TechWealth Insights

Recent advances in artificial intelligence (AI) technology and its expanding applications have created a shortage of graphics processing units (GPUs). As AI takes over more of our lives, especially as generative AI becomes more mainstream, the computing power it requires becomes a critical factor, and GPUs are at the center of this.

While large GPU manufacturers like Nvidia are responding by announcing that they are ramping up production, the current demand for GPUs is far outpacing supply. This situation has forced cloud service providers like Amazon Web Services (AWS) and Microsoft Azure to limit the amount of computing power they offer, even to the point where Oracle and Microsoft are discussing an unusual deal that would allow them to lend each other servers when they run out of computing power for their cloud customers.

Why does computing power matter in generative AI? Because the performance of generative AI models is highly dependent on the input of computing power, especially GPUs. UBS analysts estimate that the initial version of the chatbot GPT required around 10,000 GPUs, with updated versions requiring three to five times that number.

As a solution to the GPU shortage, Nvidia announced at its recent Computex event in Taiwan that it is ramping up production of its H100 chips, but many AI founders expect the shortage to continue for at least the next year.

As a result, AI startups and venture capitalists (VCs) are scrambling to secure GPUs. Some VCs are ordering servers and processors in bulk to share among the AI startups they invest in. Some AI startups are buying their own servers with GPUs, while others are using minor cloud services until they can scale their AI models or the shortage is resolved.

The GPU shortage is spawning a new market, the GPU secondary market. This includes cryptocurrency miners. Companies that have accumulated large amounts of GPUs for mining but no longer need them due to the market downturn are finding themselves in the secondary market.

Chinese internet company Baidu is expected to announce its first smartphone product next week, according to Bloomberg. Baidu has an artificial intelligence dialog system through its DuerOS software and aims to embed artificial intelligence in all of its major business areas, including search, autonomous driving, and smart home appliances. The smartphone launch is expected to play an important strategic role in the company's entry into the mobile market.

Recently, a YouTuber named The Q created a bicycle tire with a square-shaped caterpillar. It's an impractical choice, to be sure, but it also has an unexpected advantage: it has a lot of ground clearance and doesn't require any upright parts.

But, while these innovations are unique and creative, bicycle tires are round for a variety of practical reasons.

First, round tires are more stable and maneuverable. The circular shape allows for consistent rotation, which provides a smoother ride and gives you more control over your bike. A square shape, on the other hand, is unstable and can cause the bike to wobble or move unpredictably.

Second, round tires are more efficient than square tires. The constant rotation of a round tire provides better energy transfer and requires less effort to move the bike. Square tires, on the other hand, require more energy and effort than round tires due to their unbalanced rotation and high friction with the ground.

Round tires conform better to the terrain. The narrow contact area of a bicycle tire requires a stable rotation to follow the irregularities of the terrain. The shape of a round tire ensures that it follows the contours of the terrain and provides a comfortable and stable ride.

Finally, round tires are easier and more cost-effective to make and maintain. Square tires require additional components and mechanisms to maintain stability and balance, which can be more difficult and expensive to manufacture and maintain.

Despite the practical limitations of square bicycle tires, it is important to respect the creativity and innovative ideas of individuals who seek to push the boundaries of conventional design. These innovative ideas spark new discussions and ideas, which ultimately contribute to the development of better bicycle design.

OpenAI, a San Francisco-based artificial intelligence (AI) startup, recently attracted new investors. According to TechCrunch, VC firms including Tiger Global, Sequoia Capital, Andreessen Horowitz, Thrive and K2 Global, as well as Founders Fund, have acquired new shares. These VC firms collectively invested more than $300 million, valuing the company at between $27 billion and $29 billion. Microsoft is also said to have invested around $10 billion.

OpenAI, the creator of ChatGPT, a technologically advanced AI model, has also gained popularity with many companies that are leveraging GPT and ChatGPT to advance their products and services. While this adds value to OpenAI, the technology and its rapidly evolving ecosystem also attracted investors. Microsoft announced GPT-4 integration by integrating OpenAI's technology into its Azure infrastructure to support the computational needs of its models, challenging Google's search service.

Last time I wrote about the shutdown of Paxful, I mentioned that alternatives were popping up all over the place. Now, Venmo joins the ranks.

Venmo is a money transfer app owned by Paypal, after launching its cryptocurrency trading service in April 2021, the company will begin offering cryptocurrency remittance services next month. Starting in May, Venmo users will be able to send cryptocurrency to their Venmo and PayPal accounts, as well as to external wallets and exchanges. However, users will have to pay blockchain network fees to send to external wallets and exchanges, and the service will not be available to Venmo users in New York.

