(Image Credit: spotify)

The Gemini extension was first spotted in the code of the Google app in June, and now it’s rolling out to compatible Android devices. According to an update Google has just Gemini Spotify rollout, users can now ask the Gemini app to search or play music on Spotify by specifying the song, album, artist, playlist, and more. Google already offers similar functionality with Gemini integration for its YouTube Music platform.

Google Gemini Spotify feature

Spotify is the second third-party app to integrate with Google Gemini AI assistant. Previously, Google launched a Gemini extension for WhatsApp, allowing users to send messages or make calls using WhatsApp through Gemini prompts. Additionally, Google has begun rolling out a Utilities extension for Gemini, which adds functionality such as opening specific apps, websites, or Android settings.

Users should know that if they are already linked to another music service like YouTube Music, they shall need to verbally say or type in which service they want Gemini to use as part of initial request, and from there, it will default to whatever was used last. Also users would not be able to use the Spotify extension inside Google Messages, the Gemini web app, or the Gemini app on iOS, and it only works while Gemini’s language is set to English to start.

Spotify Gemini extension default assistant

Google clarified that the Spotify extension works only within the Gemini app or when Gemini is set as the default assistant. Currently, this feature is not available for the Gemini app on iOS. To use the Spotify extension, users must link their Spotify account to their Google Account. Spotify will play specific song requests only for users with a Spotify Premium subscription.

How to enable Gemini extension on Spotify

Gemini Extensions require the “Gemini Apps Activity” to be enabled. To enable it, tap on your Google Account menu in the Gemini app, select “Gemini Apps Activity” and turn it on. To connect Spotify to the Gemini extension, users will first need to log into their Spotify account and then link to their Google account. Once done, users can open the Gemini app, tap on their profile picture, and go to Extensions. There, they can manually toggle on the Spotify option.

How to use the Google Gemini extension on Spotify

Link your Spotify account to your Google Account through the Spotify app and ensure you have an active Spotify Premium subscription.

Open the Google Gemini app and tap on the Google Account menu in the top-right corner, navigate to the “Extensions” option and enable the Spotify extension. Once enabled, you can use Gemini to play, search, or browse music on Spotify.

Google Gemini extension rollout

While Google has announced rolling out the Gemini extension, it may take up to two weeks before it is available to all users globally.

The post Spotify Joins the Google Gemini Ecosystem: What You Need to Know appeared first on Technowize.

(Image Credit: apple)

It seems that iPhone 17 Air prototypes are between 5 and 6 millimeters thick, which could possibly make it thinner than the iPhone 6, released a decade ago. While the iPhone 16 is 7.8mm thick, this could make the iPhone 17 Air as thin as the new M4 iPad Pro.

iPhone 17 Air design compromises

Apple iPhone 17 Air changes shall be major to keep it ultra-slim. Apple will have to sacrifice several crucial features to achieve this design. As per report Apple engineers are “finding it hard to fit the battery and thermal materials into the device,” which means it will likely have to add a smaller battery to this iPhone. Besides, a major compromise on the iPhone 17 Air might be a single earpiece speaker.

Apple ultra-thin iPhone 17 design missing features

Apple aims to keep the iPhone 17 Air model’s thickness between five and six millimeters by making several internal design changes.

Prototype models of the iPhone 17 Air are expected to measure between 5mm and 6mm in thickness, presenting challenges for fitting components within the compact frame. Early supply chain data indicates difficulties in sourcing appropriate batteries and thermal materials to support the ultra-thin design. To accommodate these constraints, Apple may replace the traditional stereo speaker setup with a single speaker for this model.

iPhone 17 Air changes to camera

Apple may also simplify the camera system for the iPhone 17 Air. This model is anticipated to feature a single rear camera housed in a large, centered camera bump. Though it is a premium phone, the iPhone 17 Air model may lack some high-end camera specifications, such as the periscopic telephoto lens with 5x zoom, which is likely to remain exclusive to the Pro variants. Earlier speculation suggested the inclusion of the telephoto lens in the Air model, but this now seems unlikely.

