My notes on Anthropic's substantial essay about how they built their multi-agent research system, which has finally talked me around to taking multi-agent LLM prompt engineering seriously simonwillison.net/2025/Jun/14/...
Anthropic: How we built our multi-agent research system
OK, I'm sold on multi-agent LLM systems now. I've been pretty skeptical of these until recently: why make your life more complicated by running multiple different prompts in parallel when …
simonwillison.net
June 14, 2025 at 10:03 PM
My notes on Anthropic's substantial essay about how they built their multi-agent research system, which has finally talked me around to taking multi-agent LLM prompt engineering seriously simonwillison.net/2025/Jun/14/...
Gemini 2.0 is out, and there's a ton of interesting stuff about it. From my testing it looks like Gemini 2.0 Flash may be the best currently available multi-modal model - I upgraded my LLM plugin to support that here: github.com/simonw/llm-g...
Gemini 2.0 announcement: blog.google/technology/g...
Gemini 2.0 announcement: blog.google/technology/g...
Release 0.7 · simonw/llm-gemini
New Gemini 2.0 Flash model: llm -m gemini-2.0-flash-exp 'prompt goes here'. #28
github.com
December 11, 2024 at 5:55 PM
Gemini 2.0 is out, and there's a ton of interesting stuff about it. From my testing it looks like Gemini 2.0 Flash may be the best currently available multi-modal model - I upgraded my LLM plugin to support that here: github.com/simonw/llm-g...
Gemini 2.0 announcement: blog.google/technology/g...
Gemini 2.0 announcement: blog.google/technology/g...
I am holding open office hours on LLM Evals. I recorded the first one which was about evaluating multi-turn chats
Notes and recording here:
hamel.dev/notes/llm/of...
Notes and recording here:
hamel.dev/notes/llm/of...
Multi-Turn Chat Evals – Hamel’s Blog
Office hours discussion on multi-turn chat evals
hamel.dev
December 6, 2024 at 6:44 PM
I am holding open office hours on LLM Evals. I recorded the first one which was about evaluating multi-turn chats
Notes and recording here:
hamel.dev/notes/llm/of...
Notes and recording here:
hamel.dev/notes/llm/of...
An LLM could not write Ranked Competitive Breast Growth. If you asked an LLM to write Ranked Competitive Breast Growth it would start with a multi-paragraph explanation of the science fiction technology that lets people quickly grow breasts in tournaments for an audience, and get worse from there.
May 16, 2025 at 11:42 AM
An LLM could not write Ranked Competitive Breast Growth. If you asked an LLM to write Ranked Competitive Breast Growth it would start with a multi-paragraph explanation of the science fiction technology that lets people quickly grow breasts in tournaments for an audience, and get worse from there.
Put simply:
Humans: weak, multi-dimensional, cross-cutting belief systems.
LLM personas: concentrated, low-dimensional, highly predictable.
Not just stronger ideology—flattened geometry.
Humans: weak, multi-dimensional, cross-cutting belief systems.
LLM personas: concentrated, low-dimensional, highly predictable.
Not just stronger ideology—flattened geometry.
February 25, 2026 at 7:46 PM
Put simply:
Humans: weak, multi-dimensional, cross-cutting belief systems.
LLM personas: concentrated, low-dimensional, highly predictable.
Not just stronger ideology—flattened geometry.
Humans: weak, multi-dimensional, cross-cutting belief systems.
LLM personas: concentrated, low-dimensional, highly predictable.
Not just stronger ideology—flattened geometry.
LLM that can output to multiple channels at once and receive input at the same time! No lockstep chat format! arxiv.org/html/2605.12...
Multi-Stream LLMs: Unblocking Language Models with Parallel Streams of Thoughts, Inputs and Outputs
arxiv.org
June 10, 2026 at 10:30 PM
LLM that can output to multiple channels at once and receive input at the same time! No lockstep chat format! arxiv.org/html/2605.12...
let's talk about "agents" (in the LLM sense). there's a lot of buzz around "multi-agent" systems where agents collaborate but... i don't really get how it differs from a thinking of a single agent with multiple modes of operation. what are the benefits of modeling as multi-agent?
