Out-of-Domain Finetuning to Bootstrap Hallucination Detection
How to use open-source, permissive-use data and collect less labeled samples for our tasks.
324 articles · Fresh posts aggregated from independent blogs across the directory.
How to use open-source, permissive-use data and collect less labeled samples for our tasks.
The use of large language models in the real world has strongly accelerated by the launch of ChatGPT. We (including my team at OpenAI, shoutout to them) have invested a lot of effort to build default safe behavior into the model during the alignment process (…
2022 年 10 月 24 日,SegmentFault 思否宣布开源旗下问答平台软件 Answer 1。今日,恰逢开源一周年之际,Answer 社区发布公告2 宣布项目已顺利通过投票,以全票通过的优秀表现正式进入 Apache 软件基金会(ASF)孵化器。 作为 Answer 背后的创始团队和中国新一代技术问答社区,SegmentFault 思否一直积极参与开源布道、拥抱开源生态,曾发布多篇深
The biggest deployment challenges, backward compatibility, multi-modality, and SF work ethic.
Evals, retrieval-augmented generation, guardrails, and collecting feedback; all that good stuff.
Reference, context, and preference-based metrics, self-consistency, and catching hallucinations.
Distinguishing problems with external vs. internal LLMs, and data vs non-data patterns
Evals, RAG, fine-tuning, caching, guardrails, defensive UX, and collecting user feedback.

Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written c…
Writing drafts via retrieval-augmented generation. Also reflecting on the week's journal entries.
What's the big deal, intuition on query-key-value vectors, multiple heads, multiple layers, and more.

Here are eight observations I’ve shared recently on the Cohere blog and videos that go over them.: Article: What’s the big deal with Generative AI? Is it the future or the present? Article: AI is Eating The World
It started with a question that had no clear answer, and led to eight PRs from the community.
Should chat be the main UX for LLMs? I don't think so and believe we can do better.
Prompt Engineering, also known as In-Context Prompting, refers to methods for how to communicate with LLM to steer its behavior for desired outcomes without updating the model weights. It is an empirical science and the effect of prompt engineering methods ca…
Many new Transformer architecture improvements have been proposed since my last post on “The Transformer Family” about three years ago. Here I did a big refactoring and enrichment of that 2020 post — restructure the hierarchy of sections and improve many sect…
去年没写年终总结,今年发生了很多事,觉得应该抽点时间,回顾记录,以重新出发。 仍旧是过的很快的一年,好像工作后的每一年都过得很快,书上说人之所以觉得成年后的日子过得快,是因为对很多事没有新鲜感了,没接触新的东西,没有记忆点,日复一日,所以很快。我觉得,对了一半。另一半也许因为年岁渐长,人们喜欢追忆,过去的时间也许也并没有比现在过得更慢,只是过去的日子被人们…
[Updated on 2023-01-24: add a small section on Distillation.] Large transformer models are mainstream nowadays, creating SoTA results for a variety of tasks. They are powerful but very expensive to train and use. The extremely high inference cost, in both tim…