Dan Biderman

Person

Mentions

Episodes

Claims

Dan Biderman, mentioned 14 times across podcast episodes and expert conversations analyzed by Sonic.

What Dan Biderman has said

Dan Biderman previously believed that vision and action were the keys to AI progress, but the 'ChatGPT moment' and his work at MosaicML shifted his perspective to the effectiveness of language models.

neutral·Dan Biderman

Engram aims to use offline computation to compress large KV caches, potentially reducing an 80-gigabyte cache by a factor of 1,000x.

bullish·Dan Biderman

The entire set of weights for a 70-billion-parameter Llama model occupies approximately 100 gigabytes.

neutral·Dan Biderman

Expert perspectiveDan BidermanJun 24

Engram aims to use offline computation to compress large KV caches, potentially reducing an 80-gigabyte cache by a factor of 1,000x.

Expert perspectiveDan BidermanJun 24

The entire set of weights for a 70-billion-parameter Llama model occupies approximately 100 gigabytes.

Expert perspectiveDan BidermanJun 24

Engram's continual learning approach can reduce token inference consumption by up to two orders of magnitude (100x) by internalizing context that would otherwise require large system prompts.

Expert perspectiveDan BidermanJun 24

Engram's technology requires white-box access to model weights, making it most easily applicable to open-source models.

Expert perspectiveDan BidermanJun 24

Engram considers individual computers and phones to be future targets for its continual learning technology.

Expert perspectiveDan BidermanJun 24

Demis Hassabis stated at a Sequoia event approximately one month prior to this interview that new breakthroughs are needed around the topics of AI memory and continual learning.

Expert perspectiveDan BidermanJun 24

The KV cache for a single Wikipedia article can require 80 gigabytes of HBM memory on a GPU.

Expert perspectiveDan BidermanJun 24

Dan Biderman's vision is for Engram to become the primary LLM interface to the data plane for all users, drawing a parallel to the roles of companies like Databricks and Oracle.

Expert perspectiveDan BidermanJun 24

Dan Biderman argues that if an AI lab like OpenAI needed to win a math Olympiad within a week, the superior strategy would be to synthesize training data and launch a new training job rather than rely...

Expert perspectiveDan BidermanJun 24

Some of the best large language models from China incorporate layers inspired by state-space architectures, which allows them to operate with sub-quadratic cost.

Expert perspectiveDan BidermanJun 24

Dan Biderman identifies the launches of GitHub Copilot and ChatGPT as the main surprising events in AI for him.

Expert perspectiveDan BidermanJun 24

A model trained with Engram's technology can answer certain queries in 100 tokens, whereas a frontier model might consume 100,000 tokens for the same task.

Expert perspectiveDan BidermanJun 24

Engram aims to create hundreds of millions of personalized models that function as a 'brain state' representation of a file system, which is more associative and efficient than a literal representatio...

Expert perspectiveDan BidermanJun 24

Create a free account to see Dan Biderman's full intelligence report — every claim, the relationship network, and AI Q&A across all sources. No card needed.

Get started free

Back to Entities Entity Detail