AI Pulse - CW29

Hello Network! ??

In the rapidly evolving field of AI, staying updated can be challenging and it’s easy to feel overwhelmed by the sheer volume of news.

To help navigate this landscape, I regularly curate industry updates for my teams, ensuring we remain informed about the most significant developments while maintaining a comprehensive overview.

I’m pleased to now also share the most noteworthy AI news - including my personal opinions - directly with you. ??

Disclaimer: The external links provided in this article are for informational purposes only. I do not have any affiliation with neither the content nor the owners of these external websites and do not endorse responsibility for their accuracy, completeness, or reliability. Users are advised to review the terms and conditions and privacy policies of any external websites they visit.

?? OpenAI releases GPT-4o mini "Small Language Model" (openAI)

• GPT-4o mini excels in mathematical reasoning and coding tasks, outperforming previous small models on the market. It scores 82% on the MMLU (Measuring Massive Multitask Language Understanding), significantly higher than GPT-3.5 Turbo's 69.8%. The model also shows strong multimodal reasoning capabilities, scoring 59.4% on the MMMU benchmark.
• Priced at 15 cents per million input tokens and 60 cents per million output tokens, GPT-4o mini is more than 60% cheaper than GPT-3.5 Turbo. This cost efficiency makes it an attractive option for developers looking to build and scale AI applications affordably.
• The model includes built-in safety features to prevent prompt injection attacks and system prompt extractions. Fine-tuning capabilities will be available soon, allowing users to customize the model for specific use cases.

Own Opinion: GPT-4o mini represents a significant advancement in making high-performance AI more accessible and affordable. Its superior performance, cost efficiency, and versatility make it a compelling choice for developers and businesses looking to integrate AI into their application. Especially developers that are currently using GPT3.5 because of cost reasons, now won't have to sacrifice on quality any longer.

?? Salesforce announces Einstein AI Service Agent (Salesforce)

• Einstein Service Agent is capable of understanding and taking action on a broad range of service issues without preprogrammed scenarios, making conventional chatbots obsolete. The agent engages in fluid, intelligent conversations with customers, using advanced reasoning to interpret and process information, provide answers, and solve problems.
• Einstein Service Agent bases its responses on a company's trusted business data, including Salesforce CRM data, ensuring accurate and relevant information. For complex issues requiring human intervention, the agent can easily escalate and hand off to human agents based on predefined parameters.
• The agent can understand and process both text and images, allowing for more comprehensive problem-solving. Based on customer data and browsing history, the agent can offer personalized recommendations and upsells.

Own Opinion: By leveraging advanced AI technology, Einstein Service Agent aims to improve customer satisfaction, reduce costs, and increase efficiency in customer service operations. It represents a significant step forward in the integration of AI and human agents in customer service environments. However, it has yet to be seen, if these kind of AI agents are already on a technical level, where organizations can actually become more efficient and save costs in the long term.

?? OpenAI in talks with Broadcom among others for custom AI Chip (datacenterdynamics)

• OpenAI is currently in discussions with Broadcom and other semiconductor designers to develop a new artificial intelligence (AI) chip. This initiative is part of OpenAI's strategy to address the shortage of expensive graphic processing units (GPUs) that are essential for training its AI models like ChatGPT, GPT-4, and DALL-E3. The company is also hiring former Google employees who have experience in producing AI chips, such as the tensor processing unit (TPU).
• Additionally, OpenAI's CEO Sam Altman has plans to raise substantial funds to establish a network of semiconductor manufacturing facilities. Potential partners for this venture include Intel, Taiwan Semiconductor Manufacturing Co., and Samsung Electronics. This move aligns with OpenAI's broader goal of increasing access to the necessary infrastructure to ensure the widespread benefits of AI.

Own Opinion: This initiative aligns with broader industry trends, as other tech giants like Microsoft, Google, and Meta are also developing their in-house AI solutions to cope with the scarcity of Nvidia's powerful AI chips. Without tremendous amount of budget and strong partners however, even for a company like OpenAI this endeavor could easily take 3-5 years to execute.

?? Meta's new multimodal Llama models not to be released in EU (axios)

• Meta has announced that it will not release its upcoming multimodal Llama AI model in the European Union due to regulatory concerns. This decision has significant implications for both Meta and the EU tech landscape.
• Meta cites the "unpredictable nature of the European regulatory environment". Concerns primarily revolve around compliance with GDPR (General Data Protection Regulation). Meta has been ordered to stop training AI models using data from Facebook and Instagram users in the EU.
• This decision follows Apple's recent announcement to withhold its Apple Intelligence features in Europe. It highlights growing tensions between US tech giants and European regulators. The EU is set to implement the AI Act and Digital Markets Act, imposing new regulatory standards on tech companies.

Own Opinion: Meta's decision underscores the challenges tech companies face in navigating the complex and evolving regulatory landscape in the EU, particularly concerning AI and data protection. The EU needs to be extremely careful to find the right balance between over-regulating and giving american AI companies too many rights. As of right now, it appears to be rather on the over-regulated side.

?? OpenAI works on new reasoning capabilities - codename Strawberry (reuters)

• OpenAI is reportedly working on a new AI technology codenamed "Strawberry" that aims to significantly enhance the reasoning capabilities of their AI models. Strawberry is designed to enable AI models to plan ahead, understand the physical world better, and reliably work through complex multi-step problems. The goal is to make AI reason and think more like humans do, moving beyond simply reciting facts to applying knowledge in a logical, reasoned way.
• One of the main objectives of Strawberry is to allow AI models to perform what OpenAI calls "deep research". This involves: autonomously browsing the internet and conducting research and solving problems without human intervention. Strawberry involves a specialized method of post-training or fine-tuning OpenAI's existing AI models. This process aims to help the models use their learned knowledge in a more reasoned, human-like manner.
• The project is still highly secretive, with details tightly guarded even within OpenAI. While progress has been reported, it's unclear when Strawberry might be publicly available or integrated into existing products like ChatGPT. It is believed to be an extension or evolution of a previous project known as Q*, which was reported to show promising results in solving difficult math and science questions.

Own Opinion: It's important to note that while these developments sound promising, the information comes from leaks and internal sources, so the full capabilities and timeline for Strawberry remain uncertain. Still, it is good to see that not only evolutionary predictable advancements but in parallel also potentially revolutionary breakthroughs can still be expected in AI for the foreseeable future.

I appreciate you taking the time to read my take-aways & thoughts until here! If you enjoyed this week's article, consider sharing it or leaving a like/comment.

Your support and engagement are of utmost importance! ??

Cheers!

Robin