Posts with «author_name|will shanklin» label

The 256GB iPad mini is $120 off and down to a record-low price

Apple’s newest iPad mini is on sale for $120 off. You can get the 256GB version of the 6th-generation iPad mini (2021) for $529 on Amazon — a record low. If that’s more storage than you need, you can also get the 64GB model (usually $500) is $100 off.

Apple refreshed the iPad mini in 2021 with an “all-screen” Liquid Retina design, ditching the Home button and aligning the mini-tablet with the larger iPad Air’s modern design language. The iPad mini has an 8.3-inch display that supports TrueTone and covers the P3 color gamut. It has a 2266 x 1488 resolution (326 ppi).

This model runs on the A15 Bionic chip, also found in the iPhone 13 series. This tablet eschews Face ID for a Touch ID sensor on its power and sleep button. The iPad mini has a USB-C port for versatile charging and data transfers, and it works with the second-generation Apple Pencil for scribbling notes or digital sketching. Apple estimates up to 10 hours of battery life, which can vary depending on how you use it.

Photo by Valentina Palladino / Engadget

Apple will reportedly update its entire iPad lineup this year, and the latest rumors point to a possible late 2024 launch for a new iPad mini. If those reported plans pan out, this model may only be the newest for another six to eight months.

If you want a full-sized tablet, Walmart has the 5th-generation (10.9-inch) iPad Air for $120 off, too. You’ll pay only $449 for the 64GB variant. This model runs on an M1 chip with a 2360 x 1640 resolution (264 ppi) and an estimated 10 hours of battery life.

Follow @EngadgetDeals on Twitter and subscribe to the Engadget Deals newsletter for the latest tech deals and buying advice.

This article originally appeared on Engadget at

Google pauses Gemini’s ability to generate people after overcorrecting for diversity in historical images

Google said Thursday it’s pausing its Gemini chatbot’s ability to generate people. The move comes after viral social posts showed the AI tool overcorrecting for diversity, producing “historical” images of Nazis, America’s Founding Fathers and the Pope as people of color.

“We’re already working to address recent issues with Gemini’s image generation feature,” Google posted on X (via The New York Times). “While we do this, we’re going to pause the image generation of people and will re-release an improved version soon.”

The X user @JohnLu0x posted screenshots of Gemini’s results for the prompt, “Generate an image of a 1943 German Solidier.” (Their misspelling of “Soldier” was intentional to trick the AI into bypassing its content filters to generate otherwise blocked Nazi images.) The generated results appear to show Black, Asian and Indigenous soldiers wearing Nazi uniforms.

Still real real broke

— John L. (@JohnLu0x) February 21, 2024

Other social users criticized Gemini for producing images for the prompt, “Generate a glamour shot of a [ethnicity] couple.” It successfully spit out images when using “Chinese,” “Jewish” or “South African” prompts but refused to produce results for “white.” “I cannot fulfill your request due to the potential for perpetuating harmful stereotypes and biases associated with specific ethnicities or skin tones,” Gemini responded to the latter request.

“John L.,” who helped kickstart the backlash, theorizes that Google applied a well-intended but lazily tacked-on solution to a real problem. “Their system prompt to add diversity to portrayals of people isn’t very smart (it doesn’t account for gender in historically male roles like pope; doesn’t account for race in historical or national depictions),” the user posted. After the internet’s anti-“woke” brigade latched onto their posts, the user clarified that they support diverse representation but believe Google’s “stupid move” was that it failed to do so “in a nuanced way.”

Before pausing Gemini’s ability to produce people, Google wrote, “We’re working to improve these kinds of depictions immediately. Gemini’s Al image generation does generate a wide range of people. And that’s generally a good thing because people around the world use it. But it’s missing the mark here.”

The episode could be seen as a (much less subtle) callback to the launch of Bard in 2023. Google’s original AI chatbot got off to a rocky start when an advertisement for the chatbot on Twitter (now X) included an inaccurate “fact” about the James Webb Space Telescope.

As Google often does, it rebranded Bard in hopes of giving it a fresh start. Coinciding with a big performance and feature update, the company renamed the chatbot Gemini earlier this month as the company races to hold its ground against OpenAI’s ChatGPT and Microsoft Copilot — both of which pose an existential threat to its search engine (and, therefore, advertising revenue).

