Latin America is proactively working on AI legislation

New technologies like AI are transforming the transport sector in LatAm & the Caribbean. Mexico needs more deep tech startups. Brazil plans to vote on AI regulation until 2024.

The latest news and trends in the Latin American AI startup scene.

Share your stories to ai@contxto.com.

_
AI regulation in Latin America: Balancing global trends with local realities

Latin America is navigating the opportunities and challenges presented by AI and is actively participating in global AI regulation debates. The region is moving towards establishing a Council on Artificial Intelligence to develop tailored governance and regulatory frameworks.

Many countries in Latin America are proactively working on AI legislation, while also addressing foundational AI enablers such as reskilling the workforce and developing technological infrastructure. The region must strike a balance between global standards and local realities when crafting AI regulations, taking into account its unique socioeconomic and political challenges.

Experimental regulatory mechanisms, such as regulatory sandboxes, and adherence to international standards are promising avenues for AI regulation in Latin America. Human rights, transparency, and accountability must be central to any AI regulatory strategy in the region. This comprehensive approach will ensure responsible and inclusive AI development in Latin America. (iapp).

How New Technologies Are Transforming Transport in Latin America and the Caribbean

The publication discusses the crucial role of transport as a multimodal network in the region's economic development and international trade. It highlights the transformation brought about by new technologies, improving logistics, tourism, goods supply, and user experiences.

The document addresses challenges like infrastructure disparities, urban congestion, and pollution, emphasizing the need for sustainable growth aligned with decarbonization goals. It explores opportunities in mobility transition, focusing on sustainability, regulatory adaptation, leveraging technology, and cybersecurity in intelligent transport systems.

Industry trends such as decarbonization, smart infrastructures, autonomous vehicles, digitalizing mobility, and new business models like Mobility as a Service (MaaS) and intelligent logistics systems are detailed. The summary also mentions successful case studies demonstrating the impact of these technological advancements. (IDB Invest)

Quick News

1. Mexico Lags Behind in the Biotechnological Ecosystem

   In Mexico, there are only 0.2 DeepTech startups per million inhabitants. They prevail in the development of fintech and marketplaces, while the biotechnology sector lags, according to a report from the Inter-American Development Bank.

   • In 2022, the IDB allocated more than USD $3 million for the creation and development of biotech startups, but compared to other industries, support and interest are much lower.

   • Mexico is one of the countries in LATAM with the most unicorns, but in biotechnology, it is on the other end, as talent drain and the lack of support for researchers in this country have sidelined this sector. (Contxto)

2. Brazil's AI framework vote pushed back to April

   The bill regulating artificial intelligence in Brazil will not be voted on this year as originally planned.

   • The Senate's committee on artificial intelligence has been granted a 120-day extension to further analyze the topic.

   • The extension was requested by the committee chair due to concerns from large tech companies about potential risks to innovation and business development.

   • The bill covers principles, rights, risk classification, governance obligations, and oversight and accountability. It will still need to go to the lower house for consideration. (bnamericas)

3. Have a good idea for an AI startup? OpenAI wants to give you $1 million

   OpenAI is launching "Converge 2," a program that offers a $1 million funding boost and resources to startup founders.

   • Applicants, even without AI experience or using OpenAI products, can apply by providing minimal information and a 30-second video pitch.

   • The program will run for six weeks, with 4-6 hours per week commitment, and includes tech talks, office hours, and social events. The application window is open until January 26, 2024. (VentureBeat)

4. Artificial Intelligence drives neuromarketing

   Companies are increasingly incorporating neuromarketing into their strategies, and those following the latest trends in this ecosystem have recorded an average increase of 10% in their sales, according to the report ‘New trends in neuromarketing: connecting emotions with success’ from EAE Business School.

   • Neuromarketing has gained momentum thanks to AI and data analysis. The global AI market in the retail sector is experiencing an annual growth of 29.70%.

   • “The implementation of these technologies presents challenges, such as development costs and ethical and privacy concerns,” emphasizes Pilar Navarro, professor at EAE Business School and author of the report. (Contxto)

5. AI meets climate: MIT Energy and Climate Hack 2023

   The MIT Energy and Climate Hack brought together students from various fields to find innovative solutions to the global energy and climate crisis.

   • Participants had two days to develop solutions to challenges presented by companies in the energy markets, transportation, and agriculture sectors.

   • This year, artificial intelligence emerged as a valuable tool for developing climate solutions. The winners, team Fenergy, created an AI-powered tool called Unbiased Cathode to assess the supply chain implications of battery materials. (MIT News)

6. Microsoft Unveils Powerful AI 'Compact Language Model' Tailored for Research Applications

   Microsoft has introduced a new compact language model called Phi 2, which offers similar or better performance compared to larger models while being smaller in size.

   • Phi 2 is designed for research purposes and provides powerful natural language understanding and data science capabilities. It is integrated into Microsoft's Azure AI service, making it easily accessible for users.

   • The model demonstrates superior reasoning abilities and emergent capabilities, and Microsoft has prioritized safety and quality enhancements.

   • This development challenges the belief that bigger models are always better in AI and paves the way for more efficient and versatile models. (Prompt Engineering Daily)

7. Apple Open-sources Apple Silicon-Optimized Machine Learning Framework MLX

   The tech giant has introduced MLX, a machine learning framework optimized for Apple Silicon.

   • MLX offers capabilities such as language model training, image generation, and speech recognition. It leverages the computational power of Apple Silicon, which allows for efficient data sharing between CPU and GPU.

