Facebook’s parent company, Meta, is making a strategic shift in artificial intelligence (AI) development by testing its first in-house chip designed for training AI systems. This move marks a significant step toward reducing reliance on external suppliers like Nvidia and cutting down on infrastructure costs.

A Step Towards Self-Sufficiency

The world’s largest social media giant has commenced a small-scale deployment of the AI training chip, with plans to expand production if initial tests yield positive results. This development is part of Meta’s broader strategy to control its AI-driven future by designing custom silicon, rather than depending solely on industry leaders like Nvidia.

Meta’s commitment to AI is evident in its financial projections. The company anticipates total expenses of $114 billion to $119 billion in 2025, with up to $65 billion allocated for capital expenditures primarily driven by AI infrastructure investments.

The Meta Training and Inference Accelerator (MTIA) Program

The new chip is part of Meta’s MTIA series, which has faced setbacks in the past. One of the sources noted that the training chip is a dedicated accelerator, meaning it is optimized specifically for AI workloads, potentially making it more power-efficient than traditional integrated GPUs used for such tasks.

Taiwan Semiconductor Manufacturing Company (TSMC) (2330.TW) is producing the chip, which recently completed its first “tape-out” a critical milestone in silicon development. Tape-outs involve sending an initial chip design to a factory for manufacturing, a process that costs millions and takes several months. However, the risk of failure remains high, as an unsuccessful test would require Meta to diagnose issues and redo the tape-out process.

Learning from Past Failures

Meta’s journey in AI chip development has not been without obstacles. The company previously scrapped an in-house custom inference chip after an unsuccessful small-scale deployment, forcing it to reverse course and invest billions in Nvidia GPUs. This decision made Meta one of Nvidia’s largest customers, utilizing GPUs for tasks like ad systems, recommendation engines, and training large-scale AI models, including its Llama foundation model series.

However, Meta’s executives are determined to make their own chips viable. Last year, the company successfully deployed an MTIA chip for inference tasks, such as curating content for Facebook and Instagram news feeds. This success has encouraged Meta to push forward with training-specific chips, with plans to integrate them into AI-powered recommendation systems and, eventually, generative AI applications like the Meta AI chatbot.

A New Era for AI Training?

Meta Chief Product Officer Chris Cox likened the company’s AI chip development to a “walk, crawl, run” approach, emphasizing gradual advancements. He highlighted the first-generation inference chip as a major success and underscored the long-term vision of scaling up Meta’s AI capabilities through proprietary hardware.

Despite this push, doubts persist about the long-term sustainability of scaling large language models through sheer computational power. Recent breakthroughs by Chinese startup DeepSeek have demonstrated that optimizing computational efficiency rather than simply increasing GPU power could be a more effective approach. This realization triggered a temporary decline in AI stock values, including a sharp drop in Nvidia’s share price.

Nonetheless, investors remain confident in Nvidia’s dominance in the AI chip industry, even as companies like Meta explore alternative solutions. Whether Meta’s AI chip strategy will eventually disrupt the reliance on third-party GPUs remains to be seen, but one thing is clear: the battle for AI hardware supremacy is just beginning.

Meta’s shift toward custom AI chips marks a significant inflection point in its technological evolution. If successful, this endeavor could lower costs, improve efficiency, and reduce dependency on Nvidia and other external suppliers. However, the risks remain high, and the company’s ability to execute this vision will determine whether it can establish itself as a leader in AI hardware.

Also read: Microsoft is Reportedly Working on an Xbox Handheld Console

Microsoft is making a bold move in gaming by reportedly working on a handheld Xbox console, which could launch later this year. This comes alongside plans to introduce new versions of the Xbox Series S and X consoles by 2027. If these reports are accurate, it signals a shift in Microsoft’s gaming strategy towards more flexibility and portability.

Xbox Handheld: What We Know So Far

According to Windows Central, Microsoft is collaborating with a PC gaming OEM potentially brands like ASUS, Lenovo, MSI, or Razer to develop the device. Unlike a traditional console, this new handheld will run Windows and focus on the Microsoft Store and PC Game Pass as core features. This suggests it will function more like a gaming PC in a portable form factor, rather than a fully custom-built Xbox console.

Reports indicate that the handheld, codenamed “Keenan,” may not introduce groundbreaking hardware but will refine Windows for handheld gaming. The device could be similar to existing PC handhelds, like the Asus ROG Ally X or Lenovo Legion Go, and could feature an AMD Strix Point APU, ensuring high performance for modern games.

Microsoft’s Strategy – Expanding the Xbox Ecosystem

Rather than being a major revenue generator, this new handheld is believed to be a research project for Microsoft, helping the company explore ways to expand Xbox beyond traditional consoles. It aligns with Microsoft’s vision of making Xbox an ecosystem rather than a single hardware platform.