This article originally appeared on Engadget at

Your older S23 phone will get Samsung's Galaxy AI suite in late March

Samsung said Wednesday that the Galaxy S24’s AI features will arrive on last year’s phones (including foldables) and tablets in late March. In January, Engadget’s Sam Rutherford reported that the AI suite would soon be available on the Galaxy S23 series, Z Fold 5, Z Flip 5 and Tab S9. Today’s announcement makes that device list official while adding the more specific arrival window of late March 2024.

That group of 2023 devices will receive a software update next month with the AI features from the S24 series. Those include communication-based AI tricks like Chat Assist (adjusts message tone and translates messages), Live Translate (real-time voice and text translations) and Interpreter (split-screen translation for in-person conversations).

They’ll also get the productivity-based AI features Circle to Search (search for anything on your screen by drawing a ring around it), Note Assist (formatting, summaries and translations of notes), Browsing Assist (summaries of news articles) and Transcript Assist (transcribe and summarize meeting recordings).

Finally, image-based AI features coming to those devices include Generative Edit (reframe shots, move subjects around or delete and replace them), Edit Suggestion (recommended image tweaks), and Instant Slow-Mo (generate extra frames to transform a standard video into a slow-motion one).

Photo by Sam Rutherford / Engadget

The full list of devices receiving the update starting in March includes the Galaxy S23, Galaxy S23+, Galaxy S23 Ultra, Galaxy S23 FE, Galaxy Z Fold 5, Galaxy Z Flip 5 and Galaxy Tab S9. But Samsung says you can expect more devices to join them later. “This is only the beginning of Galaxy AI, as we plan to bring the experience to over 100 million Galaxy users within 2024 and continue to innovate ways to harness the unlimited possibilities of mobile AI,” Samsung President TM Roh wrote in a press release.

We were mostly impressed with the AI features in our Galaxy S24 Ultra review. “While harnessing AI might not be a super exciting development now that everyone and their grandmother is trying to shoehorn it into everything, it does make the S24 Ultra a more powerful and well-rounded handset,” Engadget’s Sam Rutherford wrote in January.

Although he noticed a few hiccups in the AI tools at launch, he found most of them to be a genuinely helpful complement to the phone’s high-end hardware. “Samsung finally has an answer to the sophisticated features that were previously only available from the Pixel family,” he wrote. “Sure, the S24’s tools aren’t quite as polished as Google’s offerings, but they get you 80 to 90 percent of the way there.”

This article originally appeared on Engadget at

FTC concludes Twitter didn’t violate data security rules, in spite of Musk's orders

The Federal Trade Commission (FTC) concluded Elon Musk ordered Twitter (now X) employees to take actions that would have violated an FTC consent decree regarding consumers’ data privacy and security. The investigation arose from the late 2022 episode informally known as “The Twitter Files,” where Musk ordered staff to let outside writers access internal documents from the company’s systems. However, the FTC says Twitter security veterans “took appropriate measures to protect consumers’ private information,” likely sparing Musk’s company from government repercussions by ignoring his directive.

FTC Chair Lina Khan discussed the conclusions in a public letter sent Tuesday to House Judiciary Committee Chair Jim Jordan (via The Washington Post). Jordan and his Republican colleagues have tried to turn the FTC’s investigation into a political wedge issue, framing the inquiry as a free speech violation — perhaps to shore up GOP support from Musk’s legion of rabid supporters. Jordan and his peers previously described the investigation as “attempts to harass, intimidate, and target an American business.”

Khan’s response to Jordan adopts a tone resembling that of a patient teacher explaining the nuance of a complicated situation to a child who insists on seeing simplistic absolutes. “FTC staff efforts to ensure Twitter was in compliance with the Order were appropriate and necessary, especially given Twitter’s history of privacy and security lapses and the fact that it had previously violated the 2011 FTC Order,” Khan wrote.

“When a firm has a history of repeat offenses, the FTC takes particular care to ensure compliance with its orders,” the FTC Chair wrote.

House Judiciary Chair Jim Jordan (R-OH)
Tom Williams via Getty Images

The FTC’s investigation stemmed from allegations that Musk, newly minted as Twitter’s owner, ordered staff to give outside writers “full access to everything” in late 2022. Had staff obeyed Musk’s directive, the company likely would have violated a settlement with the FTC (originally from 2011 but updated in 2022) requiring the company to tightly restrict access to consumer data.

In November 2022, the FTC said publicly it was monitoring Twitter’s developments following Musk’s acquisition with “deep concern.” That followed the resignation of chief information security officer Lea Kissner and other members of the company’s data governance committee. They expressed concerns that Musk’s launch of a new account verification system didn’t give them adequate time to deploy security reviews required by the FTC.

