This Week In AI: Latin America, Russia & US Build Regional AI Models As Apple, Amazon & Midjourney Innovate

In every corner of the world, a technology has been evolving in the blink of an eye—no points for guessing that it is artificial intelligence (AI).

The growing impact of AI across the creative sector and diverse industries has led to the rise of a regional AI race, specifically in the United States (U.S.) and Latin America (LatAm). With U.S. AI startups leading the race, businesses in LatAm and Russia have raised the stakes with AI announcements of their own.

Let’s see what’s unfolded this week.
 

Latin America Unites To Launch Latam-GPT

Latin America has come up with a novel initiative that could transform the AI outlook for the region.

A dozen countries across the region will unite in September for the launch of Latam-GPT, the first large language model specifically designed to understand the diverse cultures and linguistic nuances unique to Latin America.

With development for Latam-GPT commencing in January 2023, officials have maintained that this model is a core technology, solely created for building different applications, such as chatbots, rather than directly competing with existing consumer products like ChatGPT.

Chile's state-run National Center for Artificial Intelligence (CENIA) will be heading this open-source project, involving over 30 regional institutions.


The model is built upon Llama 3 AI technology and is being trained using a distributed network of computers across the region, which includes facilities at Chile's University of Tarapaca and cloud-based systems.

Global AI models, precisely trained on English-language data, faced struggles with Latin American languages and cultural contexts. Latam-GPT intends to overcome the inaccuracies and limitations of such hurdles.

The objective behind the project is to considerably increase the adaptation and accessibility of AI throughout the region. Other objectives include the crucial goal of preserving Indigenous languages. In fact, an initial translator has already been developed for Rapa Nui, the native language of Easter Island.

Plans are on the go to amplify this capability to other Indigenous languages, opening the doors for supplementary applications, such as virtual public service assistants and personalized educational systems adapted to these communities.

Chilean Science Minister Aisen Etcheverry expressed hopefulness about the project's potential, stating, "This could be a democratizing element for AI." Etcheverry contemplates the model’s application in critical sectors like schools and hospitals, which provide solutions that truly resonate with local culture and language.

The regional development bank CAF and Amazon Web Services (AWS) have provided major support for this ambitious effort.

CENIA head Alvaro Soto has high expectations, though the project presently operates without a resolute budget. He believes that Latam-GPT's capabilities will be potent and influential in attracting the necessary funding to advance its development and expand its reach to promote a more inclusive and culturally relevant AI future for Latin America.

While Latin America prioritizes inclusive AI, Russia is all set for a strategic move.
 

Sberbank Introduces A New Reasoning AI Model

Russia's largest financial institution, Sberbank, has taken a gigantic step in the AI race.

Sberbank has planned the launch of a new version of its proprietary GigaChat large language model (LLM) trained with modern reasoning capabilities. The purpose is to go beyond basic conversational tasks and solve more challenging problems in areas such as science, coding, and mathematics.


"In the near future, we will release a new model. Our GigaChat will have reasoning functions, meaning it will be capable of scientific research,” First Deputy CEO Alexander Vedyakhin revealed in a recent interview. “I am already testing the beta version,” he added.

This is a huge opportunity for Sberbank to catch a second wind alongside global AI leaders like the United States and China, who have already seen such reasoning LLMs in action, with OpenAI making headlines by introducing its LLM version last September.

While Vedyakhin admitted Russia presently straggles by an approximated six to nine months, he pointed out GigaChat's constitutional perks for Russian clients. Its powerful language adaptation, steady performance, and the use of domestic cloud servers make it a preferred option for roughly 15,000 Russian companies already using the platform.

Interestingly, the timing aligns with Yandex’s May announcement of reasoning functionality for its search engine, which is Sberbank's major competitor in the AI space.

Apart from Russia building AI for local needs, Apple is using AI to rethink its chip design.
 

Apple Explores AI For Chip Design, Signals Future Innovation

After the most-awaited WWDC 25—yes, how can we miss the “liquid glass” update—Apple is apparently exploring the use of generative AI to expedite the design of its proprietary chips.

According to the company's top hardware technology executive, this step holds immense potential to increase productivity in the complex world of semiconductor evolution.

Johny Srouji, Apple’s Senior Vice President of Hardware Technologies, recently hinted at this strategic direction during a speech in Belgium where he was honored by Imec, a distinguished semiconductor research and development.


In a recorded speech, Srouji outlined Apple’s journey in creating custom silicon, starting from the original A4 chip in the 2010 iPhone to the powerful processors found in today’s Mac desktops and the Vision Pro.

According to Srouji, the necessity of utilizing the most advanced chip design software available was the key lesson. This includes tools from electronic design automation (EDA) firms, where industry leaders like Cadence Design Systems and Synopsys are already integrating AI into their services.

"EDA companies are super critical in supporting our chip design complexities," Srouji mentioned. He added with a clear belief, "Generative AI techniques have a high potential in getting more design work in less time, and it can be a huge productivity boost."

Srouji further added about the mainstay of Apple's chip design philosophy of making bold, irrevocable determination. He emphasized Apple’s transformation from Mac computers from Intel’s chips to their own custom silicon in 2020, which was an unplanned switch.

With Apple making strides in AI, Midjourney, another US-based AI startup, has entered the conversation with the launch of V1.
 

Midjourney Steps Into AI Video Arena With V1 Model

After OpenAI’s Sora, Midjourney, a company that has created a unique identity in the generative AI space with its diverse image creation tools, has introduced its most awaited AI video generation model, V1.

With the V1 image-to-video model, you can turn existing images into dynamic five-second video clips using self-generated images or ones using Midjourney's own models.


By simply selecting an image and pressing the "Animate" button, the model creates a set of four variations. Though the initial videos are short, you still get an option to extend them by four seconds, up to a total of four times.

You can choose between automatic or manual settings, with the automatic setting generating motion prompts and the manual giving more control over the desired actions.

On top of that, "low motion" and "high motion" settings are also accessible to direct the camera and subject movement within the generated videos.

The initial trials turn out to be the outputs that maintain Midjourney's signature "otherworldly" and "dreamlike" aesthetic, unlike the models aiming for hyperrealism.

While speaking about the company's grander vision for its AI models, Midjourney CEO David Holz said in a blog post, “We believe the inevitable destination of this technology are models capable of real-time open-world simulations.”

This goes simply beyond video to include plans for 3D renderings and real-time AI models, which will build blocks towards a fully interactive generative ecosystem.

However, this model might be heavy on the wallet, costing eight times more than a typical image generation model. This means users will end up using the monthly allotted tokens more quickly when creating videos. The cheapest is the $10-per-month Basic plan, while the Pro plan comes at $60-a-month and the Mega plan at $120-a-month. 

The good news is, both the Pro and Mega plans will have unlimited video generations, provided in the slower "Relax" mode, and Midjourney will reconsider the pricing of its video generation over the next month.

With the launch of V1, Midjourney now enters the AI video generation race, with major players like Adobe's Firefly, Google's Veo, and Runway's Gen-4.

Well, let’s take a pause here.

Creativity often invites accusations and controversy, especially when it’s driven by AI tools. Despite the launch of V1, Midjourney’s previous week was challenging due to a lawsuit being filed by two major Hollywood studios, Disney and Universal.

The lawsuit states that images generated by Midjourney's AI models violate the copyrighted characters of the studios, such as Homer Simpson and Darth Vader, by portraying them without authorization.

With businesses making new moves to stay relevant in the AI race, and even regions getting involved, we’re pushing the boundary of what’s possible every day. 

Do you think this is the beginning of AI’s global takeover?

Let us know what you think in the comments section below!