Apple’s in-house 5G modem

The report also suggests that the iPhone 17 Air could be one of the first Apple devices to use Apple’s in-house 5G modem. This chip is reportedly less powerful than the Qualcomm 5G modems currently used in iPhones, offering lower peak speeds, slightly reduced reliability, and no support for mmWave 5G technology. Apple is still grappling with whether to include a SIM tray in the iPhone 17 Air. While the company has phased out physical SIM cards in markets like the U.S., regions such as China still require a SIM card slot, adding to the design challenges.

Apple ultra-thin iPhone 17 design is it worth the compromise?

The iPhone 17 Air represents a bold step toward creating an ultra-thin device, but Apple will need to make several trade-offs to achieve this design. While the slim profile will appeal to those seeking a sleek and lightweight phone, compromises in battery life, audio quality, camera functionality, and 5G performance may limit its appeal for users who prioritize these features. It would be interesting to see if these trade-offs are worth it for an all-new iPhone design.

The post Slimmer Than Ever: Will the iPhone 17 Air’s Design Compromises Be Worth It? appeared first on Technowize.

(Image Credit: huawei

Xu Zhijun said that while more than 15,000 Harmony-based apps are already available for basic consumer needs, there is a need for more personalized and boutique apps. According to the tech giant’s Chairman, the ecosystem requires more personalized and boutique apps.

Huawei’s HarmonyOS 100, 000 apps target

“Based on our analysis, for the Harmony ecosystem to be mature in meeting consumer needs, 100,000 apps is the milestone, and that is the key objective over the next six to 12 months,” Xu said in a speech posted on the WeChat messaging app.

Xu also invited developers to enrich application offerings, as well as encourage government agencies, state companies, and social organizations to adopt Harmony as their operating system.

The ambitious HarmonyOS apps growth target highlights the urgency in developing home-grown technologies. This is as China faces elevated tensions with the U.S. in areas ranging from trade to technology as President-elect Donald Trump threatens to be tougher on China.

U.S. sanctions on Huawei

Huawei launched its operating system five years ago after U.S. sanctions cut off support for Google’s Android. The Shenzhen-based company, which sells products ranging from smartphones to laptops, later developed an open-source version of the Harmony system.

With the U.S. sanctions, “Huawei has been forced to accelerate developing its own operating system,” Xu said. Although much progress has been made, “for any operation system, no matter how advanced it is, it would be of no value if no one uses it.”

Huawei wants impossible possible

Huawei unveiled Harmony in August 2019, three months after U.S. placed it under trade restrictions over alleged security concerns. The company denies its equipment poses a risk.

“No way back leads to victory,” Xu said. “Huawei will unwaveringly invest in developing the Harmony ecosystem, and strive to make the impossible possible.”

The initiative underscores Huawei’s strategy to reduce dependence on foreign technology while overcoming challenges posed by sanctions.

Huawei HarmonyOS apps offerings

Xu expressed hope that developers could work hard to enrich app offerings and called on government agencies, state companies and social organizations to use Harmony as their operating system at work.

Xu wants consumers to be tolerant of the Huawei Harmony OS’s immaturity, saying, “The more people use it, the more quickly it will become mature.”

Road ahead for Huawei

Huawei’s efforts to expand its app ecosystem align with broader efforts in China to achieve technological independence. This move reflects a strategic response to escalating U.S.-China tensions, particularly in technology and trade. Achieving the 100,000-app goal would signify a major step forward in Huawei’s vision for HarmonyOS. It would transform it into a competitive alternative in the global tech landscape.

The post Huawei Sets Ambitious Goal for Growth, 100,000 Apps on HarmonyOS appeared first on Technowize.

(Image Credit: stellantis)

The Franco-Italian company revealed details about its STLA Frame platform, which will support full-size trucks and SUVs. Platforms are thought of as a skateboard on which many different types of vehicles can be built, and include important electrical and mechanical components of the car.

