So long, and thanks for all the fish

Summary of So long, and thanks for all the fish

by Ben Jaffe and Katie Malone

35m•July 26, 2020
Listen to Episode

Summary — "So long, and thanks for all the fish"

Hosts: Ben Jaffe and Katie Malone
Episode type: Retrospective / farewell
Context: Recorded July 13 (released a few weeks later). The podcast began October 7, 2014 and ran ~5.5 years.

Overview

This episode is a farewell/retrospective. After roughly five and a half years of producing the Linear Digressions podcast, Ben and Katie announce they’re pausing (or ending) regular production. Instead of a technical data‑science deep dive, they reflect on the podcast’s history, process, audience impact, why they’re stepping away, lessons learned, and what they’ll do next.

Key points & main takeaways

  • Longevity & scale
    • The show ran ~5.5 years, ~293 recorded episodes (including reruns); ~280+ unique episodes.
    • Ben and Katie are surprised and grateful for the audience size and engagement.
  • Why they’re stopping
    • Natural endpoint: weekly production is time-consuming and can exhaust topic ideas.
    • Field evolution: data science has shifted toward production, scale, governance, and responsibility — areas they feel less excited to cover on a weekly basis.
    • Desire to make space for other projects and deeper pursuits.
  • Podcasting process & dynamics
    • Ben edits episodes (light touch). They moved from in-person recording to remote recordings with separate mics synced in post.
    • Reruns provided breathing room and played a practical role in sustainability.
    • The co‑host dynamic (expert + intelligent non‑expert) was a key success factor—Katie’s non-expert perspective made technical topics accessible.
  • Challenges encountered
    • Sustaining fresh, interesting weekly content.
    • Translating and simplifying technical papers accurately.
    • Practical nuisances: background noise, pets, occasional accidental uploads.
  • Benefits & impact
    • Forced continual learning: the podcast pushed them to learn new topics regularly.
    • Built a “fuzzy” but useful mental model of machine learning and connected topics.
    • Positive listener feedback was humbling; emails/reviews demonstrated real-life influence.
  • Future of the show & contact
    • Website, feed and email addresses will remain available for some time; occasional “drop-ins” may happen.
    • They invite listeners to reach out and thank the audience for support.

Topics discussed in the episode (and notable past topics referenced)

  • Reflections on podcast evolution and production
  • Why they are ending/pausing the show
  • The host dynamic (expert vs. non-expert co-host)
  • Podcast technical workflow: remote recording, editing, reruns
  • Challenges: content fatigue, maintaining accuracy, background noise
  • Impact on hosts: learning, habit formation, audience interactions
  • Example subjects the podcast covered over the years:
    • Supervised learning, neural networks, Markov chains
    • GPT-2 and natural language models
    • Medical ML (“curing cancer with ML is super hard”)
    • Gravitational waves
    • The “detector” project episode (fan favorite)
  • Topics they wished had more of:
    • Productionalization, large-scale ML systems, monitoring, organizational/process best practices
    • Broader speculative topics: generalized AI, singularity / superintelligence (suggested by hosts as an interesting separate podcast)

Notable quotes & insights

  • “We’re gonna hang up our microphones… and kind of call it a day with this podcast.”
  • “Podcasting is a strength and a weakness of the medium… it’s the least intimidating thing you can imagine.”
  • “Every podcast is like a time capsule.”
  • “The podcast forced me to learn something new—something I probably wouldn’t have learned otherwise.”
  • On co-hosting value: having a curious non‑expert co‑host made technical conversations more accessible and kept the show going.
  • On sustainability: reruns were a “game changer” that allowed them to take breaks.

Action items & recommendations

For listeners

  • If you’ve been listening since the beginning (Oct 7, 2014), they asked to hear from you (tweet or email).
  • Feed, website, and email addresses will remain active—feel free to reach out with memories, corrections, or feedback.

For podcast creators / prospective shows

  • Consider a dual-host format pairing an expert and a smart non-expert to improve accessibility.
  • Use reruns strategically to reduce burnout and preserve quality.
  • Expect and plan for editorial time (editing is time-consuming but worthwhile).
  • If interested in production/scale ML topics or AGI, those are gaps the hosts highlighted as fertile ground for new shows.

Hosts’ next steps (personal plans)

  • Katie: catch up on reading, do more coding, possibly contribute to open source.
  • Ben: learning cello; open to occasional episodes/drops in the future.
  • Both: keep listening channels open and appreciate continued listener contact.

Contact info (kept active)

  • ben at lineardigressions.com
  • katie at lineardigressions.com
  • Twitter: @linerdigressions (mentioned as lynn digressions in transcript—verify when contacting)

Bottom line

This episode is a gracious, reflective sign-off: Ben and Katie celebrate what the podcast accomplished (listener impact, consistent learning, accessible ML explanations), explain practical and motivational reasons for stopping regular production, and leave the doors open for future contact and occasional content drops. The series stands as a long-running, accessible introduction to many data‑science and machine‑learning topics, and its archive remains available as a persistent resource.