To fix this, we're separating the planning and execution into two steps: – plan what to do
– execute on each step This usually leads to more calls to the language model… but helps it stay on track for more complex tasks
LLMS
-
Two-step planning improves language model task performance
By
–
-
BabyAGI and Plan-and-Solve Pattern Recognition in AI
By
–
We first noticed this pattern in @yoheinakajima
's BabyAGI project, and then more recently saw it in the Plan-and-Solve paper Paper: https://
arxiv.org/abs/2305.04091
Code: -
LangChain Introduces Plan-and-Execute Agents for Complex Tasks
By
–
Plan-and-Execute Agents Inspired by BabyAGI and the recent Plan-and-Solve paper, we're introducing a new type of @langchain agent We think these are better for more complex tasks, at the cost of more calls to the LLM Blog: https://
blog.langchain.dev/plan-and-execu
te-agents/
… -
Tidybot : robot personnalisé avec IA pour trier
By
–
Tidybot: Personalized Robot Assistance using Large Language Models
— AI Breakfast (@AiBreakfast) 10 mai 2023
In this video, Tidybot sorts laundry, trash, and places items in locations specified by the user pic.twitter.com/OqqGS4TJNGTidybot: Personalized Robot Assistance using Large Language Models In this video, Tidybot sorts laundry, trash, and places items in locations specified by the user
-
FCCM Panel: LLM Architecture and Production Cost Optimization
By
–
Tune in to today's FCCM Panel to hear about architecture choices for Large Language Models (LLMs) through the lens of the cost metric, and why Groq is best suited to run LLMs in production. Learn more at https://
fccm.org/panel-2023/ -
Research papers on extending LLM capabilities beyond conversational chat
By
–
Bien sur ! Quelques exemple : – https://
arxiv.org/pdf/2305.05658
.pdf
…
– https://
arxiv.org/pdf/2303.18223
v10.pdf
…
– https://
arxiv.org/pdf/2303.16199
v1.pdf
…
– https://
arxiv.org/pdf/2304.06975
v1.pdf
…
– https://
arxiv.org/pdf/2305.03726
v1.pdf
… Ca tourne beaucoup autour de l'adaptation des LLMs à un sens plus large que juste du "chat" . -
StarCoder Fine-tuned as Chat Assistant Using Community Datasets
By
–
So by popular demand, we've starting exploring the finetuning of BigCode's recently released code-completion model StarCoder (
https://
x.com/BigCodeProject
/status/1654174941976068119
…) to be a…. Chat Assistant We finetuned it on two high-quality datasets created by the community:
– OpenAssistant’s -
Using LLMs as control policies for robotics
By
–
Il se passe vraiment des trucs incroyables en ce moment dans le developpement des systèmes d'IA.
— Defend Intelligence (Anis Ayari) (@DFintelligence) 10 mai 2023
Les modèles de LLM semblent être la partie "d'adaptation" qui manquait depuis longtemps au domaine. Ils permettent par exemple ici d'adapter les actions d'un robot. https://t.co/pG0NsFD0rTIl se passe vraiment des trucs incroyables en ce moment dans le developpement des systèmes d'IA. Les modèles de LLM semblent être la partie "d'adaptation" qui manquait depuis longtemps au domaine. Ils permettent par exemple ici d'adapter les actions d'un robot.
-
LLMs as Reasoning Blocks for World Modeling and Generalization
By
–
Really nice ! thank you for demonstrating this capability. LLM could then indeed be used as the "reasoning" block to achieves unseen world reasonning and allow tasks to get a better generalization to accomply them.