Stellantis EV hybrid new platform

“We are very focused on the execution of our plan, despite all the difficult challenges that the industry is facing,” said CEO Carlos Tavares on a call with reporters.

Stellantis Ram delayed

Stellantis is delaying production of its electric Ram pickups until the first half of 2025 from this year, citing the need to ensure quality as per Tavares. “We are just facing a very significant amount of workload,” he said.

Global automakers openion on new platform

Automakers in Detroit and elsewhere rushed into building EV-manufacturing capacity over the last two years, only for demand to grow more slowly than anticipated.

The decision of whether to focus on platforms that support EVs versus those that include flexibility for hybrids or gasoline-powered vehicles has split automakers. Ford Motor has leaned into selling hybrid vehicles, while General Motor has focused on battery-powered models after investing more up front into building its own EV platform. GM will start offering plug-in hybrids in 2027, it said.

Stellantis’ goal

Stellantis is mainly offering plug-in hybrids in the U.S. for now, but plans to ramp up EV sales in the coming years, chasing a goal of 100% battery electric car sales in Europe and 50% electric passenger car and light-duty truck sales in the U.S. by 2030.

The global industry faces even greater uncertainty around EV demand if U.S. President-elect carries out plans to unravel various incentives around EV production and sales.

Some of Stellantis’ plug-in hybrids, including the Jeep Grand Cherokee, qualify for half of that credit.

The carmaker is confronting slipping sales in North America, which has historically brought in sizable profits thanks to sales of its popular Jeep and Ram vehicles. Tavares changed his management team in an effort to address its high vehicle inventory and slumping stock price, and he is planning to retire at the end of his contract in 2026.

Stellantis’ shares are down about 40% this year.

Stellantis Frame platform

The company announced in 2021 it would design the Frame platform in addition to large, medium and smaller foundations to accommodate different vehicles in its lineup. Ram and Jeep vehicles are the first to use the Frame version, Stellantis said on Tuesday.

Battery electric vehicles on Frame will have up to 500 miles (805 km) of range, and it will offer a 14,000-pound (6,350-kg) towing capacity, the company said. It will also support hydrogen and extended-range electric vehicles.

The post Stellantis Launches Hybrid and EV Capable Platform, Ram Electric Pickup Delayed appeared first on Technowize.

During the launch of GPT-4o in May 2024, the company promised vision capabilities. This is supposed to take ChatGPT beyond just text and voice, allowing it to understand and interact with the real world through live video. It appears that OpenAI has worked towards that feature, and it is ready to come out of the alpha phase.

ChatGPT vision feature rollout

According to multiple strings of codes, the ChatGPT live vision feature capabilities in the Advanced Voice Mode may be gearing up for a wider beta rollout. The strings with live vision feature were spotted in the latest ChatGPT v1.2024.317 beta build.

The new “Live Camera” feature, which is expected to arrive soon in a beta version, will allow ChatGPT to “see” and engage with your surroundings in real-time. Users will be able to activate the feature by tapping a camera icon within the app, which will trigger the AI to view and comment on whatever it sees through the device’s camera.

ChatGPT live video feature

As announced during the GPT4o rollout, the feature is built on the Advanced Voice Mode, which lets ChatGPT have natural, flowing conversations. The addition of vision capabilities means that ChatGPT can now recognize objects and people, remember names, and even make associations between items in the environment.

ChatGPT’s Advanced Voice Mode with vision capabilities

During the GPT-4o announcement, OpenAI showcased the Advanced Voice Mode with vision capabilities. In the demo, the ChatGPT seamlessly recognized the subject in the camera, remembered its name, recognized the ball, and associated it with the dog. The demo was quite impressive, considering the AI assistant had little information the user had to input to elicit the responses.

ChatGPT vision capabilities

For example, in a demo during the GPT-4o event, ChatGPT was shown identifying a dog, recalling its name, recognizing a ball, and understanding the game of fetch, all without the user needing to provide specific input beyond the initial setup.

