need to sign up for the new Google One AI Premium Plan. This costs $20 per month — the same price as Copilot Pro — after a two-month free trial. Along with Gemini Advanced, this subscription includes everything from the Google One Premium Plan, including 2TB of storage and a VPN. Subscribers will also be able to use Gemini in apps such as Gmail, Docs, Slides and Sheets in the near future (this is replacing Duet AI).

Of note, Google says it sought to mitigate concerns such as bias and unsafe content while building Gemini Advanced and other AI products. The company says it carried out "extensive trust and safety checks, including external red-teaming" (i.e. testing by third-party ethical hackers) on Gemini Advanced before refining the model with reinforcement learning and fine tuning.

This article originally appeared on Engadget at https://www.engadget.com/google-rebrands-its-bard-ai-chatbot-as-gemini-which-now-has-its-own-android-app-151303210.html?src=rss

How security experts unravel ransomware

Posted on 06/02/2024 by Katie Malone

Hackers use ransomware to go after every industry, charging as much money as they can to return access to a victim's files. It’s a lucrative business to be in. In the first six months of 2023, ransomware gangs bilked $449 million from their targets, even though most governments advise against paying ransoms. Increasingly, security professionals are coming together with law enforcement to provide free decryption tools — freeing locked files and eliminating the temptation for victims to pony up.

There are a couple main ways that ransomware decryptors go about coming up with tools: reverse engineering for mistakes, working with law enforcement and gathering publicly available encryption keys. The length of the process varies depending on how complex the code is, but it usually requires information on the encrypted files, unencrypted versions of the files and server information from the hacking group. “Just having the output encrypted file is usually useless. You need the sample itself, the executable file,” said Jakub Kroustek, malware research director at antivirus business Avast. It’s not easy, but does pay dividends to the impacted victims when it works.

First, we have to understand how encryption works. For a very basic example, let's say a piece of data might have started as a cognizable sentence, but appears like "J qsfgfs dbut up epht" once it's been encrypted. If we know that one of the unencrypted words in "J qsfgfs dbut up epht" is supposed to be "cats," we can start to determine what pattern was applied to the original text to get the encrypted result. In this case, it's just the standard English alphabet with each letter moved forward one place: A becomes B, B becomes C, and "I prefer cats to dogs" becomes the string of nonsense above. It’s much more complex for the sorts of encryption used by ransomware gangs, but the principle remains the same. The pattern of encryption is also known as the 'key', and by deducing the key, researchers can create a tool that can decrypt the files.

Some forms of encryption, like the Advanced Encryption Standard of 128, 192 or 256 bit keys, are virtually unbreakable. At its most advanced level, bits of unencrypted "plaintext" data, divided into chunks called "blocks," are put through 14 rounds of transformation, and then output in their encrypted — or "ciphertext" — form. “We don’t have the quantum computing technology yet that can break encryption technology,” said Jon Clay, vice president of threat intelligence at security software company Trend Micro. But luckily for victims, hackers don’t always use strong methods like AES to encrypt files.

While some cryptographic schemes are virtually uncrackable it’s a difficult science to perfect, and inexperienced hackers will likely make mistakes. If the hackers don’t apply a standard scheme, like AES, and instead opt to build their own, the researchers can then dig around for errors. Why would they do this? Mostly ego. “They want to do something themselves because they like it or they think it's better for speed purposes,” Jornt van der Wiel, a cybersecurity researcher at Kaspersky, said.

For example, here’s how Kaspersky decrypted the Yanluowang ransomware strain. It was a targeted strain aimed at specific companies, with an unknown list of victims. Yanluowang used the Sosemanuk stream cipher to encrypt data: a free-for-use process that encrypts the plaintext file one digit at a time. Then, it encrypted the key using an RSA algorithm, another type of encryption standard. But there was a flaw in the pattern. The researchers were able to compare the plaintext to the encrypted version, as explained above, and reverse engineer a decryption tool now made available for free. In fact, there are tons that have already been cracked by the No More Ransom project.

Ransomware decryptors will use their knowledge of software engineering and cryptography to get the ransomware key and, from there, create a decryption tool, according to Kroustek. More advanced cryptographic processes may require either brute forcing, or making educated guesses based on the information available. Sometimes hackers use a pseudo-random number generator to create the key. A true RNG will be random, duh, but that means it won’t be easily predicted. A pseudo-RNG, as explained by van der Wiel, may rely on an existing pattern in order to appear random when it's actually not — the pattern might be based on the time it was created, for example. If researchers know a portion of that, they can try different time values until they deduce the key.

But getting that key often relies on working with law enforcement to get more information about how the hacking groups work. If researchers are able to get the hacker’s IP address, they can request the local police to seize servers and get a memory dump of their contents. Or, if hackers have used a proxy server to obscure their location, police might use traffic analyzers like NetFlow to determine where the traffic goes and get the information from there, according to van der Wiel. The Budapest Convention on Cybercrime makes this possible across international borders because it lets police request an image of a server in another country urgently while they wait for the official request to go through.