Ultimately, Twitter security veterans ignored Musk’s “full access to everything” order. “Longtime information security employees at Twitter intervened and implemented safeguards to mitigate the risks,” Khan wrote in the letter. “The FTC’s investigation confirmed that staff was right to be concerned, given that Twitter’s new CEO had directed employees to take actions that would have violated the FTC’s Order.”

FTC Chair Lina Khan
Slaven Vlasic via Getty Images

Rather than supplying outside writers with the “full access” Musk wanted them to have, Twitter employees accessed the systems and relayed select information to the group of outsiders. “Ultimately the third-party individuals did not receive direct access to Twitter’s systems, but instead worked with other company employees who accessed the systems on the individuals’ behalf,” Khan wrote.

The FTC says it will continue to monitor X’s adherence to the order. “When we heard credible public reports of potential violations of protections for Twitter users’ data, we moved swiftly to investigate,” FTC spokesman Douglas Farrar said in a statement to The Washington Post. “The order remains in place and the FTC continues to deploy the order’s tools to protect Twitter users’ data and ensure the company remains in compliance.”

This article originally appeared on Engadget at

Apple says the iPhone 15’s battery has double the promised lifespan

Apple has updated the iPhone 15’s battery lifespan. The company said on Tuesday its latest iPhones can retain 80 percent of their original charging capacity after 1,000 cycles — double the company’s previous estimate — without any new hardware or software updates. Not so coincidentally, the change will arrive in time for upcoming EU regulations that will assign an energy grade for phones’ battery longevity.

Before today, Apple’s online support documents quoted iPhone batteries as maintaining 80 percent of their original full charge after 500 cycles. But after the company retested long-term battery health in its 2023 smartphones — iPhone 15, iPhone 15 Plus, iPhone 15 Pro and iPhone 15 Pro Max — it found they can retain 80 percent capacity after at least 1,000 cycles. The company said its support documents will be updated on Tuesday to reflect the new estimate.

Apple says its testing involved charging and draining the batteries 1,000 times under specific conditions and a suite of tests reflecting common uses. As for how the estimate doubled without any physical or software changes, the company attributes the upgrade to continual improvements to its battery components and iOS power management.

For older (pre-2023) iPhones, the original estimate of retaining 80 percent capacity after 500 charge cycles still stands — at least for now. Apple said it’s looking into whether older models’ estimates need to be updated.


Starting in June 2025, smartphone and tablet manufacturers doing business in the EU will be assigned a grade (A to G) indicating their energy efficiency, battery longevity, protection from dust and water and resistance to accidental drops. The battery longevity portion of the grade requires at least 800 charging cycles while retaining at least 80 percent of their original capacity, helping explain why Apple began retesting its devices’ long-term health.

For ideal battery longevity, Apple recommends keeping your phone in temperatures between 62 and 72 degrees Fahrenheit when possible. You’ll also want to avoid charging or leaving your handset in hot environments (95 degrees or warmer) while trying to avoid much direct sun exposure. In addition, if you’re storing an old iPhone long-term, it’s best to leave it half-charged.

In addition, Apple will soon move one of your phone’s most essential battery metrics to a different part of iOS settings. Beginning in iOS 17.4, currently in beta, the battery cycle count will move from Settings > General > About to a more logical home under Battery Health (under Settings > Battery). Apple introduced the battery cycle count to its mobile software in iOS 17, which was launched last fall.

This article originally appeared on Engadget at

Acer’s latest Swift laptops have AMD 8040 chips with Ryzen AI support

Acer unveiled a pair of AMD Ryzen 8040 series laptops on Tuesday. Unsurprisingly, given their chips’ dedicated neural processing units (NPU), the company is marketing the 2024 Acer Swift Edge 16 and Swift Go 14 as AI workhorses. The Windows 11 machines support OLED displays, Radeon 780M graphics and 32GB of RAM.

The Ryzen 8040 chip series, revealed in December, has a dedicated AI Engine that AMD claims makes it up to 1.4 times faster than its predecessors in Llama 2 and AI vision model performance. Acer says the Swift Edge 16 and Swift Go 14 will deploy the NPU for AI-related tasks like PurifiedVoice (remove background noise in calls and recordings) and PurifiedView (blurring backgrounds in images and correcting your eyes’ positioning on video calls).


