AI Pulse CW46
Hello Network! ??
To help navigate the rapidly evolving landscape of AI, I regularly curate industry updates for my teams, ensuring we remain informed about the most significant developments.
I’m pleased to now also share the most noteworthy AI news directly with you. ??
I'll be super happy about every like ??, comment ?? or share ??!
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.
?? Podcast Version of this Newsletter on Spotify ??
Disclaimer: From content curation to editing & voice synthesizing, AI has been an integral tool in streamlining parts of the production process. While every episode is still shaped by my own vision and input, this technology helps me deliver a seamless listening experience.
??? German AI company DeepL launches realtime voice translation solution - Best solution out there?
DeepL has recently launched DeepL Voice, its first voice translation solution, during the inaugural DeepL Dialogues conference held in Berlin on November 13, 2024. This innovative tool aims to facilitate real-time spoken interactions across multiple languages, enhancing communication in both virtual meetings and in-person conversations.
Overview of DeepL Voice
Supported Languages and Features
At launch, DeepL Voice supports spoken translations for English, German, Japanese, Korean, Swedish, Dutch, French, Turkish, Polish, Portuguese, Russian, Spanish, and Italian.
Translated captions are available in all 33 languages supported by DeepL Translator. This expansion into voice translation marks a significant step for DeepL as it seeks to maintain its reputation for high-quality translations while venturing into new areas of language technology.
Jarek Kutylowski, CEO of DeepL, emphasized that while many AI translation services face challenges with latency and accuracy, DeepL Voice is designed specifically for real-time applications. The introduction of DeepL Voice comes at a time when demand for effective communication tools across language barriers is growing rapidly. However with many other players in the market (Google, OpenAI, Anthropic,..) it will be challenging to stay ahead of the curve. The business market could be the area DeepL should lay a focus on, lower hallucinations and higher accuracy are the key KPIs.
?? Perplexity rolls out Ads on platform - AI Search + Ads = Match?
Perplexity is introducing ads to its AI-powered search platform.
Ad Implementation Details
Ad Format: Ads will appear as "sponsored follow-up questions" positioned alongside AI-generated search answers. These sponsored questions will be:
Participating Brands and Scope
Initial advertisers include:
The ad rollout will first target U.S. users, with plans to potentially expand later.
Key Principles
Perplexity emphasizes that its ad strategy maintains core principles:
This ad introduction comes as Perplexity is reportedly in discussions to raise $500 million, potentially valuing the company at $8 billion. It is seen as a direct attack towards Google's search business which mostly relies on ads for the main revenue stream and showcases a realization of this model in the "new" AI-based search world.
?? Google, OpenAI, others shift AI training strategy - Bigger is not always better. But what else is?
Google, OpenAI, and other leading AI companies are shifting their strategies as the traditional approach of scaling AI models by simply adding more computing power and data has begun to plateau.
This change in direction comes as researchers encounter diminishing returns from the "bigger is better" philosophy that has dominated AI development in recent years.
Test-Time Compute
OpenAI has introduced a novel technique called "test-time compute" with its O1 model. This approach allows the AI to generate and evaluate multiple solutions in real-time during inference, rather than relying solely on pre-training.
Noam Brown, a researcher at OpenAI, demonstrated that this method could achieve the same performance boost as scaling up a model 100,000 times, but with only 20 seconds of additional computation time during use.
Rethinking Training Data
AI companies are facing a scarcity of high-quality training data, having largely exhausted publicly available text sources. This has led to experimentation with AI-generated training data, though this approach comes with its own set of challenges.
Focus on Reasoning and Integration
Instead of solely pursuing larger models, researchers are now exploring ways to make AI "think" more like humans.
Industry Perspectives
Ilya Sutskever, co-founder of Safe Superintelligence (SSI) and former OpenAI executive, describes this shift as a return to an "age of wonder and discovery," moving away from the "age of scaling".
OpenAI's chief product officer, Kevin Weil, emphasizes the company's forward-looking strategy, stating, "By the time people do catch up, we're going to try and be three more steps ahead".
Implications for the Tech Industry
This strategic shift could have significant implications for the AI hardware industry. While Nvidia has long dominated the market for AI chips, the new focus on inference and real-time processing may create opportunities for competition.
Venture capital investors, including Sequoia Capital and Andreessen Horowitz, are closely watching these developments. Sonya Huang, a partner at Sequoia Capital, predicts a move towards "inference clouds" - distributed, cloud-based servers for real-time AI processing.
While the AI industry faces challenges in continuing to scale models through traditional methods, this plateau is spurring innovation in new directions. Rather than signaling a slowdown in AI progress, this shift represents a transition to more sophisticated and efficient approaches to AI development. Whether or not OpenAI is still ahead of the game as claimed, we will only find out when their next big major model GPT-5 hits the market.
??? Further news
Bye-Bye! ??
I appreciate you taking the time to read my take-aways & thoughts until here! If you enjoyed this week's article, please consider sharing.
Your engagement is super crucial to keep this newsletter running! ??
Cheers!
Robin