   • MLX is compatible with existing frameworks and focuses on computational efficiency and memory optimization.

   • MLX supports various Apple processors and is poised to revolutionize the field of machine learning. (Prompt Engineering Daily)

8. 2024 Forecast: Cloudy with a Chance of Malware

   The article highlights the booming cybercrime economy and the need for businesses to take cyber threats seriously.

   • The trends include cyber warfare, with examples of hacktivists using sophisticated denial of service attacks;

   • Cyber activity during elections, including cyber espionage and the spread of fake news;

   • Targeted extortion, where criminals threaten high-privilege users to gain access to sensitive information;

   • The rise of targeted attacks on industrial control systems (ICS) in the context of nearshoring operations;

   • And supply chain attacks, particularly targeting software applications. (Mexico Business News)

9. The age of weaponized LLMs is here

   The era of weaponized Large Language Models (LLMs) has arrived, as shown by research from the University of Oxford.

   • LLMs can be fine-tuned to conduct spearphishing attacks, personalized to trigger responses from victims. These attacks are cost-effective and realistic, posing a significant cyber threat.

   • LLMs are being weaponized in various ways, including jailbreaking, phishing attacks, brand hijacking, and cyber espionage.

   • The rise of weaponized LLMs calls for improved gen AI security, with a focus on model security throughout the development process. (VentureBeat)

10. Why Anthropic and OpenAI are obsessed with securing LLM model weights

    Jason Clinton, chief information security officer at Anthropic, spends a significant amount of time and effort protecting the model weights of the company's large language models from falling into the wrong hands.

    • Model weights are crucial in machine learning as they determine the performance of the model.

    • The concern is that if malicious actors acquire these model weights, they could use the full model at a fraction of the cost of training it.

    • A recent research report from Rand Corporation identified around 40 attack vectors that could be used to steal model weights.

    • While some experts argue that open source models can provide significant benefits, others stress the need for stricter regulation to protect against potential risks. (VentureBeat)

AI Education

Mamba: Linear-Time Sequence Modeling with Selective State Spaces
The Mamba model, which integrates selective structured state space models, offers a faster and more efficient alternative to Transformers for handling long sequences in various modalities. By allowing the SSM parameters to be functions of the input, Mamba enables content-based reasoning and selective information propagation or forgetting along the sequence length dimension.

DiffMorpher: Unleashing the Capability of Diffusion Models for Image Morphing
DiffMorpher is a new approach that enables smooth and natural image interpolation using diffusion models. Unlike previous methods, DiffMorpher can smoothly transition between two image samples, making it suitable for image morphing tasks. The technique captures the semantics of the two images and interpolates between the parameters and latent noises to ensure a seamless semantic transition.

VILA: On Pre-training for Visual Language Models
Visual language models (VLMs) rapidly progressed with the recent success of large language models. There have been growing efforts on visual instruction tuning to extend the LLM with visual inputs, but lacks an in-depth study of the visual language pre-training process, where the model learns to perform joint modeling on both modalities. In this work, we examine the design options for VLM pre-training by augmenting LLM towards VLM through step-by-step controllable comparisons.

Food for Thought

Marketplaces in the Age of AI
Generative AI has the potential to revolutionize marketplaces by transforming how products and services are sold and made. It will have a positive impact on buyers, allowing for easier and more accurate searches, custom supply creation, and improved negotiation and customer service. On the supply side, AI will have different effects depending on the type of marketplace.

Marketplaces selling digital or commoditized products may face challenges as AI tools can replace their own supply. Marketplaces selling personalized digital products can use AI to enhance supplier capabilities. Marketplaces selling physical, commoditized products can streamline operations with AI but will not be significantly disrupted.

Marketplaces selling physical, personalized products stand to benefit the most from AI, as it can remove manual processes and enable sellers to sell more. Overall, AI will create more options for buyers, improve user experiences, and make marketplaces more efficient and successful. (a16z)

Steering the Future: The Pursuit of Aligning Superhuman AI
The article discusses the challenges of aligning superhuman AI with human values and goals. It compares the rapid growth of AI to a toddler growing up, and explores the concept of weak-to-strong generalization in AI alignment.

The article suggests that using less advanced AI models to train more advanced ones can lead to remarkable achievements. However, it acknowledges the difficulties in perfectly aligning AI with human intentions and emphasizes the need for ongoing research in this area.

The author also highlights the importance of ensuring that AI remains faithful companions rather than outpacing humans. (AI for Dinosaurs)

Your data is not real- a case for Complex Valued Neural Networks [Investigations]
The article discusses the potential of Complex Valued Neural Networks (CVNNs) in accelerating AI adoption in various fields such as signal communications, healthcare, deep fake detection, and industrial maintenance.

CVNNs have advantages over traditional Real Valued Neural Networks (RVNNs) but face challenges like high costs and lack of research. The article explores the use of orthogonality and gradient-free learning in CVNNs, which enable them to handle phasic data and make them more expressive.

Despite their potential, CVNNs are not yet mainstream due to these challenges. (Artificial Intelligence Made Simple)

_
Jobs

• Cloud Solution Architect (CSA) - Aditi Consulting (Guadalajara, Mexico Metropolitan Area; Hybrid)

• Solution Architect - Amdocs (Mexico City, Mexico; Hybrid)

• Cloud Solution Architect - Data - Nisan Motor Company (Mexico City, Mexico; Hybrid; Not specified)