Olatomiwa Bifarin
@bifarinthefifth.bsky.social
essayist @ http://youtube.com/@epistemeEngine
AI blog @ http://theepsilon.substack.com
μgrants @ https://mbbgrant.substack.com
AI blog @ http://theepsilon.substack.com
μgrants @ https://mbbgrant.substack.com
The authors of this article asked "Will AI become the de-facto lab Co-PI?"
The answer is yes, and went on to show possible paths, with relevant caveats.
www.nature.com/articles/s4...
The answer is yes, and went on to show possible paths, with relevant caveats.
www.nature.com/articles/s4...
Will AI become our Co-PI?
npj Digital Medicine - Will AI become our Co-PI?
www.nature.com
July 19, 2025 at 2:10 PM
The authors of this article asked "Will AI become the de-facto lab Co-PI?"
The answer is yes, and went on to show possible paths, with relevant caveats.
www.nature.com/articles/s4...
The answer is yes, and went on to show possible paths, with relevant caveats.
www.nature.com/articles/s4...
The results were a mix of hilarious & costly mistakes, from tungsten cubes orders to offering wild discounts, showing the current limits of AI's real-world common sense.
But the experiment also suggests that with bespoke agent engineering, autonomous AI agents aren't far off.
But the experiment also suggests that with bespoke agent engineering, autonomous AI agents aren't far off.
July 10, 2025 at 6:53 PM
The results were a mix of hilarious & costly mistakes, from tungsten cubes orders to offering wild discounts, showing the current limits of AI's real-world common sense.
But the experiment also suggests that with bespoke agent engineering, autonomous AI agents aren't far off.
But the experiment also suggests that with bespoke agent engineering, autonomous AI agents aren't far off.
It's a fairly detailed piece articulating the power of leveraging collaborative subagents and the challenges of prompt engineering and evaluation.
The first sentence of the conclusion: "When building AI agents, the last mile often becomes most of the journey."
The first sentence of the conclusion: "When building AI agents, the last mile often becomes most of the journey."
July 5, 2025 at 9:15 PM
It's a fairly detailed piece articulating the power of leveraging collaborative subagents and the challenges of prompt engineering and evaluation.
The first sentence of the conclusion: "When building AI agents, the last mile often becomes most of the journey."
The first sentence of the conclusion: "When building AI agents, the last mile often becomes most of the journey."
Essentially the framework merges two foundational things: 1) "Anthropic's" agent architectures (qua parallel, evaluator-optimizer, orchestrator, etc) 2) MCP native (It's opinionated about the world being MCP native). It's also super lightweight.
June 24, 2025 at 9:28 PM
Essentially the framework merges two foundational things: 1) "Anthropic's" agent architectures (qua parallel, evaluator-optimizer, orchestrator, etc) 2) MCP native (It's opinionated about the world being MCP native). It's also super lightweight.
A key challenge is "superposition," where language models compress numerous concepts into a limited space.
researchers often use sparse autoencoders/transcoders to isolate and understand individual concepts, or "features," providing a clearer picture of how models represent and connect ideas.
researchers often use sparse autoencoders/transcoders to isolate and understand individual concepts, or "features," providing a clearer picture of how models represent and connect ideas.
June 21, 2025 at 10:42 PM
A key challenge is "superposition," where language models compress numerous concepts into a limited space.
researchers often use sparse autoencoders/transcoders to isolate and understand individual concepts, or "features," providing a clearer picture of how models represent and connect ideas.
researchers often use sparse autoencoders/transcoders to isolate and understand individual concepts, or "features," providing a clearer picture of how models represent and connect ideas.