The server provides information on the hacker’s activities, like who they might be targeting or their process for extorting a ransom. This can tell ransomware decryptors the process the hackers went through in order to encrypt the data, details about the encryption key or access to files that can help them reverse engineer the process. The researchers comb through the server logs for details in the same way you may help your friend dig up details on their Tinder date to make sure they’re legit, looking for clues or details about malicious patterns that can help suss out true intentions. Researchers may, for example, discover part of the plaintext file to compare to the encrypted file to begin the process of reverse engineering the key, or maybe they’ll find parts of the pseudo-RNG that can begin to explain the encryption pattern.

Working with law enforcement helped Cisco Talos create a decryption tool for the Babuk Tortilla ransomware. This version of ransomware targeted healthcare, manufacturing and national infrastructure, encrypting victims' devices and deleting valuable backups. Avast had already created a generic Babuk decryptor, but the Tortilla strain proved difficult to crack. The Dutch Police and Cisco Talos worked together to apprehend the person behind the strain, and gained access to the Tortilla decryptor in the process.

But often the easiest way to come up with these decryption tools stems from the ransomware gangs themselves. Maybe they’re retiring, or just feeling generous, but attackers will sometimes publicly release their encryption key. Security experts can then use the key to make a decryption tool and release that for victims to use going forward.

Generally, experts can’t share a lot about the process without giving ransomware gangs a leg up. If they divulge common mistakes, hackers can use that to easily improve their next ransomware attempts. If researchers tell us what encrypted files they’re working on now, gangs will know they’re on to them. But the best way to avoid paying is to be proactive. “If you’ve done a good job of backing up your data, you have a much higher opportunity to not have to pay,” said Clay.

This article originally appeared on Engadget at https://www.engadget.com/how-security-experts-unravel-ransomware-184531451.html?src=rss

Roblox adds real-time AI chat translation using its own language model

Posted on 06/02/2024 by Richard Lai

Currently serving over 70 million daily active users, Roblox is still going strong since its September 2006 launch — almost 18 years ago. The development team is now taking one step further to boost the platform's massive community, by way of providing real-time AI chat translation to connect gamers around the world. According to CTO Daniel Sturman, his team needed to build their own "unified, transformer-based translation LLM (large language model)" in order to seamlessly handle all 16 languages supported on Roblox, as well as to recognize Roblox-specific slangs and abbreviations (this writer just learned that "obby" refers to an obstacle course in the game).

As a result, the chat window always displays the conversation in the user's own tongue — with a small latency of around 100 milliseconds, so it's pretty much real time. You can also click on the translation icon on the left of each line to see it in its original language. Sturman claims that thanks to the language model's efficient architecture and iterative training, it "outperforms commercial translation APIs on Roblox content." The development team will later roll out a feedback tool to help improve translation quality, in addition to its ongoing updates with whatever new catchphrases it picks up on the platform.

Roblox built its own large language model to support real-time chat translation for all 16 languages on its platform. It recognizes Roblox-specific slang and abbreviations.

Roblox's translation efforts don't stop there. Sturman adds that his team is already looking into automatically translating "text on images, textures, 3D models" and more. As Roblox supports voice chat, the exec also teases the possibility of automatic voice chat translations, so gamers from around the world can seamlessly talk to one another in their own tongue on the platform. Given that Samsung already offers a similar feature via Galaxy AI, it probably won't be long before we hear another update from Roblox on this end.

This article originally appeared on Engadget at https://www.engadget.com/roblox-adds-real-time-ai-chat-translation-using-its-own-language-model-061929902.html?src=rss

Google is reportedly rebranding Bard to Gemini and plans to launch a dedicated app

Posted on 03/02/2024 by Cheyenne MacDonald

According to a document leaked on X, Google is planning to introduce some major changes to its Bard AI tool as soon as this coming week. The plans, which have not been publicly confirmed, reportedly include changing the Bard name to Gemini. It would make sense for Google to do so, if only for simplicity’s sake — the company introduced its new multimodal AI model, Gemini, at the end of 2023 and has begun integrating it into some of its products, including Bard.

The changelog shared by Android app developer Dylan Roussel is dated February 7, and also notes that the paid Gemini Advanced tier will become available at this time. It mentions a Gemini app for Android is “coming soon,” as well.

Google added a new changelog for Bard, and — oh boy — it's a big one!

The availability in Canada is awesome! That said I don't really understand the limitations with the app. That's disappointing as someone who lives in Europe.

