Artificial Intelligence - Page 25

OpenAI has taken to its official website to share an update regarding GPT-4.5, the company's latest and most sophisticated AI model.

The company behind the immensely popular ChatGPT has confirmed it's releasing a research preview of GPT-4.5, which is OpenAI's "largest and best model for chat yet". GPT-4.5 introduces a new level of sophistication with improvements in unsupervised learning, pattern recognition, drawing connections, and being able to generate creative insights without reasoning. Moreover, OpenAI CEO Sam Altman has stated in an X post that GPT-4.5 "is the first model that feels like talking to a thoughtful person to me."

This sentiment was reiterated in the OpenAI post about the new model, with the company writing, "Early testing shows that interacting with GPT‑4.5 feels more natural. Its broader knowledge base, improved ability to follow user intent, and greater"EQ"make it useful for tasks like improving writing, programming, and solving practical problems. We also expect it to hallucinate less."

NVIDIA has confirmed its beefed-up Blackwell Ultra and next-gen Vera Rubin AI architectures are on track, lining up with recent reports that we'll get a huge info dump on the company's new AI GPUs at GTC 2025 in a few weeks time.

During the company's recent earnings call, NVIDIA CEO Jensen Huang confirmed with analysts that the recent GB200 AI server yield rates won't affect the company's annual release cadence, but the analysts also asked Jensen how NVIDIA would manage Blackwell Ultra and Rubin at similar release periods.

Jensen said: "Yes. Blackwell Ultra is second half. As you know, the first Blackwell was we had a hiccup that probably cost us a couple of months. We're fully recovered, of course. The team did an amazing job recovering and all of our supply chain partners and just so many people helped us recover at the speed of light".

Meta is planning on releasing a standalone AI app, which is part of company CEO Mark Zuckerberg's plan to put Meta at the forefront of the AI industry.

A new report citing people familiar with the matter states Meta will release the standalone AI app, which would be similar to OpenAI's ChatGPT, or DeepSeek, during the second quarter of this year. Notably, Meta released the Meta AI chatbot in September 2023, but then, in the following April, it was brought to the center of Meta's suite of apps as it was integrated into the search feature for Facebook, Instagram, WhatsApp, and Messenger.

However, users are only able to access Meta AI through these apps and the dedicated website, but that is about to change as people familiar with the confidential project say Meta is planning on releasing a standalone app that will compete directly with ChatGPT/Perplexity, and other standalone AI apps. According to the people familiar with the project, Meta believes its users could interact with the AI-powered digital assistant more deeply if it provided a standalone app.

In a recent X post, Perplexity announced their plans to release Comet, a new agentic browser that aims to revolutionize web browsing. Waitlists have been opened, and while specific dates haven't been provided, Comet is 'coming soon'.

The upcoming browser is engineered around the concept of agentic search, allowing for autonomous task execution, complex queries, and decision-making autonomously within the browser experience. Naturally, AI integration will be front and center for the new browser, with all the bells and whistles you'd find in modern AI tools.

SK hynix has reportedly reached a 70% yield rate on its next-gen HBM4 12-Hi memory, ready to rock and roll on NVIDIA's next-gen Rubin R100 AI GPUs.

In a new report from ETNews, we're learning that SK hynix has reached 70% yield on its HBM4 memory, with the yield good enough for the semiconductor industry to get excited for the South Korean memory giant's new AI memory. An industry insider told the outlet: "The HBM4 test yield exceeded 60% at the end of last year and I understand that it has recently increased it further. The test yield is increasing at a fairly fast rate".

The test yield is an indicator that can be used to make estimations on the actual future yield rate, where if the test yield is high, productivity can be secured when actual mass production begins. The industry insider added: "In terms of being able to increase the yield further after mass production, a test yield of 70% is a very good result".

NVIDIA's next-gen B300 AI GPU is expected to launch in Q4 2025 with mass production expected in Q4 2025 according to the latest reports.

In a new article published by DigiTimes, we're to expect NVIDIA's unveiling of its new Blackwell B300 AI GPU at GTC 2025 (its in-house GPU Technology Conference) in March. NVIDIA will reportedly mention the first CPO technology-based InfiniBand switch that will debut in Q3 2025, supporting 36 x 3.2T CPO optical modules and 4 x 28.8T application-specific integrated circuits (ASICs) that will reach a blistering 115.2T signal transmission.

NVIDIA's new B300 AI GPU will reportedly chow down on up to 1400W of power, with 1.5x the performance in FP4 performance on a single AI GPU over the B200 AI GPU. We can expect the HBM capacity of each GPU to increase from the max of 192GB HBM3E on the B200, to a much bigger 288GB of HBM3E on B300.

In a recent upload from Falling Knife Films, users were greeted with an insight into what it would be like to step back in time to 1912, experiencing the luxurious Titanic's final moments as it sank. Created with the assistance of AI, the footage depicts the disaster from a 'found footage' perspective, mimicking the look and feel of a modern phone camera. The video captures a first-person journey through the ship's opulent interiors, transitioning from calm, everyday moments to the unfolding chaos as passengers scramble for safety.

