Home
Technology
Gaming
Entertainment
World & Business
Science
Sports
AI

World & Business

Home
Technology
Gaming
Entertainment
World & Business
Science
Sports
AI

ActICL reportedly beats similarity retrieval by 11.6 points · Digg

Announcement

ActICL reportedly beats similarity retrieval by 11.6 points

ActICL's developers say its agents search a demonstration bank and select in-context examples within token budgets.

2 Sources, 2h ago, first seen 2h ago

TLDR

ActICL's developers say they trained multimodal agents to select their own in-context examples. Across seven benchmarks and three backbones, they report a 4.4-point lead over the strongest trained baseline while using about 3.5 demonstrations on average. They also say a 9B ActICL agent built contexts that raised GPT-5.6-Sol by 3.5 points over standard retrieval and beat a Gemini3.5-Flash-powered workflow on all nine tested model–benchmark pairs.

Combined views

—

2 Sources, first seen 2h ago

Combined views

—

2 Sources, first seen 2h ago

— likes

Today's Rank

—

Not ranked yet

Today's Rank

—

Not ranked yet

— likes

— comments

— saves

— reposts

— comments

— saves

— reposts

2 Sources

@karrsen0713🚨Agentic Multimodal ICL: Can an MLLM learn to build its own in-context examples: what to look for, what to keep, and when it has enough? We post-train MLLM agents to do this with tailored tools. ActICL beats similarity retrieval by +11.6 points, and boosts frontier models including GPT 5.6! Key takeaways: 1️⃣ We cast ICL retrieval/context construction as an agentic task: the agent searches a demonstration bank with retrieval goals it writes itself, reads promising candidates, curates its context, and commits it, all under token budgets. 2️⃣ We train ActICL agents with rejection-sampling SFT followed by joint RL over acquisition and answering. Across 7 benchmarks and 3 backbones, ActICL beats the strongest trained baseline by +4.4 points while using only ~3.5 demonstrations on average. 3️⃣ ActICL contexts transfer: a 9B ActICL agent builds contexts that raise GPT-5.6-Sol by +3.5 points over standard retrieval, and it beats a Gemini3.5-Flash-powered agentic workflow on all 9 model–benchmark pairs we test. 👇🧵2h

@EliasEskinRT @karrsen0713: 🚨Agentic Multimodal ICL: Can an MLLM learn to build its own in-context examples: what to look for, what to keep, and when…2h

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

2 Sources

@karrsen0713🚨Agentic Multimodal ICL: Can an MLLM learn to build its own in-context examples: what to look for, what to keep, and when it has enough? We post-train MLLM agents to do this with tailored tools. ActICL beats similarity retrieval by +11.6 points, and boosts frontier models including GPT 5.6! Key takeaways: 1️⃣ We cast ICL retrieval/context construction as an agentic task: the agent searches a demonstration bank with retrieval goals it writes itself, reads promising candidates, curates its context, and commits it, all under token budgets. 2️⃣ We train ActICL agents with rejection-sampling SFT followed by joint RL over acquisition and answering. Across 7 benchmarks and 3 backbones, ActICL beats the strongest trained baseline by +4.4 points while using only ~3.5 demonstrations on average. 3️⃣ ActICL contexts transfer: a 9B ActICL agent builds contexts that raise GPT-5.6-Sol by +3.5 points over standard retrieval, and it beats a Gemini3.5-Flash-powered agentic workflow on all 9 model–benchmark pairs we test. 👇🧵2h

@EliasEskinRT @karrsen0713: 🚨Agentic Multimodal ICL: Can an MLLM learn to build its own in-context examples: what to look for, what to keep, and when…2h