Nils Durner's Blog Ahas, Breadcrumbs, Coding Epiphanies

Anthropic's Claude 2: Initial Impressions

Anthropic have released Claude 2, touting “two times better at giving harmless answers”. Initial reactions on the web seem favorable, particularly regarding the support of larger documents that ChatGPT. What I have seen so far with the regular end-user chat UI: A book of 1027 pages didn’t load: “Text extraction failed”. Loading a 62-pages le... Read more

Non-Uniform Performance in In-Context Learning

Non-uniform performance reported for in-context learning: How Language Models Use Long Contexts: Model performance is highest when relevant information occurs at the beginning or end of its input context Model performance substantially decreases as input contexts grow longer Extended-context models are not necessarily better at using i... Read more

GPT-4 General Availability, Deprecation

GPT-4 API was just made generally available, and: We are working on safely enabling fine-tuning for GPT-4 and GPT-3.5 Turbo and expect this feature to be available later this year. Source Also: Developers wishing to continue using their fine-tuned models beyond January 4, 2024 will need to fine-tune replacements atop the new base GPT-... Read more

SAP invests in Aleph Alpha

Prompted by a colleague posting the Handelblatt article about SAP investing in Aleph Alpha, I elaborated that the beauty of Aleph Alpha is GDPR-compliance and generally good data hygiene. With OpenAI, for me it’s oftentimes a “write code to do X”, with X actually being done on data that sits with me locally. But if X requires advanced natural la... Read more

Aleph Alpha

First glance at & hands-on Aleph Alpha, the German competitor of OpenAI: Their stated strong points: Explainable & trustworthy AI Deployment On-premise / any cloud infrastructures / their own datacenter in Germany so this is the closest thing I can currently think of regarding actually usable “GPT-on-premise” G... Read more