Now, thanks to new code strings spotted in the latest beta update by Android Authority, it looks like OpenAI is preparing for a wider rollout of the ChatGPT with vision capabilities. OpenAI made sure to add a cautionary warning, stating users are advised not to rely on the “Live Camera” for important decisions, like navigation or anything that could impact health or safety.

ChatGPT Live Camera availability

While it’s still in the beta stage, we expect the “Live Camera” feature to soon be available to ChatGPT Plus subscribers and possibly other paid tiers. If this ends up happening, it would give it a step up over competitors like Google’s Gemini, whose closest offering to this is Google Lens which doesn’t offer live capabilities like ChatGPT is claiming.

A feather in ChatGPT’s hat

ChatGPT has about 10 million paying users which is nothing compared to Google’s 100 million Google One subscribers that have access to Gemini Advanced. With more additions like this, ChatGPT could give many users valid reasons to jump ship from whatever chatbot they currently serve to this.

The post OpenAI Expands ChatGPT with Vision Capabilities for Real World Engagement appeared first on Technowize.

(Image Credit: apple)

The list of vintage iPhone models now includes the iPhone 6s Plus and iPhone XS Max. The smaller iPhone XS and iPhone 6s still count among the active models. But as reported, the iPhone 6s and iPhone 6s Plus with 32GB configurations were previously moved to the obsolete list. As for the Apple Watch, the Apple Watch Series 2 models made from aluminum and stainless steel were moved to the obsolete list.

Apple vintage products

Vintage and obsolete status refers to whether Apple and Apple-certified repair services can fix devices with expired warranties and supply parts for those models. A vintage model is when it “stopped distributing them for sale more than five and less than seven years ago,” as per Apple.

Apple Vintage and Obsolete List

The latest Apple products are making their way to the vintage and obsolete lists and these include some of the popular iPhone and Apple Watch models.

Apple iPhones

Here are the details of the vintage products

• iPhone Xs Max
• iPhone 6s Plus

The mentioning of the iPhone 6s Plus with the Xs Max is peculiar, mostly because the former is 3 years older than when the Xs Max was launched in 2018. Typically, Apple is very fixed about its decision to make a product obsolete but the 6s Plus moving to the vintage list now means the device has outlived its purpose and is finally ready to end its services in the market.

Obsolete list

• Apple Watch Series 2 Aluminum (both 38mm and 42mm sizes)
• Apple Watch Series 2 Stainless Steel (both 38mm and 42mm sizes)

What the Apple vintage and obsolete means?

Apple’s vintage is defined as a product which has stopped selling through the company for more than 5 years but less than 7 years. Similarly, the obsolete list includes products from Apple that have stopped selling in the market for more than 7 years. These are not the first Apple products to be added to this list and surely won’t be the last. But the fact that a 2018 iPhone model is now part of the list shows you how quickly the paradigm of the tech space evolves and new products show their dated value.

Apple vintage repair

Those models can be repaired as long as supplies for relevant parts are available. Obsolete models are devices that haven’t been distributed for over seven years. Hardware for obsolete products are discontinued and service providers can no longer order parts for them.

Active Apple devices that are not on the vintage or obsolete lists are still supported by OS updates. However, some new capabilities, like Apple Intelligence, aren’t compatible with the older still-active models. The iPhones that support Apple Intelligence are the full iPhone 16 generation and iPhone 15 Pro and iPhone 15 Pro Max.

The post Apple Adds iPhone 6s Plus, XS Max to Vintage List and Watch Series 2 to Obsolete appeared first on Technowize.

(Image Credit: nvidia)

Nvidia Corporation’s new Blackwell AI chips are reportedly grappling with overheating issues. The Blackwell graphics processing units (GPU), overheat when connected together in server racks designed to hold up to 72 chips, as per the report.

Nvidia’s Blackwell AI chips overheating issue

Nvidia, the chipmaker has asked its suppliers to change the design of the racks several times to resolve overheating problems. This problem has been reported by Nvidia employees who have been working on the issue, as well as customers and suppliers with knowledge of the issue.