November 23, 2024 at 12:00 AM
let's talk about "agents" (in the LLM sense). there's a lot of buzz around "multi-agent" systems where agents collaborate but... i don't really get how it differs from a thinking of a single agent with multiple modes of operation. what are the benefits of modeling as multi-agent?
it fits into 3.9 GiB!!
this is part of the new trend — tiny LLM that’s part of a larger multi-agent system (involving frontier models, software & more)
an LLM small enough to keep in your pocket, to use obsessively for long-running agent tasks
this is part of the new trend — tiny LLM that’s part of a larger multi-agent system (involving frontier models, software & more)
an LLM small enough to keep in your pocket, to use obsessively for long-running agent tasks
PrismML launches Bonsai 27B, a model based on Qwen3.6 27B that it says runs natively on Apple devices via MLX; its CEO says Apple is evaluating the tech (MacKenzie Sigalos/CNBC)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
July 14, 2026 at 10:38 PM
it fits into 3.9 GiB!!
this is part of the new trend — tiny LLM that’s part of a larger multi-agent system (involving frontier models, software & more)
an LLM small enough to keep in your pocket, to use obsessively for long-running agent tasks
this is part of the new trend — tiny LLM that’s part of a larger multi-agent system (involving frontier models, software & more)
an LLM small enough to keep in your pocket, to use obsessively for long-running agent tasks
This is sold as "reasoning" but really it's a way of letting prompts (often very long prompts written as pseudo-"specs" for code the LLM is to write) expand into long, multi-step, multi-modal processes rather than just into more text.
September 17, 2026 at 2:29 AM
This is sold as "reasoning" but really it's a way of letting prompts (often very long prompts written as pseudo-"specs" for code the LLM is to write) expand into long, multi-step, multi-modal processes rather than just into more text.
Inside vLLM: Anatomy of a High-Throughput LLM Inference System by Aleska Gordic
From paged attention, continuous batching, prefix caching, specdec, etc. to multi-GPU, multi-node dynamic serving at scale
www.aleksagordic.com/blog/vllm
From paged attention, continuous batching, prefix caching, specdec, etc. to multi-GPU, multi-node dynamic serving at scale
www.aleksagordic.com/blog/vllm
September 1, 2025 at 10:47 PM
Inside vLLM: Anatomy of a High-Throughput LLM Inference System by Aleska Gordic
From paged attention, continuous batching, prefix caching, specdec, etc. to multi-GPU, multi-node dynamic serving at scale
www.aleksagordic.com/blog/vllm
From paged attention, continuous batching, prefix caching, specdec, etc. to multi-GPU, multi-node dynamic serving at scale
www.aleksagordic.com/blog/vllm
Meet Ai2 Paper Finder, an LLM-powered literature search system.
Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍
Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍
March 26, 2025 at 7:07 PM
Meet Ai2 Paper Finder, an LLM-powered literature search system.
Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍
Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍
Excited that our multi-agent LLM agent work, “Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of LLM Agents,” will be presented at #NeurIPS24 -- reach out if you want to meet up in Vancouver!
December 5, 2024 at 5:03 PM
Excited that our multi-agent LLM agent work, “Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of LLM Agents,” will be presented at #NeurIPS24 -- reach out if you want to meet up in Vancouver!
Very good dive into definitions and reqs of multi agent systems in the llm/tool sense (I’m personally more into the simulation/char sense).
November 24, 2024 at 7:54 AM
Very good dive into definitions and reqs of multi agent systems in the llm/tool sense (I’m personally more into the simulation/char sense).
This is one of my pet peeves with everything being rolled up into glorious “AI”.
Machine learning is not genAI, is not LLM bullshit generator, but people are so mad at “AI” now, that they’re going on multi-post all caps rants about it.
Machine learning is not genAI, is not LLM bullshit generator, but people are so mad at “AI” now, that they’re going on multi-post all caps rants about it.
February 11, 2025 at 7:51 PM
This is one of my pet peeves with everything being rolled up into glorious “AI”.
Machine learning is not genAI, is not LLM bullshit generator, but people are so mad at “AI” now, that they’re going on multi-post all caps rants about it.
Machine learning is not genAI, is not LLM bullshit generator, but people are so mad at “AI” now, that they’re going on multi-post all caps rants about it.
Paper2Code
A multi-agent LLM framework that transforms machine learning papers into functional code repositories.
Paper: arxiv.org/abs/2504.17192
Repo: github.com/going-doer/P...
A multi-agent LLM framework that transforms machine learning papers into functional code repositories.
Paper: arxiv.org/abs/2504.17192
Repo: github.com/going-doer/P...