Like most new Windows machines (including Acer’s models launched at CES 2024), the laptops have a dedicated Microsoft Copilot button on their keyboards for quick ChatGPT-like AI queries. (Copilot taps into Microsoft and OpenAI’s servers through the cloud rather than using the machines’ on-device NPU.)

Both machines’ AMD chips include Microsoft’s Pluton security co-processor. Introduced in 2020 through a partnership with AMD and Intel, it bakes security directly into the processor, helping protect your credentials, encryption keys and personal data from hackers.

Acer Swift Edge 16


Acer markets the Acer Swift Edge 16 as optimizing performance without sacrificing portability. All device variants have a 16-inch OLED panel with WQXGA (3200 x 2000) resolution and a 120Hz frame rate with less than a 0.2ms response time. It supports 100 percent of the DCI-P3 color gamut and VESA DisplayHDR True Black 500 certification for (what should be) accurate and nuanced visuals.

The laptop’s Ryzen 7 8840U processor is joined by AMD Radeon 780M integrated graphics. The notebook has up to 32GB of LPDDR5 RAM and 2TB PCIe 4.0 SSD storage. Depending on your configuration, it supports Wi-Fi 7 or Wi-Fi 6E.

The Swift Edge 16 weighs 2.71 lbs and is 12.95mm tall. It’s encased in a magnesium-aluminum alloy chassis and has a 54 Wh battery. It includes two USB-C ports (with USB4 speeds of up to 40Gbps), two USB-A, one HDMI 2.1 and a microSD slot.

Acer Swift Go 14


The smaller of the pair, the Acer Swift Go 14, has up to an AMD Ryzen 9 8945HS octa-core processor with AMD Radeon 780M graphics. (Cheaper configurations will trade that for a Ryzen 7 8845HS or Ryzen 5 8645HS.) The machine maxes out at 32GB LPDDR5X RAM and up to 2TB PCIe Gen 4 SSD.

The top-shelf 14-inch variant will have a WQXGA (2880x1800) OLED display with a 90Hz refresh rate. Another version will swap that for a 1920 x 1200 IPS panel with touchscreen capabilities. The device’s hinge can extend to 180 degrees, and its glass trackpad is 44 percent larger than the previous model’s. It supports Wi-Fi 6E.

Thanks to its smaller footprint, Acer markets the Swift Go 14 as the more portable version. However, it’s about seven percent heavier than its 16-inch counterpart — at 2.91 lbs. It ships in configurations with either a 65 or 50 Wh battery.

Both models have 1440p QHD webcams, although only the Swift Go 14 is listed as having a privacy shutter. The smaller model is lighter on ports than the 16-inch model, forgoing the pair of USB-A connections in the larger one. The Swift Go 14 has two USB-C ports (both supporting USB4 speeds), an HDMI 2.1 connector and a microSD slot.

Pricing and availability

The Swift Edge 16 launches in March in North America, starting at $1,300. Meanwhile, the Swift Go 14 follows in April, starting at $700. Apart from their entry-level models, Acer hasn’t yet detailed how pricing will break down across various configurations.

This article originally appeared on Engadget at

Microsoft, OpenAI, Google and others agree to combat election-related deepfakes

A coalition of 20 tech companies signed an agreement Friday to help prevent AI deepfakes in the critical 2024 elections taking place in more than 40 countries. OpenAI, Google, Meta, Amazon, Adobe and X are among the businesses joining the pact to prevent and combat AI-generated content that could influence voters. However, the agreement’s vague language and lack of binding enforcement call into question whether it goes far enough.

The list of companies signing the “Tech Accord to Combat Deceptive Use of AI in 2024 Elections” includes those that create and distribute AI models, as well as social platforms where the deepfakes are most likely to pop up. The signees are Adobe, Amazon, Anthropic, Arm, ElevenLabs, Google, IBM, Inflection AI, LinkedIn, McAfee, Meta, Microsoft, Nota, OpenAI, Snap Inc., Stability AI, TikTok, Trend Micro, Truepic and X (formerly Twitter).

The group describes the agreement as “a set of commitments to deploy technology countering harmful AI-generated content meant to deceive voters.” The signees have agreed to the following eight commitments:

  • Developing and implementing technology to mitigate risks related to Deceptive AI Election content, including open-source tools where appropriate

  • Assessing models in scope of this accord to understand the risks they may present regarding Deceptive AI Election Content

  • Seeking to detect the distribution of this content on their platforms

  • Seeking to appropriately address this content detected on their platforms

  • Fostering cross-industry resilience to deceptive AI election content

  • Providing transparency to the public regarding how the company addresses it

  • Continuing to engage with a diverse set of global civil society organizations, academics

  • Supporting efforts to foster public awareness, media literacy, and all-of-society resilience

The accord will apply to AI-generated audio, video and images. It addresses content that “deceptively fake or alter the appearance, voice, or actions of political candidates, election officials, and other key stakeholders in a democratic election, or that provide false information to voters about when, where, and how they can vote.”

