AI News · The field notes
The Daily AI Drop - Morning Edition, October 28, 2026
Alex and Jordan cover running Liquid AI's LFM3 locally, fixing the context window problem, DeepSeek's big benchmark win, and how Legora is transforming legal work with AI.
Alex and Jordan cover running Liquid AI's LFM3 locally, fixing the context window problem, DeepSeek's big benchmark win, and how Legora is transforming legal work with AI.
Listen: Listen to this episode
Run LFM3 1.7B Locally: Model for Specialized Applications — Fahd Mirza
Fahd installs and tests Liquid AI's LFM3-1.7B, a small text model built for local agents and specialized applications. It matters because a capable small model means speed, privacy, and low cost without sending data to the cloud — a practical entry point if you have been curious about running models on your own machine.
The Context Window Problem Nobody Talks About — Aish Reganti
Aish breaks down why long AI conversations quietly lose track of earlier instructions and decisions, and how context compaction preserves what matters. Techniques like sliding windows help an assistant hold onto critical decisions over long sessions instead of only remembering the last few messages — a mental model worth having the next time you design a prompt or build an agent.
Deepseek crushed everyone — Matthew Berman
Matthew digs into a result that puts DeepSeek ahead of the field, resetting expectations about who is leading. It matters beyond one leaderboard: a decisive win like this tends to move developers, pricing, and product roadmaps, and it is a good prompt to pick models on current results rather than reputation.
Legora legal AI series — OpenAI (four videos)
- How Legora is transforming legal work with OpenAI: Watch source video ↗
- Faster contract review. Broader legal analysis. | Legora's Alex Fortescue-Webb: Watch source video ↗
- Building the future of legal work with AI | Legora CEO Max Junestrand: Watch source video ↗
- From legal bottleneck to business partner | Legora's Emma Kadri Bergström: Watch source video ↗
OpenAI's four-part Legora series shows a working legal platform using OpenAI models, ChatGPT, and Codex. General counsel Emma Kadri Bergström says NDA reviews once took around 20% of her team's capacity and now take under 1%, with lawyers staying responsible for decisions. Global head of legal engineering Alex Fortescue-Webb explains how teams can analyze thousands of contracts across multiple questions in minutes, expanding due diligence, while CEO Max Junestrand frames the ambition of making legal expertise available wherever it is needed. Together, the series is a concrete example of AI returning expert time for higher-value work rather than just demoing a feature.