While the hardware itself may resemble existing PC handhelds, Microsoft’s real innovation may come from optimizing Windows for portable gaming. Improvements could include better power management, fan control, console-style input integration, and a more user-friendly touch interface. If Microsoft succeeds, this could set a new standard for Windows-based handheld gaming.

How Does This Compare to Valve’s Steam Deck?

Microsoft’s move mirrors Valve’s SteamOS strategy, where the company optimized its OS for the Steam Deck and later expanded it to third-party devices. A dedicated handheld Xbox, optimized for Game Pass, could be a direct response to the Steam Deck’s success in dominating the portable gaming market. The challenge for Microsoft will be creating a seamless handheld gaming experience that truly competes with Valve’s optimized SteamOS.

What’s Next for Xbox Consoles?

Alongside the handheld, Windows Central reports that Microsoft has also finalized plans for the next generation of Xbox consoles, set to replace the Xbox Series S and X in 2027. These consoles will likely continue the trend of becoming more PC-like, making it easier for developers to port games between PC and Xbox.

While no official specs have been revealed, it’s speculated that the next Xbox will use AMD’s future GPU architecture, likely beyond RDNA 4. Microsoft also plans to ensure backward compatibility, meaning older Xbox games will still be playable on newer hardware.

Final Thoughts – Will This Xbox Handheld Succeed?

The handheld gaming market has remained niche, despite the rise of devices like the Steam Deck, ROG Ally, and Lenovo Legion Go. If Microsoft can successfully optimize Windows for handheld gaming and deeply integrate Game Pass, it could change the landscape of portable gaming.

However, competition is fierce. The Steam Deck remains the dominant player, and success will depend on how well Microsoft balances hardware, software, and pricing. If done right, an Xbox handheld could redefine portable gaming and push Microsoft further into the world of PC-style gaming devices.

OpenAI researchers have revealed that even their most cutting-edge AI models still fall short of matching human coders, despite CEO Sam Altman’s assertion that these systems could outdo “low-level” software engineers by the end of the year. In a newly released paper, the team notes that frontier AI models those at the very forefront of the field remain incapable of handling the bulk of coding tasks.

To reach this conclusion, the researchers introduced a fresh benchmark called SWE-Lancer. This tool draws on more than 1,400 real-world software engineering tasks from the freelancing site Upwork. Using SWE-Lancer, OpenAI evaluated three large language models: its own o1 reasoning model, the flagship GPT-4o, and Anthropic’s Claude 3.5 Sonnet.

Researchers used a newly created benchmark to see how large language models would handle two categories of Upwork tasks: fixing bugs directly or managing projects at a higher level. To prevent any reliance on existing online code solutions, the models were barred from internet access. Although they collectively tackled jobs worth hundreds of thousands of dollars, the models mostly addressed small, surface-level errors. They consistently fell short when it came to spotting more serious bugs or diagnosing their deeper causes an outcome often noted by those familiar with AI’s knack for confidently supplying answers that can unravel upon close inspection.

Despite working at speeds far exceeding human programmers, all three models struggled to recognize how pervasive the errors were or how they impacted the broader project. This led to incomplete or inaccurate outcomes. Among them, Claude 3.5 Sonnet was the top earner on these simulated Upwork tasks, outperforming the two OpenAI models. However, the majority of its solutions were still flawed, and the paper concluded that any AI system would need much greater consistency and accuracy to be suitable for real coding work. In short, while frontier AI models can solve focused issues and do it quickly, they remain nowhere near as adept as human engineers when it comes to tackling complex software challenges.

Although these large language models have made huge strides in recent years and will likely keep improving, they still lack the expertise to fully replace human software developers. Despite this shortfall, it hasn’t prevented some CEOs from taking the risky step of dismissing their programmers in favor of AI systems that aren’t quite ready for prime time.

Also read: China’s EV Giant to Mass Produce Solid-State Batteries for Extended Range

BYD, one of China’s leading electric vehicle (EV) manufacturers, has announced a major milestone in battery technology. The company is set to mass-produce all-solid-state batteries (SSBs), marking a shift towards safer and more efficient energy storage solutions. These batteries replace the liquid electrolytes found in traditional lithium-ion batteries with solid electrolytes, enhancing safety and performance.

What Makes Solid-State Batteries Special?

Solid-state batteries bring several advantages over conventional lithium-ion batteries. By using sulfide-based solid electrolytes, BYD aims to offer:

• Longer battery life: Reduced degradation over time.
• Faster charging: More efficient energy transfer.
• Improved safety: Less risk of overheating or fire.
• Higher energy density: Increased driving range for EVs.

These benefits make solid-state batteries an ideal choice for high-end electric vehicles, where performance and safety are top priorities.

BYD’s Plan for Solid-State EVs

BYD’s Chief Technology Officer, Sun Huajun, confirmed that the company expects its first solid-state EVs to be available within the next two years. The initial phase will focus on limited production, allowing BYD to demonstrate the capabilities of its new battery technology. Over the next five years, the company plans to scale up production and integrate these batteries into more EV models.

