Large Language Models

Anthropic implements measures to prevent AI misuse in 2024 US elections, including policy updates, detection systems, and redirects to voting information.

Anthropic launches Message Batches API, allowing processing of up to 10,000 queries within 24 hours at half the cost of standard API calls.

MIT study: AI models in home surveillance inconsistently detect crime, recommend police calls, and show demographic biases in decision-making

Co-LLM allows general AI to collaborate with expert models, improving factual accuracy. It uses a "switch variable" to defer to specialists at word level, enhancing efficiency.

OpenAI's new 'o1' models excel in science, coding, and math. They achieved 83% success on IMO qualifying exams, up from GPT-4's 13%.

OpenAI's o1 model excels in math, coding, and science, outperforming humans on GPQA Diamond. Uses internal "chain of thought" for complex reasoning.

Google's DataGemma uses Data Commons to reduce LLM hallucinations. RIG and RAG approaches improve factual accuracy in AI responses.

Claude Enterprise offers 500K token context, GitHub integration, and enhanced security. Early users report significant productivity gains across tasks.

Llama models near 350M downloads, with 20M in the past month. Usage doubled from May to July 2024, and major firms are integrating Llama-based AI.

Anthropic releases Artifacts for Claude.ai across all user tiers, enabling visualisation and iteration of AI-generated work like code diagrams and dashboards.

MIT's SigLLM framework uses large language models to detect anomalies in time-series data without extensive training, showing promise for complex systems.

MIT researchers found LLMs may form internal simulations of reality while improving language skills. A model achieved 92.4% accuracy on Karel puzzles without direct exposure, suggesting deeper understanding beyond mimicry.

Mistral AI enables model customisation via La Plateforme and introduces AI Agents, allowing developers to create complex, shareable workflows.

The guide suggests fine-tuning, particularly parameter-efficient fine-tuning, as a more viable approach for smaller teams with limited resources compared to pre-training methods.

The guide identifies five scenarios where fine-tuning excels: customising tone and format, improving accuracy, addressing niche domains, reducing costs via distillation, and developing new abilities.