How Microsoft ML Researcher Actually Uses ChatGPT

Item: How Microsoft ML Researcher Actually Uses ChatGPT
Rating: 4
Author: PostSingularity Institute

Humanities, Social Sciences & Thought AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence

🎙 The AI Advantage 👥 480K 📅 February 14, 2024 ⏱ 23 min 👁 10K 📄 expert opinion 🧭 2026-09-08

Available in: English (current) Français

Keywords

prompt engineeringcontext windowfew-shot promptingchain-of-thoughtdebugging

Summary

In this video, the host interviews Besmira Nushi, a principal researcher at Microsoft, about her practical use of ChatGPT. The conversation covers prompt engineering basics, emphasizing the ‘instructions plus context’ framework. Nushi discusses the importance of tailoring prompts to the use case, whether for productivity, learning, or documentation browsing. They explore advanced techniques like few-shot prompting and chain-of-thought reasoning, noting the challenges of applying few-shot to code generation. The discussion also touches on custom instructions, with Nushi revealing she uses two: one for general factual responses and one for coding. For coding, they emphasize the importance of verifying libraries, using follow-up questions, and modularizing code for effective debugging. The host shares his own experiences and tips, such as feeding updated documentation to avoid outdated code. The video concludes with a recommendation to check Nushi’s LinkedIn and Microsoft for Startups resources.

140 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable practical insights from a machine learning researcher, offering a perspective that differs from typical power-user advice. The argumentation is based on personal experience and professional expertise, which lends credibility. However, the claims are not backed by empirical evidence or formal studies, and the conversational format sometimes leads to less structured reasoning. The host’s questions guide the discussion effectively, but the depth of technical explanation is limited by the target audience’s presumed familiarity with AI concepts.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is moderate: the content is anecdotal and not peer-reviewed, but the speaker’s credentials (Microsoft researcher) add authority. The sources cited are limited to LinkedIn profiles and Microsoft for Startups links, which are relevant but not primary research. The title accurately reflects the content, and the video’s structure with chapters helps navigation. The host’s commentary is clear, and the discussion stays on topic.

159 words

Title / Content Match

The title accurately reflects the content: a conversation with a Microsoft ML researcher about her practical use of ChatGPT.

Quality & Reliability

7/10

The video features a credible Microsoft ML researcher sharing practical insights, but the content is largely anecdotal and lacks rigorous scientific backing. The host's framing and the conversational format limit the depth of technical validation.

Chapters

Cited Sources

Microsoft for Startups LinkedIn — Mentioned in the video as a resource for startups.
Besmira Nushi LinkedIn — The guest's professional profile.

Concurring Sources

OpenAI Prompt Engineering Guide — General guidance on prompt engineering that aligns with the video's advice.

Dissenting Sources

On the Dangers of Stochastic Parrots — Raises concerns about LLM reliability, contrasting with the video's optimistic view.

External References

Contribution & Novelties

The video offers a unique perspective by featuring a Microsoft researcher’s daily use of ChatGPT, highlighting practical tips like using custom instructions for temperature control and verifying libraries. It bridges the gap between research and application.

Pour aller plus loin :

Prompt engineering — Overview of techniques discussed.
Chain-of-thought prompting — Research paper on the technique mentioned.
Few-shot learning — Context for the few-shot prompting discussion.

66 words

Radar Profile

The radar profile shows a balanced but moderate performance across all dimensions, with slightly higher scores in information quantity and quality, reflecting the video's practical value but limited technical depth.

Reliability 6/10