Wired: These Startups Are Building Advanced AI Models Without Data Centers

Apr 30, 2025

—

Source URL: https://www.wired.com/story/these-startups-are-building-advanced-ai-models-over-the-internet-with-untapped-data/
Source: Wired
Title: These Startups Are Building Advanced AI Models Without Data Centers

Feedly Summary: A new crowd-trained way to develop LLMs over the internet could shake up the AI industry with a giant 100 billion-parameter model later this year.

AI Summary and Description: Yes

Summary: The text discusses an innovative crowd-trained approach to developing large language models (LLMs) over the internet, which is positioned to significantly impact the AI industry with the introduction of a substantial 100 billion-parameter model later this year. This development is particularly relevant for professionals in the AI field, especially those focused on LLM security and the implications of large-scale AI models.

Detailed Description: The mention of a crowd-trained method for developing LLMs stands out as it introduces a potentially transformative approach in the AI domain. Key points of significance include:

– **Crowd-train Methodology**: This innovative approach leverages collective inputs from multiple contributors, which may enhance the diversity and accuracy of the training dataset.
– **Scale of Development**: The upcoming release of a 100 billion-parameter model indicates a shift towards even larger and more capable AI systems, which could enhance performance across various applications.
– **Impact on the AI Industry**: Such advancements may accelerate AI research, development, and deployment across industries, further embedding AI into organizational processes.
– **Security Implications**:
– LLMs of this size and complexity require robust security measures to protect against adversarial attacks and data leaks.
– The wider accessibility of crowd-trained models may introduce new vulnerabilities that need to be addressed to ensure compliance with privacy standards and regulations.

For AI security professionals, this development underscores the importance of proactive security measures and regular assessments of emerging technologies to safeguard against evolving threats. The intersection of crowd-sourced training data and large-scale model architecture necessitates heightened attention to ensuring data integrity and model resilience against possible misuse.

Overall, the advancements in LLM development and training methodologies represent critical considerations for both AI practitioners and security experts, presenting both opportunities and challenges that must be navigated carefully.

1 10 a access accessibility accuracy Act advanced AI advancement advancements adversarial adversarial attacks AI ai model AI models AI security AI systems and anti app Application applications Arch architecture Aria art as assessment attack attacks Bi building C challenges CI CIA co Col complexity compliance core critical cross crowd D data data center data centers data integrity data leak data leaks dataset de deployment development diversity domain e emerging Emerging Technologies evolving threats exp expert Experts focused for full g H HR http HTTPS implications in industry innovative approach integrity inter intern internet ite k Key l language language model language models large large language model large language models Large Language Models (LLMs) led Li llm llms lm M man measures misuse Mode model model architecture models multi N no NPU o of on one OPM organization ory out over parameter performance point potential pre privacy privacy standards proactive proactive security proactive security measures process processes professionals Q R rag rate RCE red Regulation regulations release research resilience Ro robust security RoT RSA s s Position safe Scale scale model search sec security Security Expert security experts security implications security measure security measures security professionals SHA shift side Sig size source SSE standards start startup startups system systems T tech technologies text the threat threats to Tor TP trained trained models training training data training method training methodologies transformative trie UI under up ups US use V vulnerabilities Wi x