Nvidia’s Blackwell AI chips are experiencing overheating problems when connected to server racks designed to hold up to 72 chips,

Nvidia on the overheating servers

“Nvidia is working with leading cloud service providers as an integral part of our engineering team and process. The engineering iterations are normal and expected,” as per the company’s spokesperson.

Blackwell chips delays

The Blackwell chips, unveiled by Nvidia in March, were initially slated to ship in the second quarter.

However, Nvidia Blackwell chips faced delays, potentially affecting customers such as Meta Platforms Inc., Microsoft Corporation, Alphabet Inc.’s Google and Elon Musk’s xAI.

Why new Nvidia  AI chips Matter?

The demand for the Blackwell GPU platform was described as “insane” by Nvidia CEO Jensen Huang last month.

Nvidia’s Blackwell chip revenue was projected to hit $6 billion in the next quarter, according to a Morgan Stanley analyst. This projection was made despite Nvidia being fully supply-constrained on new products.

Nvidia’s Blackwell chip takes two squares of silicon the size of the company’s previous offering and binds them into a single component that is 30 times speedier at tasks like providing responses from chatbots.

While NVIDIA often adjusts server designs before a launch, the changes to the Blackwell racks are said to be coming late in the game. That said, Nvidia may still be able to deliver the racks to customers on time, and it hasn’t notified customers of any delay.

An NVIDIA spokesperson declined to comment on whether the Blackwell rack designs have been finalized.

All eyes are on NVIDIA ahead of earnings this Wednesday, November 20th.

This development highlights the complexity of scaling AI hardware to meet surging demand while managing engineering and logistical hurdles.

Nvidia the most valuable company

Earlier this month, Nvidia surpassed Apple to become the world’s most valuable company, largely due to its dominance in the artificial intelligence chip market. Despite these challenges, Blackwell chips promise groundbreaking performance. The AI giant’s current market capitalization stands at $3.482 trillion. Nvidia shares closed Friday’s trading session with a 3.26% decline, settling at $141.98. In after-hours trading, the stock saw a slight uptick, reaching $142.70. NVIDIA’s stock is up 187% year-to-date.

The post NVIDIA’s Blackwell AI Chips Face Overheating Issues Amid High Demand and Delays appeared first on Technowize.

(Image Credit: apple)

Apple says the new feature will allow you to quickly mask people and objects in a shot to color grade them or add effects separately from the rest of a scene.

Apple’s Final Cut Pro 11 AI update

In a suite of updates rolled out across its creative software lineup, the tech giant has sharpened its pitch to content creators, touting the efficiency of its M-series chips and the potential of its Apple Vision Pro for immersive media.

At the core of Final Cut Pro 11 are two standout tools: Magnetic Mask and Transcribe to Captions. Both powered by Apple’s Neural Engine. These new AI tools and features aim to reduce editing time by automating processes that traditionally required significant manual effort. Magnetic Mask uses machine learning to isolate people or objects in a frame, eliminating the need for green screens or complex rotoscoping. Transcribe to Captions, auto-generates captions on the timeline directly, giving faster workflows.

Vision Pro video editing

The Pro 11 release marks Apple’s move to position Final Cut Pro as a leader in spatial video editing, particularly for use with Apple Vision Pro. Users can now edit and preview spatial video, shot on iPhones or Canon’s new dual-lens system, directly on their Mac. In an upcoming update, users will also gain access to an ultra-wide virtual display that simulates a 32:9 panoramic view, offering a fully immersive editing environment designed for the Apple Vision Pro.

Final Cut Pro for iPad

Final Cut Pro updates extend to the iPad, where Apple has refined the touch-based editing experience. The iPad version now supports high-frame-rate recording, Live Drawing with new ink styles, and advanced haptic feedback with Apple Pencil, underscoring Apple’s push to make the iPad a legitimate mobile production tool. These updates also include the ability to apply effects, like automatic color and contrast adjustments, optimized for a range of formats from HDR to RAW.