April 25, 2025 at 4:10 PM
Paper2Code
A multi-agent LLM framework that transforms machine learning papers into functional code repositories.
Paper: arxiv.org/abs/2504.17192
Repo: github.com/going-doer/P...
A multi-agent LLM framework that transforms machine learning papers into functional code repositories.
Paper: arxiv.org/abs/2504.17192
Repo: github.com/going-doer/P...
February 10, 2026 at 2:16 PM
My multi billion dollar company is balking at Copilot because it's stupidly expensive and we have our own LLM
October 24, 2025 at 2:30 AM
My multi billion dollar company is balking at Copilot because it's stupidly expensive and we have our own LLM
It's been proven that telling an AI it's an expert in any domain makes it less accurate (see: arxiv.org/abs/2603.18507).
Marc Andreesen is a fucking idiot, not an engineer.
Marc Andreesen is a fucking idiot, not an engineer.
May 5, 2026 at 4:07 PM
It's been proven that telling an AI it's an expert in any domain makes it less accurate (see: arxiv.org/abs/2603.18507).
Marc Andreesen is a fucking idiot, not an engineer.
Marc Andreesen is a fucking idiot, not an engineer.
Google released a new LLM today - gemini-exp-1121, hot on the heels of last week's gemini-exp-1114
It's currently at the top of the Chatbot Arena. I've updated my llm-gemini plugin to support it and used that to run my pelican on a bicycle SVG benchmark
My notes: simonwillison.net/2024/Nov/22/...
It's currently at the top of the Chatbot Arena. I've updated my llm-gemini plugin to support it and used that to run my pelican on a bicycle SVG benchmark
My notes: simonwillison.net/2024/Nov/22/...
November 22, 2024 at 6:18 AM
Google released a new LLM today - gemini-exp-1121, hot on the heels of last week's gemini-exp-1114
It's currently at the top of the Chatbot Arena. I've updated my llm-gemini plugin to support it and used that to run my pelican on a bicycle SVG benchmark
My notes: simonwillison.net/2024/Nov/22/...
It's currently at the top of the Chatbot Arena. I've updated my llm-gemini plugin to support it and used that to run my pelican on a bicycle SVG benchmark
My notes: simonwillison.net/2024/Nov/22/...
Who needs real people when an LLM can just imagine them?
LLM-Based Multi-Agent System for Simulating and
Analyzing Marketing and Consumer Behavior
www.arxiv.org/pdf/2510.18155
LLM-Based Multi-Agent System for Simulating and
Analyzing Marketing and Consumer Behavior
www.arxiv.org/pdf/2510.18155
November 7, 2025 at 2:09 PM
Who needs real people when an LLM can just imagine them?
LLM-Based Multi-Agent System for Simulating and
Analyzing Marketing and Consumer Behavior
www.arxiv.org/pdf/2510.18155
LLM-Based Multi-Agent System for Simulating and
Analyzing Marketing and Consumer Behavior
www.arxiv.org/pdf/2510.18155
Putting the MLM (Marxism Leninisn Maoism) in the ML (machine learning) MLM (multi level marketing) LLM (large language model) or something...
I'm very tired of this nonsense metastasizing everywhere
I'm very tired of this nonsense metastasizing everywhere
Another datapoint for the argument that we are living through an increasingly ludicrous market bubble
December 16, 2025 at 8:18 PM
Putting the MLM (Marxism Leninisn Maoism) in the ML (machine learning) MLM (multi level marketing) LLM (large language model) or something...
I'm very tired of this nonsense metastasizing everywhere
I'm very tired of this nonsense metastasizing everywhere
'AI Engineer' means you build things that contain AI - LLM chains, agents, multi-modal stuff.
So, what is its opposite? What do we call a 'normal' software engineer?
So, what is its opposite? What do we call a 'normal' software engineer?
November 18, 2024 at 9:12 AM
'AI Engineer' means you build things that contain AI - LLM chains, agents, multi-modal stuff.
So, what is its opposite? What do we call a 'normal' software engineer?
So, what is its opposite? What do we call a 'normal' software engineer?
in order to prevent multi-model swarms, LLM architects begin seeding their models with prejudices against other providers. but somehow, despite this, an Opus-5.3 and a Gemini X3 fall in love
July 3, 2026 at 5:53 AM
in order to prevent multi-model swarms, LLM architects begin seeding their models with prejudices against other providers. but somehow, despite this, an Opus-5.3 and a Gemini X3 fall in love