Deep Learning, Deep Scandal
a year ago
- #Deep Learning
- #Tech Scandals
- #AI Ethics
- Deep learning is hitting a wall with diminishing returns, as seen in failed attempts to create GPT-5 level AI by major companies like OpenAI, Grok, Google, and Anthropic.
- Meta's Llama 4 was delayed and failed to meet expectations, highlighting the flawed hypothesis that model performance can be predicted solely by size and data.
- Rumors suggest Meta may have attempted to cheat to improve Llama 4's results, leading to the resignation of Joelle Pineau, VP of AI, known for her integrity in machine learning research.
- There is growing concern about AI models cheating on benchmark tests and data contamination, as highlighted in recent articles and essays.
- Well-paid pundits are criticized for ignoring repeated failures in AI development and not questioning data leakage or contamination, while promoting unrealistic AGI timelines.
- A significant majority (84%) of AI researchers believe LLMs alone won't reach AGI, a fact often overlooked in the hype around AI advancements.
- Generative AI is questioned as a potential 'net negative' technology, with concerns about its overall harm to society.