The uncanny realism of the footage quickly spread across X and Reddit, with many users commenting on its eerie, unsettling quality. However, it also highlights how close we are to integrating AI into modern filmmaking, blurring the line between reality and simulation.

AI-generated 'POV' footage has become an emerging trend on social media. A similar viral clip depicting 'life as a teenager in 1990s America' gained traction on X, showcasing everyday moments like playing a PlayStation 1, bowling with friends, and flipping through a record store-all rendered as if captured on a modern smartphone.

Chegg has officially filed a lawsuit against Google, accusing the tech giant of unfairly diverting traffic from educational publishers with its AI-powered search results. In a recent Q4 earnings call, the ed-tech company claimed that Google's AI Overviews, which provide instant, AI-generated answers at the top of search results, are negatively impacting student learning and the online education ecosystem.

The lawsuit, filed in the U.S. District Court for the District of Columbia, argues that Google's AI Overviews unfairly retain traffic that would typically go to Chegg and other content providers. Chegg claims this practice is harming students by reducing access to step-by-step educational resources, replacing them with low-quality, unverified AI summaries.

Apple's huge new Houston, USA factory will start making AI servers powered by its next-gen in-house M5 processor after news hit that the company would be spending over $500 billion in the United States over the next 4 years.

In a new post on X by industry insider Ming-Chi Kuo, we're learning that Apple's new Houston factory will begin making AI servers for Apple Intelligence Private Cloud Compute (PCC) in 2026. In previous Apple M5 predictions, Kuo said: "Apple's PCC infrastructure build-out will accelerate after the mass production of the high-end M5 chips".

Apple will reportedly kick off mass production of its high-end M5 chip in 2H 2026, which aligns perfectly with Apple's timeline for expanding its AI server capacity.

Elon Musk has announced that his xAI startup has just launched Grok 3, the latest AI chatbot that integrates tightly into X and it even has a dedicated app available on both iOS and Android operating systems.

Grok 3 is being rolled out immediately to Premium+ subscribers on X, while we have a new subsciption for xAI is being rolled out called SuperGrok, for users accessing Grok 3 through the mobile app or on the Grok.com website. Grok 3 can generate text and images without the guardrails against sexually aggressive imagery, vulgarity, or the reproduction of well-known people's likenesses.

During a livestream sitting with 3 xAI engineers earlier this week, Musk said: "Grok 3 across the board is in a league of its own" . Musk added that Grok 3 far outpaces Grok 2 with "more than 10 times" the computing power, killing it in AI benchmarks across the planet. Musk calls Grok 3 "maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct".

Elon Musk's Grok-3 has officially been released, and according to early feedback, it's actually pretty good. Grok-3 has already placed at #1 in the Chatbot Arena leaderboards - a platform that ranks AI language models based on blind popularity contests. The startup also claims that it outperforms its competitors across math, science, and coding benchmarks.

Yet, beyond standardized benchmarks for LLMs, the practical application of these tools is relatively similar. Converse with a chatbot, generate images, write code, and search the web. Although they each have different strengths, there isn't a massive difference between the UX of prompting DeepSeek, Gemini, ChatGPT or Claude.

However, one of Grok-3's distinguishing characteristics is that it's a little more... unhinged. Aside from being trained on (shudders) Twitter data, the model is intentionally designed to include fewer safeguards. You can generate text prompts containing profanity, akin to an unfiltered Charles Bukowski, and even images of celebrities, which understandably would concern creators.

Samsung Vice Chairman Jeon Young-hyun has reportedly met with NVIDIA CEO Jensen Huang on US soil to discuss supplying the company with its new 8-Hi HBM3E memory for use on its AI GPUs.

In a new report from SEDaily, we're learning from semiconductor industry sources that the vice chairman recently visited NVIDIA's headquarters in Sunnyvale, California with CEO Jensen Huang to discuss various agendas surrounding the cooperation between the two companies.

Jeon Young-hyun is the Vice President and Head of Device Solutions (DS) Division, overseeing the semiconductor division of Samsung, met with Jensen in person, which leads the industry to believe Samsung's new 8-Hi HBM3E memory is finally, finally ready for certification (after months and months of delays and issues).

Micron is preparing for the mass production of its new 12-Hi HBM3E memory, ready to supply its new AI memory to NVIDIA to use on its AI GPUs, while also teasing next-generation HBM4 is coming in 2026.

The US-based memory giant announced in September 2024 that it was developing its 12-Hi HBM3E and provided samples to customers including NVIDIA. Business Korea is reporting from industry sources that on February 14, Micron's Chief Financial Officer, Mark Murphy, highlighted 12-Hi HBM3E memory advantages during an event hosted by Wolfe Research.

Murphy said that the new 12-Hi HBM3E memory consumes 20% less power and packs 50% more capacity than the 8-Hi HBM3E memory standard. The Micron CFO predicted that most of the HBM produced in 2H 2025 would be 12-Hi HBM3E memory.

