AI Disruption: DeepSeek and Cerebras

DeepSeek was a disruptive surprise at the start of 2025--an open weights model trained at a fraction of the cost of previous models. Bryan and Adam were joined by Andy Hock and James Wang from Cerebras, whose wafer-scale silicon executes these models faster than is possible with any number of GPUs.
In addition to Bryan Cantrill (bsky.app/profile/bcantrill.bsky.social) and Adam Leventhal (bsky.app/profile/ahl.bsky.social), we were joined by Andy Hock (github.com/oxidecomputer/oxide-and-friends/blo…), and James Wang (github.com/oxidecomputer/oxide-and-friends/blo…), both of Cerebras.
Some of the topics we hit on, in the order that we hit them:
• interactive inference with Cerebras (inference.cerebras.ai)
• 100x Defect Tolerance: How Cerebras Solved the Yield Problem (cerebras.ai/blog/100x-defect-tolerance-how-cer…)
• Tweet from Eric Meijer (x.com/headinthebox/status/1883940072623595840)
• Ouroborus (en.wikipedia.org/wiki/Ouroboros)
• Quine Relay (github.com/mame/quine-relay)
• Simon Willison’s Weblog when DeepSeek fell from space (simonwillison.net/2024/Dec/25/deepseek-v3)
• Tweet from Naveen Rao (x.com/NaveenGRao/status/1886544584588619840)
BONUS
• MST3K archive (gizmoplex.com/browse)
If we got something wrong or missed something, please file a PR! Our next show will likely be on Monday at 5p Pacific Time on our Discord server; stay tuned to our Mastodon feeds for details, or subscribe to this calendar (calendar.google.com/calendar/ical/c_318925f418…). We'd love to have you join us, as we always love to hear from new speakers!