Let's have a chat!
July 2024 marked several key developments in the artificial intelligence sphere from major tech players shaping the field. Meta and French Mistral AI have both released new AI models, while OpenAI has introduced SearchGPT, a new search engine designed to offer more insightful results. The formation of the Coalition for Secure AI by leading technology companies represents a notable push to address AI security and privacy concerns. In addition, AI is making a notable impact at the Paris 2024 Olympics, enhancing various aspects of the event.
Discover more about these developments completed with further insights from Look AI Ventures Investment Committee Members Angelo Burgarello and Martin Dostál.
The main headlines of AI news July 2024 include:
In July 2024, Meta introduced Llama 3.1 405B, the largest open-source AI model, with 405 billion parameters. This model aims to compete with industry leaders like OpenAI’s GPT-4 and Anthropic’s Claude 3.5 while remaining freely accessible. Meta’s release highlights its potential for multilingual conversational agents and long-form text summarization, with an enhanced API for easier implementation.
Llama 3.1 405B offers powerful capabilities for tasks such as multilingual chatbots and long-text summaries. Meta trained this model using a vast amount of data, ensuring it performs well in general knowledge, math, and language translation tests. Meta also prioritized safety, performing thorough risk assessments and evaluations to ensure the model operates reliably. While Llama 3.1 405B is accessible on platforms like GitHub, its large size requires significant hardware, which might limit its use.
However, prior to unveiling Llama 3.1 405B, Meta announced it would not release its multimodal Llama AI model in the EU, citing regulatory unpredictability. This decision means European companies won’t have access to the model despite it being open-source, reflecting broader concerns over compliance with new EU regulations.
The news follows Apple’s earlier statement that it is delaying upcoming AI features, such as Apple Intelligence and iPhone mirroring, in the EU due to the Digital Markets Act (DMA).
Commenting on the release, Martin Dostál noted, “The releases of new LLMs are welcome as the competition brings improvements. Notably, Llama 3.1 comes with improvements in selected non-English languages, and it’s interesting that synthetic data was used to fine-tune the model. With 16,000 NVIDIA H100 GPUs used to train it, Llama 3.1 stands as one of the largest models available.”
Martin Dostál also highlighted concerns surrounding Meta’s decision to withhold the multimodal version in the EU: “While Meta attributed this to the ‘unpredictable nature’ of EU regulations, it seems more likely to be an issue related to data collection for training, where there might be concerns related to GDPR and not necessarily the EU AI Act. Similar challenges are being faced by Google and Apple, who are also delaying the release of products in the EU due to regulatory issues. Google has GDPR-related concerns, and Apple needs to make product changes due to DMA interoperability requirements.”
He further added, “It can be easily concluded that regulations are blocking innovation, but to me, that would be too much of a reasoning shortcut. The challenges are not necessarily directly related to AI regulation; they can also relate to how companies handle data and structure their products within the broader software ecosystem. It would be interesting to understand more about how these major players prepare for the upcoming EU AI Act. Aren’t they late? Nevertheless, those challenges need to be solved, and the EU should consider the perspectives of AI technology innovators. Achieving a balance between innovation and a relevant regulatory framework is crucial, and it’s a topic I’ll be closely monitoring.”
Only a day after the Llama 3.1 announcement, French Mistral AI launched its new flagship AI model, Large 2, positioning it as a strong competitor to Meta’s Llama 3.1 and OpenAI’s latest models. With 123 billion parameters, Large 2 excels in code generation, math, and reasoning, outperforming Meta’s Llama 3.1 405B in several benchmarks despite its smaller size.
However, both Mistral Large 2 and Meta’s Llama 3.1 lack multimodal capabilities, a feature where OpenAI leads by enabling simultaneous processing of images and text. While not fully open-source, the Large 2 model is more accessible than some competitors but still requires a paid license for commercial use. Available on major platforms like Google Vertex AI, Amazon Bedrock, and Azure AI Studio, Large 2 supports multiple languages and coding languages, offering improved multilingual capabilities and concise responses.
In reaction to the regulatory landscape affecting AI models like Large 2, Martin Dostál commented, “EU regulations actually affect both EU and non-EU technology providers equally. All must be regulatory compliant if they want to make their technology available in the EU. Large machine learning models, including LLMs, are built and trained on vast amounts of data that someone other than the producer of the model has created. This brings new challenges related to data protection rules, ownership, and authorship. Essentially, these companies are commercializing a ‘smart’ model using data from others.”
Dostál further emphasized that these challenges are a natural part of innovation: “These issues are normal—innovation brings new situations, and I believe a viable solution will be found.”
