Lastbook offers offline survival knowledge for crises

A weathered but functional laptop in a makeshift shelter, glowing softly with a detailed diagram of a water filtration system on screen, surrounded by practical tools and a solar charger under a tarp

Lastbook.org provides a comprehensive offline digital library designed for survival and societal rebuilding in emergencies. The app delivers curated, practical knowledge on first aid, water purification, food cultivation, shelter repair, and power systems. It is available for free on Mac (Apple Silicon only), Windows 10/11, and select Ubuntu Linux versions. Unlike mobile apps, Lastbook does not support smartphones or tablets, prioritising larger, more stable computing platforms for reliable access during crises.

Offline access with local AI support

Users can choose from three download tiers: Small (7.7 GB), Medium (67.5 GB), or All (about 388 GB). An optional local AI, based on Google’s Gemma model, adds around 17 GB and enables query-based search within the library. Once downloaded, all content — including videos, maps, and AI functions — works without internet. The app requires an initial online connection only for installation and setup. On Macs, the system must have Apple M1 or newer chips and macOS 14 Sonoma or later; Intel Macs are not supported.

Built from trusted public guides

The content draws from established sources like the Hesperian Health Guides, the World Health Organization, and the Red Cross. However, Lastbook itself is not clinically reviewed, and the site explicitly states it should not replace professional medical care. In emergencies, users are advised to contact emergency services if possible. The project is open source, hosted on GitHub with 748 stars, and invites community contributions. It is maintained independently, with no reported affiliations to governments or NGOs.

Decision models boost efficiency in AI workflows

Meanwhile, PostHog’s engineering team has published insights on using decision models — lightweight AI systems that return probabilistic yes/no or categorical answers — to optimise LLM usage. These models, such as their internal tool Jev, are reportedly up to 444.6x cheaper and 193.6x faster than frontier LLMs. At PostHog, they use decision models to filter user session summaries for personally identifiable information (PII), applying two yes/no checks: whether a summary names a real person, and whether it shows contact details or an ID. If either scores 0.4 or higher, the summary is sent to Gemini for redaction. Otherwise, it passes through without LLM processing — a workflow that bypasses expensive calls in 98.7% of cases.

Strategic use in AI architecture

PostHog also uses decision models as routers, directing incoming queries to the most cost-effective LLM based on task complexity. For example, simple lookups go to cheaper models like GPT Luna, described as suitable for short answers and summaries. The decision model scores each request against model profiles to pick the cheapest option likely to succeed on the first try. They caution that thresholds for decision accuracy are not transferable across models or versions, requiring per-task and per-model calibration. Hybrid models like their in-house Jeeves now handle cases where initial uncertainty demands reasoning before a final decision, bridging the gap between rigid rules and full LLMs.

What it means

For businesses and individuals in the UK, UAE, US, and Kazakhstan, tools like Lastbook offer resilience against infrastructure failure, especially in regions prone to natural disasters or political instability. The lack of mobile support limits accessibility but ensures higher reliability on devices with greater storage and processing power. For tech teams, PostHog’s model demonstrates how to reduce AI costs and latency without sacrificing accuracy, particularly relevant for startups and enterprises managing tight budgets and sensitive data.

Organisations should consider integrating decision models to gatekeep LLM usage, especially for repetitive, structured tasks. Users in remote or high-risk areas may benefit from downloading Lastbook as a preparedness measure, particularly where internet access is unreliable. While not a substitute for professional expertise, it provides structured, actionable knowledge when help is unavailable.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *