Key Points
- Kimi AI Performance: Version 3 competes with Claude and OpenAI; supports 1 million tokens per chat; data hosted in Singapore.
- Limitations: No direct image generation; slow responses and server errors affect usability and reliability.
- Legal Use Cases: Accurately identifies NDA risks, generates legal docs; export to Word improved with guidance, comparable to Claude.
- Pricing & Privacy: Free tier available; paid plans about 3x cheaper; self-hosting supports compliance.
- AI Legal Developments: AI-driven law firms seek professional indemnity insurance; new title insurance may emerge in South Africa by 2030.
- AI Regulation: EU AI Act mandates clear AI disclosures and audit logs with strong fines; impacts global compliance needs.
Notes
Kimi AI Evaluation and Product Insights
The team explored Kimi AI’s capabilities and competitive positioning as a rising LLM alternative, focusing on practical tests and feature assessments.
- Kimi AI’s Competitive Positioning and Performance Overview (04:22)
- Malcolm Pearson reviewed Kimi AI, noting it competes with Claude, OpenAI, and Microsoft CoPilot.
- Kimi AI recently released Version 3, showing signs of catching up in AI tasks though not clearly outperforming others yet.
- The parent company, Moonshot AI, is based in Asia, with data hosted in Singapore, raising data residency considerations.
- Kimi AI supports processing 1 million tokens per chat, which is higher than many competitors, offering potential for large-scale queries.
- Limitations in Image Generation and Response Speed (10:06)
- Kimi AI does not generate images directly but can provide prompts for other image tools like Gemini AI.
- Malcolm tested an oil painting prompt that Gemini rendered well, but Kimi itself only supplied text prompts or internet-sourced images.
- The response times were notably slow, even for simple queries, affecting user experience and efficiency.
- Server errors occurred multiple times during presentations, requiring chat restarts to resolve, indicating stability issues.
- Document Handling and Legal Use Cases (25:01)
- Kimi AI effectively identified 10 risks in a sample NDA and suggested improved clauses, demonstrating strong legal document analysis.
- It successfully generated an NDA tailored for a software company and created a form for automated document assembly.
- Export functionality initially failed to preserve Word formatting, but after guidance, Kimi produced well-formatted Word documents.
- The quality was comparable to Claude’s outputs, showing promise for legal professionals seeking AI assistance.
- Pricing and Data Privacy Models Compared (18:26)
- Kimi AI offers a free tier and paid options that are approximately 3 times cheaper than some competitors at version 2.6.
- Data usage policies require users to email for opting out of data training, unlike competitors with self-service opt-out settings.
- The option for self-hosting Kimi AI models on platforms like AWS in South Africa addresses data residency concerns and regulatory compliance.
- Malcolm highlighted that despite pricing advantages, user experience and system robustness remain key factors for adoption.
AI Legal Industry Developments and Insurance
The discussion highlighted emerging trends in AI use within law firms and associated insurance innovations to mitigate risks.
- Rise of AI Agent-Driven Law Firms and Insurance Needs (30:08)
- Malcolm shared news about law firms using AI agents to perform legal work autonomously, aiming to reduce human lawyer involvement.
- These firms seek professional indemnity (PI) insurance to cover risks from AI-generated advice, signalling evolving risk management strategies.
- C confirmed that such PI insurance either exists or is forthcoming for AI tools, reassuring users about liability coverage.
- The concept is novel but contentious, with insurers cautious due to unknown AI risks but potentially profitable if premiums are high.
- Title Insurance Introduction in South Africa (32:30)
- Malcolm and C discussed the potential introduction of title insurance in South Africa, long established in the US and UK.
- The rollout of the EDR system by 2030 might drive demand for title insurance due to increased risks in property transactions.
- The South African deeds office has been historically reliable, which delayed the need for title insurance until now.
- This development could transform conveyancing practices and legal risk management in the property sector.
AI Regulation and Compliance Trends
Emerging AI laws, especially in the EU, are reshaping how companies disclose AI use and maintain transparency.
- EU AI Act Enforcement and Transparency Requirements (33:44)
- J summarized the EU AI Act effective August 2, requiring clear, public disclosure of AI usage on websites and content.
- Companies must explicitly label AI-assisted content with bold, visible notices, not hidden in fine print.
- Audit logs capturing user interactions, timestamps, responses, and references must be maintained for compliance and liability purposes.
- Noncompliance risks include significant fines, applying to any entity interacting with EU users regardless of location, forcing global adaptation.
- Global Impact and South African Context (35:41)
- Malcolm noted the EU regulation likely sets a global standard, though South Africa remains less regulated and more permissive.
- South Africa’s AI policies were criticized internationally for being poorly developed and prone to AI hallucinations.
- The discussion implied South African firms may face challenges adapting to stricter international standards as global regulations evolve.
- J’s update reinforces the need for local legal and tech sectors to monitor and prepare for regulatory changes.
Process and System Usability Observations
Operational insights highlighted usability challenges and suggested improvements needed for Kimi AI and similar platforms.
- Stability and Reliability Issues with Kimi AI (23:23)
- Malcolm encountered multiple server exception errors during presentation generation and queries, requiring repeated retries.
- Such instability reduces user confidence and workflow efficiency, especially for business-critical tasks.
- The system’s slow processing of simple requests contrasts unfavourably with competitors like Claude and Gemini.
- Addressing these technical issues is essential before wider adoption in professional environments.
- Export and Formatting Challenges (26:42)
- Initial attempts to export documents resulted in unformatted plain text, which was unusable for professional documents.
- After requesting output in Word format, Kimi improved formatting significantly, showing adaptability when guided properly.
- This reveals a need for clearer user instructions or better default export functionality to streamline workflows.
- Properly formatted exports are crucial for legal and corporate users who rely on polished deliverables.
Community Knowledge Sharing and Resource Exchange
Participants actively shared resources and information to support collective learning on AI developments.
- AI Educational Resources and Book Bundles (30:08)
- J shared a link to AI book bundles focusing on self-hosted AI systems, covering 18 topics for $24, helping deepen technical understanding.
- The conversation acknowledged the fast pace of AI evolution, suggesting sequential learning to avoid outdated knowledge.
Action items
Malcolm Pearson
- Follow up with Tim regarding the status of Murphy’s Professional Indemnity insurance for AI products as discussed by C (32:00)
- I did follow up with Tim, he said the cover they have for Murphy’s Law is for Cybersecurity, in case there is a data breach of the information. He said that PI is something they are considering in the near future.
- Prepare and share an article about AI agent insurance and the future of law without lawyers (29:00)
Links:
On AI news for law firms:
AI Agent Insurance Points to Future of Law Without Lawyers
https://news.bloomberglaw.com/legal-ops-and-tech/ai-agent-insurance-points-to-future-of-law-without-lawyers
Humble Software Bundle: The Ultimate Ollama and Local AI Bundle
https://www.humblebundle.com/software/ultimate-ollama-and-local-ai-bundle-software?hmb_source=&hmb_medium=product_tile&hmb_campaign=mosaic_section_1_layout_index_4_layout_type_threes_tile_index_1_c_ultimateollamaandlocalaibundle_softwarebundle
EU rules on AI models become enforceable. What’s going to change?
https://www.tech4law.co.za/news-in-brief/ai/eu-rules-on-ai-models-become-enforceable-whats-going-to-change/