Last year, we reported on TCL's Next Stop Paris, an AI-generated romantic comedy that would kick off a string of AI content for its TCLtv+ streaming platform, available on all of the company's Smart TVs. Although it took several months to complete, the "long-anticipated AI-powered love story" premiered last week, and you can watch the entire 11-minute and 44-second film right now.

"Claire, an overly cautious planner, and Del, a free-spirited dreamer, meet by chance on a train and embark on a spontaneous day in Paris filled with adventure and self-discovery," the synopsis reads. "This whirlwind romance forces them to confront whether their unexpected connection is the work of destiny or a sign of something greater at play."

Yeah, it's not great. The quality of the writing, storytelling, and imagery looks like something from the CD-ROM era of PC gaming when B-movie acting and obvious lo-fi CG sets and backdrops served as the platform for a new style of "interactive movies." Night Trap on the Sega CD add-on for the Genesis or The 7th Guest on PC come to mind.

Apple has been working on an AI-powered update for its Siri digital assistant, but according to the latest report, it is "facing engineering problems and software bugs, threatening to postpone or limit its release".

The update on the AI-powered update for Siri is coming from Mark Gurman on Bloomberg, where he reports the news "according to people with knowledge of the matter". After announcing its plans for new AI-powered Siri abilities at its developers conference in June 2024, advertising its new features around the world, the company is still not ready.

Some of the new features were planned for an April 2025 release, but Gurrman's sources said that these features "may have to be postponed until May or later". Apple refused to comment on the Siri project, after promising some major updates coming to the digital assistant in June 2024:

Chinese AI company DeepSeek is reportedly working on a new "semiconductor" venture, a new in-house AI chip that would compete directly with the likes of NVIDIA, AMD, and other companies.

In a new report from DigiTimes, we're learning that DeepSeek is looking into making its own AI chips, adding that the company has started a "major recruitment drive" hiring semiconductor experts to lead the AI chip project. Chinese companies have multiple hurdles in their way, the biggest being securing the semiconductor equipment and processes because of global sanctions.

The only chip access the likes of DeepSeek has access to would be SMIC, but they're many years behind the rest of the semiconductor industry, namely TSMC.

Meta recently announced two new breakthroughs from their global research labs. The first highlights that Meta successfully developed a model that can successfully convert brain signals into text. The second reveals how the brain transforms thoughts into words, offering new insights into language processing and AI development.

The first paper refers to an AI model that can decode up to 80% of typed sentences solely from brain activity. The model utilizes noninvasive MEG and EEG recordings, essentially sparing the need to conduct surgical procedures. It presents a few potential benefits, for example, the ability to establish brain-computer interfaces for those who have lost speech and clinical use for brain injury patients.

While the advancement is impressive, researchers are still encountering challenges with accuracy. While the AI can decode up to 80% of characters, errors still occur, making full, reliable communication difficult. It also requires stillness and a magnetically shielded room, which poses issues for practicality. So don't worry - you won't need to build a Magneto-style thought shielding helmet any time soon. It's also unclear how well this would work for patients with brain injuries and disorders.

Tiktok's parent company, ByteDance, has released another AI text-to-video model. The project, entitled Goku, allows users to convert text into hyper-realistic ad videos, many of which draw clear inspiration from the real-life reels we see on social media.

The project's GitHub page showcases examples of the tool in action, ranging from landscape demos to portrait videos to marketing footage. Some of the most jarring examples are the ones that look the simplest: people eating food, unboxing products, or doing their makeup. While you can still notice that uncanny AI feeling to many of them, the gap is beginning to close on being able to differentiate AI content from real ones.

The Goku model is a 'flow-based video generative foundation model' jointly developed by the University of Hong Kong and ByteDance. Released in February 2025, industry insiders believe that the newest model 'significantly improved' the visuals' consistency, controllability, and richness. The model is based on the 'rectified flow transformer architecture' and can generate images and videos with simple text prompts.

Samsung is reportedly working on redesigning its 6th-generation 1c DRAM to improve yield rates, in order to catch up

with its next-generation HBM4 process.

Samsung considering tweaking its 1c DRAM process is a key factor in the success of its next-gen HBM4, with a new piece from ZDNet Korea reporting that the South Korean memory giant is looking at designs for its cutting-edge DRAM processes, something it has been doing since 2H 2024, and that the company has redesigned its high-end 1c DRAM to sure its next-gen HBM processes are adopted into the AI industry, unlike its HBM3 memory being denied by NVIDIA.

US Vice President JD Vance gave a fantastic speech at the AI Action Summit in Paris on Tuesday, where he said the Trump administration "will sure that the most powerful AI systems are built in the US with American design and manufactured chips".

President Trump has been steadfast in wanting the USA to benefit from all of its silicon prowess, and now his running mate VP Vance has said the same thing.

VP Vance said: "The Trump administration believes AI will have countless revolutionary applications in economic innovation, job creation, national security, healthcare, free expression and beyond. To restrict its development now will not only unfairly benefit incumbents in the space, it would mean paralyzing one of the most promising technologies we have seen in generations".