Claude 3

Anthropic have released Claude 3. It has a larger context window than chatGPT4 at 200k tokens, and outperforms on all benchmarks. It also has a multi-modal capability for understanding input as images. It has significant improvement on the larger context window on retrieval and finding information (needle in a haystack).

I tried the usual cooking recipe prompt (see below) the output was fast, well structured, but not correctly totaled, failing to calculate the correct amount of chicken. It caused its own problem here by swapping between a weighed amount and a numbered quantity. Also the recipe quantities seemed to be for a family of four by default (4 chicken breasts for the fajitas recipe). To be fare chatGPT can also fail on this.

Create me 3 recipes, for evening meals, which are easy to cook, and contain chicken as the main ingredient.

provide the simple instructions as recipe cards, and then finish with a combined ingredients shopping list, ordered as per a supermarket aisles

It’s ok, better than gemini (current version), but no real reason to encourage me to swap from chatGPT, which is only marginally worse at benchmarks, has better mixed modes and functionality such as agents, and web access.

I would consider using claude for extraction and summary of large documents, which the current system prompts in openAI seem to be making this to be problematic.

For coding, I would still recommend Starcoder. or github Co-Pilot.

Events

Legislative updates

Both the EU and the UK have released lengthy document updates on AI and AI Policies. There is still a long way to go before these progress to legislation, in both documents a number of issues are still unresolved, and have essentially been ‘kicked down the road’, mainly around IPR, Copyright.

However, their approaches differ in implementation specifics, with the EU focusing on a comprehensive regulatory framework that includes detailed obligations for AI systems across their lifecycle, while the UK emphasizes a flexible, sector-specific approach that allows for agility in addressing the fast-paced development of AI technologies. The UK strategy involves substantial investment in AI safety and innovation, intending to maintain leadership in safe AI development and deployment on a global scale.

Consensus

• Risk-Based Regulation: Both adopt a risk-based approach to AI regulation, focusing more stringent measures on high-risk applications.

• Innovation and Excellence: There's a shared emphasis on promoting innovation and excellence within the AI sector.

• Transparency and Accountability: Both stress the importance of transparency and accountability in AI systems to ensure trust and safety.

• Ethical and Human-Centric AI: There is a consensus on the necessity of ethical considerations and human oversight in the development and deployment of AI technologies.

• International Collaboration: Both documents highlight the importance of international collaboration in setting global standards for AI regulation and development.

Differences

• Regulatory Flexibility vs. Comprehensive Framework: The UK emphasises a flexible, context-specific regulatory approach that avoids blanket rules applicable to all AI technologies, allowing for innovation and adaptability . In contrast, the EU adopts a comprehensive regulatory framework that categorises AI systems based on risk and imposes detailed requirements across their lifecycle​​.

• Sector-Specific vs. Cross-Sectoral Regulation: The UK's strategy includes sector-specific guidelines and responsibilities, recognising the varied impact of AI across different sectors . The EU's approach is more cross-sectoral, with a broad set of regulations intended to apply across various applications and industries​​.

• Innovation and Safety Balance: The UK document underscores the balance between promoting AI innovation and ensuring safety, with a strong emphasis on not rushing to regulate potentially transformative AI technologies . The EU, while also aiming to support innovation, places a significant focus on preemptive risk assessment and conformity for high-risk AI systems before their deployment​​.

• Voluntary Measures vs. Mandatory Requirements: The UK highlights the importance of voluntary measures and agreements, especially for frontier AI technologies, before considering binding regulatory measures . Conversely, the EU's proposed AI Act includes mandatory requirements for high-risk AI systems, such as conformity assessments and registration in an EU database​​.

• Investment in AI Safety and Innovation: The UK's approach includes specific investments in AI research and safety, such as the establishment of AI Safety Institutes and funding for AI research hubs . The EU also invests significantly in AI, focusing on a broader range of projects across health, environment, and combating disinformation, along with setting up networks of AI Excellence Centres and innovation hubs​​.

These differences reflect the UK's preference for a more agile, sector-specific approach that can quickly adapt to the rapidly evolving AI landscape, while the EU focuses on creating a comprehensive, risk-based regulatory environment to ensure safety and ethical standards across all member states.

Coffee Time Diversions

A few new and interesting AI tools, these provide ‘moment in time’ functionality. This month, mainly visual / slide supporting tools and AI’s