What can we expect of LLMs as software engineers?
summary
Shows the strawberry gate error to illustrate context failures, explains base model vs fine‑tuning and supporting models (compilers, calculators, RAG), cites the 26% productivity claim and the lost in the middle finding about long contexts, demonstrates when tooling actually runs code versus when it merely generates plausible text, and argues that LLMs help small scoped tasks but don’t replace judgment — investigative, evaluative, and innovation skills remain necessary.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:25How LLMs work
- 10:01Context vs meaning
- 15:38Base models vs GenAI
- 21:34LLM productivity research
- 25:52Key takeaways
- 28:19AI engineer skills
- 31:53Q&A
related talks
55mAre LLMs good software engineers?
01h 01mLike Having an Intern? The Impact of LLMs on Software Engineering Careers
45mRewriting the Role: Developers in the Age of LLMs
38mNobody Pages the LLM: Engineering Rigour for Vibe Coding
35mEvaluating LLM systems: the nuts and bolts
24mWill an LLM replace me? A Systematic study of Gen AI's impact on data jobs
01h 49mAI engineering with open access LLMs that lie, curse, and steal
16mLLMs are not the product