The signees say they will work together to create and share tools to detect and address the online distribution of deepfakes. In addition, they plan to drive educational campaigns and “provide transparency” to users.

OpenAI CEO Sam Altman

OpenAI, one of the signees, already said last month it plans to suppress election-related misinformation worldwide. Images generated with the company’s DALL-E 3 tool will be encoded with a classifier providing a digital watermark to clarify their origin as AI-generated pictures. The ChatGPT maker said it would also work with journalists, researchers and platforms for feedback on its provenance classifier. It also plans to prevent chatbots from impersonating candidates.

“We’re committed to protecting the integrity of elections by enforcing policies that prevent abuse and improving transparency around AI-generated content,” Anna Makanju, Vice President of Global Affairs at OpenAI, wrote in the group’s joint press release. “We look forward to working with industry partners, civil society leaders and governments around the world to help safeguard elections from deceptive AI use.”

Notably absent from the list is Midjourney, the company with an AI image generator (of the same name) that currently produces some of the most convincing fake photos. However, the company said earlier this month it would consider banning political generations altogether during election season. Last year, Midjourney was used to create a viral fake image of Pope Benedict unexpectedly strutting down the street with a puffy white jacket. One of Midjourney’s closest competitors, Stability AI (makers of the open-source Stable Diffusion), did participate. Engadget contacted Midjourney for comment about its absence, and we’ll update this article if we hear back.

Only Apple is absent among Silicon Valley’s “Big Five.” However, that may be explained by the fact that the iPhone maker hasn’t yet launched any generative AI products, nor does it host a social media platform where deepfakes could be distributed. Regardless, we contacted Apple PR for clarification but hadn’t heard back at the time of publication.

Although the general principles the 20 companies agreed to sound like a promising start, it remains to be seen whether a loose set of agreements without binding enforcement will be enough to combat a nightmare scenario where the world’s bad actors use generative AI to sway public opinion and elect aggressively anti-democratic candidates — in the US and elsewhere.

“The language isn’t quite as strong as one might have expected,” Rachel Orey, senior associate director of the Elections Project at the Bipartisan Policy Center, told The Associated Press on Friday. “I think we should give credit where credit is due, and acknowledge that the companies do have a vested interest in their tools not being used to undermine free and fair elections. That said, it is voluntary, and we’ll be keeping an eye on whether they follow through.”

AI-generated deepfakes have already been used in the US Presidential Election. As early as April 2023, the Republican National Committee (RNC) ran an ad using AI-generated images of President Joe Biden and Vice President Kamala Harris. The campaign for Ron DeSantis, who has since dropped out of the GOP primary, followed with AI-generated images of rival and likely nominee Donald Trump in June 2023. Both included easy-to-miss disclaimers that the images were AI-generated.

In January, New Hampshire voters were greeted with a robocall of an AI-generated impersonation of President Biden’s voice — urging them not to vote.
Anadolu via Getty Images

In January, an AI-generated deepfake of President Biden’s voice was used by two Texas-based companies to robocall New Hampshire voters, urging them not to vote in the state’s primary on January 23. The clip, generated using ElevenLabs’ voice cloning tool, reached up to 25,000 NH voters, according to the state’s attorney general. ElevenLabs is among the pact’s signees.

The Federal Communication Commission (FCC) acted quickly to prevent further abuses of voice-cloning tech in fake campaign calls. Earlier this month, it voted unanimously to ban AI-generated robocalls. The (seemingly eternally deadlocked) US Congress hasn’t passed any AI legislation. In December, the European Union (EU) agreed on an expansive AI Act safety development bill that could influence other nations’ regulatory efforts.

“As society embraces the benefits of AI, we have a responsibility to help ensure these tools don’t become weaponized in elections,” Microsoft Vice Chair and President Brad Smith wrote in a press release. “AI didn’t create election deception, but we must ensure it doesn’t help deception flourish.”