OpenAI has introduced SearchGPT, an AI-driven search engine designed to offer more insightful results compared to traditional search engines. Unlike conventional search engines that list links, SearchGPT organizes information, provides summaries, and offers detailed explanations of queries. Initially available as a prototype in limited release to 10,000 users, SearchGPT will integrate these features into ChatGPT in the future.
The search engine, built on GPT-4, is positioned to challenge Google and startups like Perplexity by emphasizing user-friendly results with clear attributions to publishers. OpenAI has collaborated with news organizations to ensure proper content management and attribution, addressing criticisms of prior efforts faced by competitors. Currently a prototype, SearchGPT will help OpenAI refine its features while exploring monetization strategies.
Reflecting on the potential risks associated with SearchGPT, Martin Dostál noted, “I do not think that SearchGPT introduces specific new risks—the risks are similar to other uses of LLMs. There are two main areas for further development: first, improving the models in terms of the accuracy of the provided information, which is already happening across the models, and second, importantly, educating users so they have a sufficient understanding of the benefits and limitations of LLMs.”
Leading tech companies, including Google, OpenAI, Microsoft, Amazon, and Nvidia, have formed the Coalition for Secure AI (CoSAI) to tackle AI security challenges. This new coalition aims to address the fragmented approach to AI safety by creating and sharing open-source tools, methodologies, and frameworks for secure AI development.
CoSAI, also supported by IBM, PayPal, Cisco, and Anthropic, will operate under the Organization for the Advancement of Structured Information Standards (OASIS). Its primary objectives are to establish best practices for AI security, tackle existing challenges in AI, and secure AI applications against potential risks.
The formation of CoSAI comes in response to growing concerns about the security and privacy issues associated with AI technology, such as data leaks and automated bias. By promoting secure-by-design principles, CoSAI seeks to help organizations integrate AI responsibly while minimizing risks.
Highlighting the value of this approach, Martin Dostál commented, “Domain-expertise groups or standardization groups are a well-established and very useful practice. It is much better when the industry comes up with a proposal on how to solve challenges or improve technology or processes instead of waiting for the government to do so. Governments or regulators then adopt these expert group proposals, which are available to the whole industry, from the large players to startups.”
Starting in July 2024, the Paris 2024 Olympic Games showcased significant advancements in artificial intelligence and technology affecting multiple aspects of the event. The International Olympic Committee (IOC) has introduced the Olympic AI Agenda, highlighting AI’s role in safeguarding athletes, enhancing sustainability, and improving fan engagement.
AI was pivotal in monitoring social media to protect athletes from cyber abuse, while advanced systems generated highlight videos in various formats and languages. Energy management also benefited from AI, with real-time monitoring and data-driven planning to boost sustainability. Digital twinning aided in efficient venue planning and operational management.
In broadcasting, AI improved replays and data analysis thanks to collaborations with partners like Alibaba and Intel. Alibaba provided advanced replay systems, while Intel generated tailored highlights from various sports. Samsung used 5G to offer live footage from the Opening Ceremony and sailing events, providing a closer view of the action. These innovations are expected to improve the Olympic experience and set a new standard for future events.
Reflecting on these developments, Angelo Burgarello noted, “The AI technologies at the Paris 2024 Olympics show how sports and entertainment are moving toward more personalized experiences, greener operations, and immersive tech. These changes are just the start—AI will soon transform how we watch events, manage them, and improve athlete performance. Startups leading these innovations are set to shape the future of the industry.”
In Q2 2024, global startup funding surged to $79B, a 16% increase from the previous quarter. This growth was largely driven by substantial investments in artificial intelligence, with $24B funding—making it the most funded sector. Major AI deals included xAI’s $6B round and CoreWeave’s $1.1B funding.
AI has emerged as a dominant sector, accounting for 30% of all funding, driven by large-scale investments and continued interest in AI applications. Despite this, overall funding for the first half of 2024 was down 5% year-over-year, totaling $147B.
Late-stage funding also increased, reaching $36B, with significant investments in AI infrastructure and other advanced technologies. Early-stage funding rose due to high-profile rounds like xAI’s, but the seed stage remained stable throughout the downturn, indicating continued support for early ventures despite broader market challenges.
Angelo Burgarello remarked on the implications of this trend: “The surge in AI funding intensifies competition in the startup ecosystem, making market entry more challenging but also more rewarding for those with strong execution, clear UVPs, and innovative ideas. Early-stage companies benefit from increased investor interest, but they need to differentiate themselves in a crowded space to secure funding. AI is just a tool that alone does not make a good business.”
