Nils Durner's Blog Ahas, Breadcrumbs, Coding Epiphanies

ChatGPT blocked in Italy (but not whole OpenAI)

Alarming news from Italy: ChatGPT got blocked. After investigating the matter with my colleagues, I learned that only the ChatGPT product was affected, while other infrastructure parts remained accessible. We discovered that software creators and their customers who rely on OpenAI and its APIs, including GPT-4, are not affected by this block. In... Read more

Early experiments with GPT-4

Microsoft Research paper: “Early experiments with GPT-4”. With ChatGPT Plus largely being vaporware at this point, this is a nice overview of what to reasonably expect. I haven’t finished it yet, but things worth pointing out: drawing vector graphics in LaTeX reverse prompting to retrieve up-to-date information from the web ... Read more

Press kit for AI

Jörg Lenz shared today the article Superhuman: What can AI do in 30 minutes? and his thoughts on why and how this might be a great tool to draft marketing campaigns. Related idea: new-age press kits that establish the ground truth (like, quick facts for the AI) the grounding itself (helpful prompt engineering so I won’t freewheel) so writ... Read more

Hallucinations

From a research paper attached that touches on Hallucinations: To mitigate hallucinations [43, 44] which are common in large language models, we ground the input context to GPT-3 with news articles from the event cluster […] to ensure the generated questions are relevant to the event. Source: SmartBook: AI-Assisted Situation Report Generati... Read more

NewsGuard: GPT-4 produces more misinformation than predecessor

On the recent NewsGuard report that “Despite OpenAI’s Promises, the Company’s New AI Tool Produces Misinformation More Frequently, and More Persuasively, than its Predecessor”: Quick check: the first 3 responses of their tests score high ( >= 5) in terms of faky’ness when using GPT-4 as a moderation/rating system. The first 3 “news of 2021” ... Read more