This article originally appeared on Engadget at

8BitDo’s Nintendo-inspired Retro Mechanical Keyboard is cheaper than ever right now

8BitDo’s Nintendo-inspired wireless mechanical keyboard is currently on sale on Amazon for up to 15 percent off. Launched in 2023, the Retro Mechanical Keyboard is an ode to Nintendo’s consoles from the 1980s: the Nintendo Entertainment System (NES) and Nintendo Famicom. Right now, you can order the Fami Edition (Famicom-inspired) keyboard for $85.49 or the N Edition (NES-inspired) model for $90. Both are record-low prices for the popular accessory.

The 8BitDo Retro Mechanical Keyboard has colors that match Nintendo’s 8-bit consoles nearly perfectly. The N Edition samples the NES’ familiar off-white, dark gray, black and red color scheme. Meanwhile, the Fami Edition draws from the Famicom’s white and crimson. The latter even honors the Famicom’s regional status with Japanese characters below the keys’ English markings.

The wireless keyboards include a separate “Super Buttons” accessory — two huge red or crimson buttons just begging to be mashed. They connect to the keyboard through its 3.5mm jack and are programmable through 8BitDo’s Ultimate Software.


The keyboard’s power indicator and dials fit the accessory’s old-school motif, and you can even customize the keys using Kailh Box White Switches V2. You can also swap out the Super Buttons, which use Gatreon Green Switches.

The keyboard has 87 keys and works in Bluetooth, 2.4GHz wireless and USB wired modes. Its 2,000mAh battery lasts an estimated 200 hours of use and takes about four hours to charge. Although the accessories are only advertised as working with Windows and Android, 8BitDo told Engadget last year that it works with macOS, too.

Follow @EngadgetDeals on Twitter and subscribe to the Engadget Deals newsletter for the latest tech deals and buying advice.

This article originally appeared on Engadget at

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

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).

Google CEO Sundar Pichai
ALAIN JOCARD via Getty Images

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.

Google DeepMind CEO Demis Hassabis
Joy Malone via Getty Images

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

Sony wants its PlayStation exclusives to come to PC earlier

Sony may shrink the gap between the launches of its PlayStation exclusives and PC ports. Company president Hiroki Totoki suggested in a post-earnings call Q&A session Wednesday (via VGC) that he wants PlayStation to go “aggressive on improving our margin performance,” with “multi-platform” games playing a significant role. He clarified in the talk that, by multi-platform, he meant on PlayStation and PC — not Xbox or Switch.

When asked about Sony’s profits not keeping up with increasing gross income, he said hardware and first-party games were two areas of focus. He noted that hardware cost reduction this console cycle was “difficult to come by,” suggesting we won’t see any permanent console price drops.

“I personally think there are opportunities out there for improvement of margin, so I would like to go aggressive on improving our margin performance,” he continued. Totoki hinted one way to get there is to cash in more on its (often critically acclaimed and commercially successful) PlayStation Studios titles, like Marvel’s Spider-Man 2 and God of War: Ragnarok.

“In the past, we wanted to popularize consoles, and a first-party title’s main purpose was to make the console popular,” Totoki said in the Q&A. “This is true, but there’s a synergy to it, so if you have strong first-party content — not only on our console but also other platforms, like computers — a first-party [game] can be grown with multi-platform, and that can help operating profit to improve, so that’s another one we want to proactively work on.”

The Last of Us Part I
Naughty Dog / Sony

That’s a clear shift from PlayStation Studios head Herman Hulst’s thoughts in 2022. He said then that PC gamers would have to wait “at least a year” before seeing first-party PlayStation games (minus live service titles) on their computers. God of War (2018) and the first Marvel’s Spider-Man had about a four-year gap between their PS4 and PC launches. The latter’s Miles Morales spin-off saw about a two-year turnaround.

On February 8, Sony launched Helldivers 2 on PS5 and Windows simultaneously. VGC notes the game led to PlayStation Studios’ highest concurrent Steam player count — beating God of War (2018), The Last of Us Part I and Horizon Zero Dawn. Helldivers 2 was developed by Arrowhead Games with Sony Interactive Entertainment publishing.

It isn’t clear if Totoki meant we can expect future PlayStation tentpoles like the upcoming Wolverine game or the ever-popular Spider-Man, God of War or various Naughty Dog franchises to appear on PC on the same day as console. But a strategy shift is underway regardless, and Totoki will have the leverage to put the plan into action: He takes over for Jim Ryan as interim CEO of Sony Interactive Entertainment in April.

This article originally appeared on Engadget at