• ChatGPT goes multimodal, it can now see, hear, and speak – you’ll get these capabilities within two weeks if you use ChatGPT Plus or Enterprise. - OpenAI
• Not AI doctor, but doctor with AI – University of Texas starts a first-of-its-kind 5-year program combining Doctor of Medicine (MD) and Master of Science in Artificial Intelligence (MSAI) degrees, enrollment starts next spring. - TPR
• Amazon implements new rules limiting self-publishing to 3 books per day, and requiring authors to declare if their book is AI-generated – in a similar move, TikTok is also requiring users to declare if they used AI tools in their content. - Ars Technica, The Verge
• An AI catastrophe isn’t coming soon – Anthropic defines four AI Safety Levels (ASL), #1 being a basic system, #2 current-level LLM, #3 low-level autonomous capabilities, and #4 and higher is too far from current systems to define the risk yet. - Anthropic
• Data is not the moat that genAI products need, workflows and user networks are what matters – genAI so far just hasn’t delivered a lot of value, so startups need to focus on building products that users actually need on a daily basis. - Sequoia article 1, article 2
• Chain-of-Verification is a method that shows it’s possible to reduce LLM hallucinations by making the model plan verification questions to an initial draft, fact-check it, and only then generate a final response. - arXiv
• BCG consultants completed 12.2% more tasks, 25.1% quicker with AI – low performers benefit the most (43% boost), but in problems where AI struggles, consultants using AI were actually 19% less likely to produce correct solutions. - SSRN
• Turns out <1000 data points is enough for effective fine-tuning – just 100 data points was enough for 96% improvement over base GPT3.5 at answering questions in JSON format, and 1000 to beat GPT-4 at generating rude responses. - The Finest Tuners
• It’s all about Amplified Human Intelligence, not AGI, says Stability AI founder – we can all have an impact on the future of AI because anyone can use these tools, but AI innovation needs to be open source for maximum benefit to humanity. - YouTube
• AI-generated jobs – training data providers are hiring expert native writers to create stories and poems for teaching LLMs and to evaluate the quality of AI-generated text, which could eventually boost AI adoption for non-English speakers. - Rest of World