For the iPhone, Apple has updated its Final Cut Camera app, enabling 4K recording at 120 fps on the iPhone 16 Pro, while also adding Log-encoded HEVC support. This feature reduces file sizes without compromising on quality—a practical update for professionals who need longer recording times without sacrificing fidelity. The app also introduces a LUT preview function, allowing filmmakers to view footage in real-time with applied color corrections, whether in HDR or SDR, bringing it closer to a full-scale professional setup.

Logic Pro AI features

Apple’s update also enhances Logic Pro, which has become a staple for music producers and video editors seeking to integrate professional sound into their projects. Logic Pro’s Quantec Room Simulator plug-in, introduced in this update, recreates the reverb profile of renowned Quantec acoustic spaces, adding a dimension of sonic authenticity that professionals prize. Other updates to Logic Pro streamline channel reordering and provide new search tools, allowing audio engineers to more efficiently manage complex project mixes.

These upgrades are small for current Logic users, but they make the digital audio workplace easier to use for no additional cost.

Apple’s pricing

Apple’s latest releases are available today, with Final Cut Pro 11 priced at $299.99. For new users on MacLogic Pro is priced at $199.99. The iPad versions are subscription-based, priced at $4.99 per month.

Upcoming updates by Apple

By next year, Apple’s suite of applications for creators will also include Pixelmator and Photomator which Apple recently acquired. These two apps are more aimed at editing photos than video or audio content.

The post Apple’s Final Cut Pro 11 Brings AI Magic and Spatial Video to Mac and iPad appeared first on Technowize.

(Image Credit: amd)

AMD has unveiled the Ryzen 7 9800X3D, a new CPU that is currently positioned as the main competitor to Intel’s Core i9-14900KS gaming chip. According to AMD’s Ryzen 9000 series performance comparison guide, the 9800X3D outperforms the previous competition in terms of gaming performance, with increased multi-threading and overclocking capabilities being the main focus. On average, the 9800X3D achieves around 8 percent more performance in 1080p games than its predecessor, the 7800X3D.

AMD Ryzen 7 9800X3D availability

The main issue with the 9800X3D currently is its availability, which is due to a heightened demand. One more reason for less availability is several retailers have either removed the chip entirely from their listings or shifted towards pre-orders for the chip only.

AMD Ryzen 7 9800X3D CPU price

The AMD Ryzen 7 9800X3D CPU is listed at $479 but due to unavailability, it has become the victim of scalping with prices on eBay hovering over $700.

AMD Ryzen competitor for Intel i9-14900KS

The AMD Ryzen 9 9950X is listed as the competition against the Core i9-14900K and the Core i9-13900KS. Although the AMD Ryzen 9 9950X’s performance in gaming is not as good as the 9800X3D, it is still a beast of a multi-threading part.

The same goes for the Ryzen 9 9900X, which is listed against the Core i9-14900K and Core i9-13900K. The Ryzen 7 9700X is listed as the competition of i7 14700K and i7-13700K.

AMD vs Intel

The AMD Ryzen 7 9800X3D is currently listed in the competitive guide as the competition against Intel’s top gaming chips, which are the Core i9-14900KS and the Core i9-14900K. Considering that the Core Ultra 9 285K comes nowhere close to its predecessors. The 9800X3D is around 8 percent faster on average at 1080p versus the 7800X3D, which was already the fastest chip on the market.

The chip offer better gaming performance than these two parts, but it does not offer the same level of multi-threading capabilities. This holds true also for the Ryzen 5 9600X which has been listed against the Core i5-14600K and the Core i5-13600K, which offer better multi-threaded performance.

AMD Ryzen Desktop CPUs market share

The AMD Ryzen Desktop CPUs have recently gained record market share. With recent price cuts, the Ryzen 9000 Non-X3D CPUs do look like very reasonable products, and with holiday season approaching fast, further price cuts can be seen on these chips which will make them even better value on the AM5 platform, which is now expected to support the latest Zen 6 “Ryzen” CPUs too.

