research

New approach to agent reliability, AgentSpec, forces agents to follow rules nuneybits_Abstract_art_of_robot_scientist_vs_robot_businessman_e9e69d34-ef1e-4ca9-94c3-091e3376e0c2.webp?w=578

nuneybits_Abstract_art_of_robot_scientist_vs_robot_businessman_e9e69d34-ef1e-4ca9-94c3-091e3376e0c2.webp?w=578

New approach to agent reliability, AgentSpec, forces agents to follow rules

admin March 28, 2025

Researchers from Singapore Management University developed a new domain-specific language for agents to remain reliable.Read More

Researchers warn of ‘catastrophic overtraining’ in Large Language Models cfr0z3n_an_overworeked_humanoid_robot_emits_visible_sweat_drops_ec3ba283-683f-4c5b-94c2-fc1b228577c4.png?w=578

cfr0z3n_an_overworeked_humanoid_robot_emits_visible_sweat_drops_ec3ba283-683f-4c5b-94c2-fc1b228577c4.png?w=578

Researchers warn of ‘catastrophic overtraining’ in Large Language Models

admin March 28, 2025

The researchers compared two versions of OLMo-1b: one pre-trained on 2.3 trillion tokens and another on 3...

METASCALE improves LLM reasoning with adaptive strategies an-ai-thinking-deeply-d1QQ83YATmWIi27u1BpWAQ-n6j3Yxn7TMuUfTDndIEbUA-transformed.jpeg?w=578

an-ai-thinking-deeply-d1QQ83YATmWIi27u1BpWAQ-n6j3Yxn7TMuUfTDndIEbUA-transformed.jpeg?w=578

METASCALE improves LLM reasoning with adaptive strategies

admin March 25, 2025

METASCALE uses a three-stage approach to dynamically choose the right reasoning technique for each promblem.Read More

Less is more: UC Berkeley and Google unlock LLM potential through simple sampling LLM-self-verification.jpeg?w=578

Less is more: UC Berkeley and Google unlock LLM potential through simple sampling

admin March 21, 2025

With multiple sampling and self-verification, Gemini 1.5 Pro can outperform o1-preview in reasoning tasks.Read More

Beyond RAG: SEARCH-R1 integrates search engines directly into reasoning models cfr0z3n_midcentury_modernist_retrofuturistic_style_det_c17aaff2-438d-490d-a7b3-dfb6b4d7c24d-1.png?w=578

cfr0z3n_midcentury_modernist_retrofuturistic_style_det_c17aaff2-438d-490d-a7b3-dfb6b4d7c24d-1.png?w=578

Beyond RAG: SEARCH-R1 integrates search engines directly into reasoning models

admin March 19, 2025

SEARCH-R1 trains LLMs to gradually think and conduct online search as they generate answers for reasoning problems.Read...

Chain-of-experts (CoE): A lower-cost LLM framework that increases efficiency and accuracy mixture-of-millions-of-experts.jpg?w=578

Chain-of-experts (CoE): A lower-cost LLM framework that increases efficiency and accuracy

admin March 10, 2025

Chain-of-experts chains LLM experts in a sequence, outperforming mixture-of-experts (MoE) with lower memory and compute costs.Read More

How the A-MEM framework supports powerful long-context memory so LLMs can take on more complicated tasks 0fDPi8NORiSvt1W7X6U3nQ.webp?w=578

How the A-MEM framework supports powerful long-context memory so LLMs can take on more complicated tasks

admin March 6, 2025

A-MEM uses embeddings and LLMs to create dynamic memory notes that automatically link to create complex knowledge...

Cohere’s first vision model Aya Vision is here with broad, multilingual understanding and open weights — but there’s a catch man-typing-world.png?w=578

Cohere’s first vision model Aya Vision is here with broad, multilingual understanding and open weights — but there’s a catch

admin March 4, 2025

Aya Vision 8B and 32B demonstrate best-in-class performance relative to their parameter size, outperforming much larger models.Read...

How test-time scaling unlocks hidden reasoning abilities in small language models (and allows them to outperform LLMs) DALL·E-2024-09-12-17.04.17-A-detailed-image-of-a-robotic-version-of-The-Thinker-sculpture-sitting-in-the-classic-pose-with-one-hand-resting-on-its-chin-and-deep-in-thought.-T-Large.jpeg?w=578

DALL·E-2024-09-12-17.04.17-A-detailed-image-of-a-robotic-version-of-The-Thinker-sculpture-sitting-in-the-classic-pose-with-one-hand-resting-on-its-chin-and-deep-in-thought.-T-Large.jpeg?w=578

How test-time scaling unlocks hidden reasoning abilities in small language models (and allows them to outperform LLMs)

admin February 21, 2025

A 1B small language model can beat a 405B large language model in reasoning tasks if provided...

Less supervision, better results: Study shows AI models generalize more effectively on their own 43OM9mruR_eWSFVQg2jVZA.jpeg?w=578

Less supervision, better results: Study shows AI models generalize more effectively on their own

admin February 12, 2025

Training LLMs and VLMs through reinforcement learning delivers better results than using hand-crafted examples.Read More

New approach to agent reliability, AgentSpec, forces agents to follow rules

Researchers warn of ‘catastrophic overtraining’ in Large Language Models

METASCALE improves LLM reasoning with adaptive strategies

Less is more: UC Berkeley and Google unlock LLM potential through simple sampling

Beyond RAG: SEARCH-R1 integrates search engines directly into reasoning models

Chain-of-experts (CoE): A lower-cost LLM framework that increases efficiency and accuracy

How the A-MEM framework supports powerful long-context memory so LLMs can take on more complicated tasks

Cohere’s first vision model Aya Vision is here with broad, multilingual understanding and open weights — but there’s a catch

How test-time scaling unlocks hidden reasoning abilities in small language models (and allows them to outperform LLMs)

Less supervision, better results: Study shows AI models generalize more effectively on their own

You may have missed

Nikolaj Coster-Waldau Would Prefer If You Moved on From Hating the Ending of ‘Game of Thrones’

Elon Musk’s xAI Sues OpenAI and Apple Over App Store Discrimination Claims

In a First, a Human Breathed Using an Implanted Pig Lung

Intrusive Thought of the Day: Is That YouTube Video Enhanced With AI?