The Global Competition in Solid-State Batteries

Currently, no automaker has mass-produced all-solid-state batteries for EVs, making BYD one of the pioneers in the field. The company is the world’s second-largest EV battery manufacturer, holding a 17.2% market share in 2023. The only company ahead of BYD is Contemporary Amperex Technology Co. (CATL), which dominates with a 37.9% global market share. CATL has also announced plans to produce solid-state batteries, though mass production is expected around 2027.

Challenges in Mass Production

While solid-state batteries are seen as the future of EVs, there are challenges to overcome before they can be widely used. These include:

• Manufacturing complexity: New methods are needed to produce these batteries at scale.
• High initial costs: Costs must be reduced to match lithium-ion batteries.
• Electrode design: The structure and materials used in electrodes impact performance and cost.

BYD is working on these challenges to ensure that solid-state batteries can be produced efficiently for future EVs.

The Future of EV Battery Technology

Solid-state batteries are considered the “holy grail” of EV battery innovation. Their potential to provide longer range, faster charging, and increased safety makes them a game-changer for the industry. As more automakers invest in this technology, we may see a major shift towards solid-state-powered EVs in the coming years.

With BYD leading the charge, the race to bring solid-state batteries to market is heating up. The next few years will be crucial in determining how quickly this technology can become mainstream and revolutionize electric mobility.

The development of solid-state batteries is a major step forward for the EV industry. With BYD and CATL competing to bring this technology to market, we can expect significant advancements in EV range, charging speed, and safety. If mass production is successful, solid-state batteries could soon replace traditional lithium-ion batteries, making electric vehicles more efficient and accessible worldwide.

Also read: Humane’s AI Pin to Shut Down as HP Buys Company for $116 Million

Humane, the startup that made headlines with its AI Pin, has announced on Tuesday that HP is acquiring most of its assets for $116 million. As part of this deal, Humane is immediately discontinuing the AI Pin, a device that was marketed as a smartphone alternative. If you own an AI Pin, here’s what you need to know about the shutdown and what HP plans to do with Humane’s technology.

AI Pin Will Stop Working Soon

If you have purchased an AI Pin, it will continue to function until February 28, 2025, at 12 p.m. PST (3 p.m. ET). After this time, AI Pins will no longer connect to Humane’s servers, which means they will lose key features like calling, messaging, AI interactions, and cloud access. The only offline functions, such as checking the battery level, will still work, but anything that requires internet connectivity will stop.

What AI Pin Users Need to Do

Humane is urging all AI Pin users to back up their photos, videos, and other important data before the shutdown date. Once the servers go offline, any stored data will be permanently deleted. The company will also disband its customer support team on February 28, so users will no longer have official assistance after that date.

Refund Policy

Humane has confirmed that only customers who bought an AI Pin in the last 90 days before the shutdown announcement are eligible for a refund. Refund requests must be submitted by February 27, 2025. If you have an active Humane subscription that extends beyond February 28, the company will issue a prorated refund. Customers who were waiting for replacement charging cases due to a previous recall will automatically receive a refund for the cost of that accessory after February 28.

What Went Wrong for Humane?

Humane had ambitious plans for the AI Pin, positioning it as a futuristic alternative to smartphones. However, after launching in April 2024, the device faced harsh criticism from reviewers and early users. Some of the biggest complaints were poor performance, limited functionality, and an unreliable AI assistant. High-profile reviewers, including MKBHD, even called it one of the worst products they had ever tested. In addition, safety concerns led to a recall of the AI Pin’s charging case due to battery fire risks. By the summer of 2024, returns had started outpacing sales, and the company had to drop the AI Pin’s price by $200 in October to try and boost interest.

HP’s Plans for Humane’s Technology

HP is not acquiring Humane just for the AI Pin. Instead, it is bringing on Humane’s engineers, product managers, and intellectual property to form a new division called HP IQ. This group will focus on integrating artificial intelligence into HP’s future products, including personal computers, printers, and smart office solutions.

Additionally, HP has acquired Humane’s CosmOS AI operating system, which was originally developed for the AI Pin. Humane had been exploring ways to adapt CosmOS for other devices, such as cars, smart speakers, TVs, and Android phones. HP might use this technology to expand AI capabilities across its product ecosystem.

The End of Humane’s AI Pin Journey

Humane’s journey has come to an abrupt end, but HP’s acquisition suggests that its technology may still play a role in future AI-driven devices. For AI Pin owners, however, this is the final chapter. If you have an AI Pin, make sure to back up your data and check if you’re eligible for a refund before February 27, 2025.

Would you have wanted to see the AI Pin succeed? Let us know your thoughts on this bold but short-lived attempt to reshape personal technology.

Also read: Grok 3 Is Almost Here! Elon Musk Says It’s ‘Scary Smart’