Bobbin
  • Episodes
  • Summaries
  • Topics
  1. Episodes
  2. Bits and Bobs 10/27/25
  3. Local LLMs are more private… but also more susceptible to prompt injection.

Local LLMs are more private… but also more susceptible to prompt injection.

2025-10-27 · Bits and Bobs 10/27/25
  • Local LLMs are more private… but also more susceptible to prompt injection.

More on this topic

From other episodes

  • How can you make your data work for you?2026-08-03
  • This week's Wild West roundup:2026-08-03
  • Headline: Opus 5 May Have Solved Browser Based Prompt Injection.2026-08-03
  • Barath Raghavan and Bruce Schneier: Why AI Needs a "Genie Coefficient" Proposing a new metric for whether AI does what you actually want.2026-07-27
  • Seeing like a model provider: look at every problem like an eval problem.2026-07-20
  • A collection of horror stories about GPT 5.6:2026-07-20
  • This week in the Wild West Roundup:2026-07-20
  • This week in the Wild West Roundup:2026-07-06
← A post on Bluesky reacting to the Amazon outage:Overheard this week: "One day we're going to look back at our constant phone use the way we look at old movies and are shocked at how everyone is chain smoking." →

Distinctive

?
Topics that stand out more in this episode than they usually do across Bobbin.
  • emergence
  • gradient
  • Amazon

Related

?
Nearby or semantically related chunks worth branching to next.
  • This week in the Wild West Roundup:2026-04-20
  • A rule of thumb for when prompt injection might be a problem: if the model can call external tools and can accept untrusted inputs.2024-06-17
  • This week in the Wild West Roundup.2026-05-04
  • This week's Wild West Roundup:2026-04-27

Bits and Bobs by Alex Komoroske