Venmo already launched cryptocurrency trading in April 2021, and this is the first time Venmo users will be able to send and receive cryptocurrency. PayPal launched crypto trading and sending in October 2020 and June 2022, respectively.

We previously wrote about ChatGPT being blocked in Italy.

Italy has lifted its ban on OpenAI's ChatGPT, implementing measures to strengthen privacy and minor protection. In March, Italian privacy authorities imposed a ban on ChatGPT, citing a lack of legal basis for its age verification controls and user data collection. The ban on ChatGPT was recently lifted after OpenAI implemented the measures required by the Italian authorities.

OpenAI's measures include age verification, transparency, and user and non-user rights. The company has decided to disclose on its website how ChatGPT handles user data and develop a feature that allows users to remove their data from the training algorithm. In addition, users under the age of 13 will be blocked, and users under the age of 18 will need parental consent to use the service.

Recently, Paxful announced that it was shutting down its marketplace and that customers should reclaim their assets with detailed instructions. Ray Youssef, the company's founder and CEO, said that key staff turnover and regulatory issues led to the company's decision. The company's decision illustrates the problems that arise from regulatory pressure, and investors should consider self-protective asset management methods to avoid these problems.

Alternatives to Paxful's decision include Phantom and VICA.

Phantom recently added support for Ethereum and the Polygon blockchain. This means that Phantom can now reach a broader user base beyond Ethereum and Polygon blockchains. Users of the Phantom wallet will be able to send assets across a wide range of DApps, blurs, and NFT marketplaces. Phantom also emphasizes its differentiation from MetaMask by being able to manage multiple seed phrases, which will allow it to compete with MetaMask as the largest wallet in terms of being able to link seed phrases and manage everything in one place.

This 3D art created by a Korean artist has a unique feel and atmosphere. If you're a Korean who remembers this era, you can't help but admire the crazy quality.

Korean artist WoonJu's work has a very 1980s Korean feel to it. She says that when she's working, it's important to think about how the three-dimensionality comes alive and how the shadows come in. It makes a difference in the mood you convey, the sense of space, and the atmosphere.

It's your sense and persistence that determines the realism and mood of your work. He wanted to cast shadows on the ceiling, but spotlights only extend downward, so he installed additional pointlights to cast shadows. In this way, the artist puts persistence and effort into determining the dimensionality and atmosphere. In total, the artist created this project in one year and five months. Kudos to her persistence and craftsmanship.

To see more of her work, visit here.

ZDNet recently compared Microsoft and Google's AI chatbots, Bing and Bard, and found that they differed in the accuracy, variety, and context of their search results.

While Bing provides simpler search results with more links to additional information, Bard provides more detailed search results. Bard also offers useful features like creating schedules and writing code. However, Bing was found to be less contextual than Bard.

The two chatbots have different language models, with Bing using a US language model and Bard using Google Translate, which may have contributed to the differences in the search results they provide.

Neither chatbot is perfect - they can provide real-time information, but still suffer from hallucinations, and sometimes fail to understand the context of a request. Therefore, it's difficult to make an absolute comparison between the two chatbots.

Is Chat GPT a one-size-fits-all?

A survey found that 66% of IT engineers use ChatGPT to solve system issues. The most common IT questions in the U.S. are basic, such as "frozen computer screen," "slow internet," "locked computer," and "how to fix a bricked phone. 83% of Gen Zers and 67% of Gen Mers utilized ChatGPT. The younger people are familiar with the internet, the more they use it.

Unrecord is a gameplay video created using Unreal Engine 5, which uses some very innovative technologies to deliver fine-grained graphics and detailed environments. The environment uses Nanite technology to render high-resolution objects and Lumen technology to create light reflections and shadows. These technologies not only enhance the game's graphical quality, but also open up new possibilities for game design. While Unrecord hasn't been released as a game yet, the video has garnered a lot of technical interest. The game has graphics that could be mistaken for real-world environments captured by bodycam footage, and the game's style and features have yet to be revealed.

It's not uncommon for AI to steal the voices of real singers in music production. This has been an ongoing debate, and Canadian singer Grimes recently became the center of controversy for his quick response to the issue.

Grimes took to Instagram to announce that if her voice is used in an AI-generated song, she will take 50% of the profits from the song. This follows the recent controversy over a song called "Heart on My Sleeve," which was created using AI technology and released on global music platforms before being removed by the record label.

'Heart on My Sleeve' was created by learning the voices of Canadian rapper Drake and singer-songwriter The Weeknd with AI technology, and was well received by many. However, the song was later removed after a record label complained, sparking controversy over the use of AI in music.

In response, Grimes said, "If an AI-generated song is successful, I'll take 50 percent of the profits," adding, "Please feel free to use my voice without penalty." He went on to mention Canadian rapper Drake and singer-songwriter The Weeknd, emphasizing that he can honor his offer because he is "under no obligation to a record label."

The AI music theft controversy is not just about music, but also about issues such as infringing on the publicity rights of existing artists, which complicates matters. In this regard, experts are arguing that the legal issues of music production using AI technology and the need for policy responses to protect artists' rights need to be addressed.

Since the legal issues of AI music production and the protection of artists' rights have been weakly addressed in the past, the controversy has at least reemphasized the need for attention and response to these issues.

At the recent ASU+GSV Summit, Microsoft (MS) co-founder Bill Gates expressed his belief that artificial intelligence (AI) chatbots will be able to act as tutors. Gates believes that AI chatbots are developing at such a rapid pace that within a year and a half, they will be able to act as tutors to help children with their reading and writing skills.

In his speech, Gates emphasized that chatbots can teach students to improve their reading and writing skills, and that technological developments will continue to make this possible. Along with this, Gates said that AI chatbots will be highly capable of improving students' reading and writing skills, which will help them learn more efficiently.

These comments reflect the fact that AI technology is advancing rapidly, with AI chatbots being applied to new fields and improving their performance. In his speech, Gates emphasized the positive aspects of AI chatbots and said that the technology will continue to evolve and help students learn.

Gates also noted that AI chatbots are remarkably fluent in reading and writing, which raises the possibility that AI chatbots could play an active teaching role for students. These remarks paint a positive picture of the possibilities of AI chatbots and suggest that the technology could help students learn even more in the future.

Square Enix's recently released "Photopia Serial Murder Case" is a remake of the 1983 adventure game "Photopia Serial Murder Case". This remake was developed using Square Enix's AI technology. It retains the high level of difficulty and story twists of the original, and uses natural language processing for command input. The goal of the game is to allow players to progress through the story in their own language.

However, after its release, user Steam reviews were very negative (8% positive). This is due to the fact that the AI technology is not yet mature, as it doesn't recognize most of the strings, causing the assistant to repeat answers like "I'm sorry, I don't know". Gamers criticized the experience, calling it "like a game from the '80s."

Stanford University students have developed RizzGPT, a new pair of artificial intelligence (AI) glasses that combine the GPT-4 language model with augmented reality (AR) technology. The glasses can listen to conversations and tell you what to say.

RizzGPT combines AR glasses with a smartphone and uses GPT-4, a large-scale language model, and Whisper, an automatic speech recognition model. GPT-4 is used to recognize questions, comments, and images from conversations in real time and generate answers, while Whisper converts speech to text. AR glasses use Monocle AR, which consists of a camera, microphone, and high-resolution display, and displays user-readable text like a prompter.

LizGPT understands spoken questions from the microphone and images from the AR camera, and displays the answers on the glasses' screen. It also utilizes multi-modal recognition for text, audio, and images to understand what's happening around you.

The evolution of ChatGPT and many other emerging natural language processing models has been a game-changer in natural language processing, with impressive performance on a variety of tasks, including language modeling, machine translation, and question and answer. However, the cost of using these models can be significant, which can be daunting for individuals or organizations looking to take advantage of them.

One of the main factors that affects the cost of using natural language processing models is the type of cloud service you use. Cloud providers like AWS, GCP, and Azure offer a variety of pricing options, and the cost can vary depending on the specifications and usage of the model. According to one source, the cost of using the GPT-3 model on AWS can range from $4.6 to $18.1 per hour, depending on the type of instance used. Similarly, the cost of using the GPT-3 model on GCP can range from $8.7 to $34.8 per hour, depending on the type of instance. And for the recently launched GPT-4, you can expect to pay at least 15 times the cost of GPT-3 and as much as 60 times more.

To reduce these costs, it's important to determine the specific use case for your natural language processing model. For example, if you're using the model to generate text for a chatbot, a smaller model may be sufficient, and a more modest cloud service may suffice. On the other hand, if you're using your model for language modeling tasks or neural network training, you might need a larger model and a more powerful cloud service.

Another way to save money is to use smaller cloud services or self-hosted servers instead of larger cloud services, which can be more cost-effective, especially for smaller projects or organizations with limited resources. And applying the latest hardware and software technologies can help optimize the use of natural language processing models and reduce costs. Newer hardware and software technologies have more processing power and can do more work faster, which can result in significant cost savings over time.