AMD Ryzen demand

The current availability problems, especially with the Ryzen 7 9800X3D, show that demand is outstripping supply. Increased interest in powerful gaming CPUs, coupled with AMD’s new focus on multi-threading and efficiency, has obviously led to strong demand. Some market observers see this as an indication of the increasing attractiveness of AMD processors in the high-end segment.

AMD addresses different performance segments with the Ryzen 9000 series and is particularly competitive in the upper performance class for gaming CPUs. AMD offers a broader selection of CPUs for different application areas and budgets.

The post AMD Ryzen 7 9800X3D Outpaces Intel’s Core i9-14900KS in Gaming, Spurs Scalping appeared first on Technowize.

(Image Credit: openai)

OpenAI’s next flagship artificial intelligence model is showing smaller improvements compared to previous iterations, as per report. This seems like a sign that the booming generative AI industry may be approaching a plateau.

Why OpenAI’s smarter AI model is needed?

After the release of the viral ChatGPT chatbot two years ago, technology companies, whose valuations have benefited greatly from the AI boom, have publicly maintained that “scaling up” current models through adding more data and computing power will consistently lead to improved AI models.

But now, some of the most prominent AI scientists are speaking out on the limitations of this “bigger is better” philosophy.

AI model limitations

As per Ilya Sutskever, co-founder of AI labs Safe Superintelligence (SSI) and OpenAI, results from scaling up pre-training have plateaued.

“The 2010s were the age of scaling, now we’re back in the age of wonder and discovery once again. Everyone is looking for the next thing,” Sutskever said. “Scaling the right thing matters more now than ever.”

OpenAI’s rivals working on smarter AI models

Sutskever declined to share more details on how his team is addressing the issue, other than saying SSI is working on an alternative approach to scaling up pre-training.

Behind the scenes, researchers at major AI labs have been running into delays and disappointing outcomes in the race to release a large language model that outperforms OpenAI’s GPT-4 model, which is nearly two years old, according to three sources familiar with private matters.

AI advancements challenges

The so-called ‘training runs’ for large models are expensive, could cost tens of millions of dollars by simultaneously running hundreds of chips. They are more likely to have hardware-induced failure given how complicated the system is; researchers may not know the eventual performance of the models until the end of the run, which can take months.

Large language models gobble up huge amounts of data, with AI models exhausting all the easily accessible data in the world. Power shortages have also hindered the training runs, as the process requires vast amounts of energy.

Test-time compute

To overcome these challenges, researchers are exploring “test-time compute,” a technique that enhances existing AI models during the so-called “inference” phase, or when the model is being used.

This method allows models to dedicate more processing power to challenging tasks like math or coding problems or complex operations that demand human-like reasoning and decision-making.

OpenAI’s new AI model

OpenAI has embraced this technique in their newly released model known as “o1,” formerly known as Q* and Strawberry. The O1 model can “think” through problems in a multi-step manner, similar to human reasoning. It also involves using data and feedback curated from PhDs and industry expert s . The secret sauce of the o1 series is another set of training carried out on top of ‘base’ models like GPT-4, and the company says it plans to apply this technique with more and bigger base models.

At the same time, researchers at other top AI labs, from Anthropic, xAI, and Google DeepMind, have also been working to develop their own versions of the technique, according to five people familiar with the efforts.

Google and xAI did not respond to requests for comment and Anthropic had no immediate comment.

Implications of OpenAI’s smarter AI model

The implications could alter the competitive landscape for AI hardware, thus far dominated by insatiable demand for Nvidia’s AI chips.

“This shift will move us from a world of massive pre-training clusters toward inference clouds, which are distributed, cloud-based servers for inference,” said Sonya Huang, a partner at Sequoia Capita.

Unlike training chips, where Nvidia dominates, the chip giant could face more competition in the inference market.

Asked about the possible impact on demand for its products, Nvidia pointed to recent company presentations on the importance of the technique behind the o1 model. Its CEO Jensen Huang has talked about increasing demand for using its chips for inference.

The post AI Plateau? OpenAI Pivots to Smarter Training Techniques in Model Evolution appeared first on Technowize.