Oh by the way... https://t.co/xM2snHVYJ9 is real. pic.twitter.com/QKgKrRjmM4
— Dylan Roussel (@evowizz) February 3, 2024

Per the document, Gemini Advanced will give users access to the Ultra 1.0 model of Gemini, which is “far more capable at highly complex tasks like coding, logical reasoning, following nuanced instructions, and creative collaboration.” It’ll be available in over 150 countries and optimized for the English language at the start. The changelog also says Gemini will expand to Canada with this release.

This article originally appeared on Engadget at https://www.engadget.com/google-is-reportedly-rebranding-bard-to-gemini-and-plans-to-launch-a-dedicated-app-204442265.html?src=rss

Niantic is bringing an AR skateboarding game to Apple Vision Pro

Posted on 01/02/2024 by Lawrence Bonk

Pokémon Go creator Niantic is bringing an AR skateboarding game to the Apple Vision Pro mixed-reality headset. The company teamed up with Reality Crisis, another player in the AR gaming space, to create Rodney Mullen’s SKATRIX. For the uninitiated, Mullen is a professional skateboarder who is credited with creating a number of iconic tricks, including the ollie and the kickflip.

This is the first augmented-reality skateboard game ever, unless you count using an actual skateboard to speed around town catching pocket monsters in Pokémon Go. Niantic says the gameplay will involve players navigating the real world to “explore and collect skatepark elements.” The game will use the same AR precision elements as the company’s other titles to “turn the world into an endless skatepark.” A gameplay demo shows an avatar skating in real-world locations like parking lots and inside of washing machines. However, there’s one really expensive elephant in the room.

The Apple Vision Pro costs $3,500 and isn’t exactly suited to removing from the living room, let alone the home. You’ll also look pretty stupid wearing Apple’s bulky ski goggles while out and about, not to mention it’d be mighty easy to snag it from your head and race away, perhaps on a real-life skateboard.

To that end, Niantic and Reality Labs are also bringing the game to standard mobile devices, including iOS and Android smartphones. Much of the gameplay is still under wraps, but Niantic promises that players will be able to share custom-made skateparks with other users on both mobile devices and mixed-reality headsets.

Rodney Mullen’s SKATRIX will be released sometime this year. As for the Apple Vision Pro, the device looks to be getting all kinds of apps. The headset will have access to Microsoft’s entire 365 productivity suite at launch, along with a Zoom app. All told, Apple says there will be 600 apps available for tomorrow’s release.

This article originally appeared on Engadget at https://www.engadget.com/niantic-is-bringing-an-ar-skateboarding-game-to-apple-vision-pro-183740925.html?src=rss

Block is reportedly laying off around 1,000 workers

Posted on 30/01/2024 by Kris Holt

Block is the latest notable tech company to lay off hundreds of workers, according to reports. CEO Jack Dorsey is said to have informed employees that the company is firing a "large number" of them, with Cash App, Square and the foundational (i.e. operations) teams bearing the brunt of the impact. According to a Business Insider source, Block is letting go nearly 1,000 people.

Dorsey reportedly wrote in his memo that the company is becoming leaner. It laid off around 40 people from the Tidal team in December. Last year, Block said it planned to limit its headcount to around 12,000 workers, a reduction from the around 13,000 it had in late 2023. Engadget has contacted Block for confirmation of the layoffs.

While it was initially expected that the layoffs would take place over a period of months, executives reportedly opted against that in favor carrying them out at the same time. "Why is so much happening in one single day? All of these teams were confident in the direction they're taking, and were ready to take action within the same 2-3 weeks," Dorsey is said to have written in his memo. "We decided it would be better to do [it] at once rather than arbitrarily space them out, which didn't seem fair to the individuals or to the company. When we know we need to take an action, we want to take it immediately, rather than let things linger on forever."

The tech industry has shed tens of thousands of workers over the last year or so, including thousands this month alone across companies including Unity, Twitch, Amazon, Meta, Microsoft, eBay and Google. It also emerged on Tuesday that PayPal is firing around 2,500 people.

This article originally appeared on Engadget at https://www.engadget.com/block-is-reportedly-laying-off-around-1000-workers-205319045.html?src=rss

GIZMODO.cz

Syndicated rss news

Tag Archives: Personal Investing Ideas & Strategies

Google’s Gemini 1.5 Pro is a new, more efficient AI model

Google’s Gemini 1.5 Pro is a new, more efficient AI model

Russian and North Korean hackers used OpenAI tools to hone cyberattacks

Uber, Lyft and DoorDash drivers are striking on February 14

Google rebrands its Bard AI chatbot as Gemini, which now has its own Android app

How security experts unravel ransomware

Roblox adds real-time AI chat translation using its own language model

Google is reportedly rebranding Bard to Gemini and plans to launch a dedicated app

Niantic is bringing an AR skateboarding game to Apple Vision Pro

Block is reportedly laying off around 1,000 workers