3/ ToolLLM – enables LLMs to interact with 16000 real-world APIs; allows data preparation, training, and evaluation; the authors claim that one of their models, ToolLLaMA, has reached the performance of ChatGPT (turbo-16k) in tool use.
PROMPT ENGINEERING
-
Skeleton-of-Thought: Parallel API Calls Strategy
By
–
4/ Skeleton-of-Thought – proposes a prompting strategy that firsts generate an answer skeleton and then performs parallel API calls to generate the content of each skeleton point; reports quality improvements in addition to speed-up of up to 2.39x.
-
RLHF and Encoder Modules in Decoder Architectures
By
–
Sure I don’t disagree, but since all these were before RLHF finetuned decoder architectures, I am curious whether it’s still necessary to have an encoder module. LLaMA-Adapter would be an example of adding img2txt capabilities w/o cross attention.
-
Emergent Abilities in LLMs: Responding to LeCun’s Critique
By
–
Yann LeCun is obviously a legend but I found this tweet to be quite misinformed. The whole point of "emergent abilities" such as few-shot prompting and chain-of-thought prompting, is that we clearly *did not* explicitly train or fine-tune them into the model. These abilities
-
Context Window Challenges in Large Language Models
By
–
Can Large Language Models effectively use knowledge in their input context? Find out more in our blog post: 'Information Overload: Challenges of Expanding Context Windows in LLMs' https://
samaya.ai/blog@nelsonfliu Ashwin Paranjape @MicheleBevila20 @Fabio_Petroni @maithra_raghu -
Large Context Windows Still Need Better Information Retrieval
By
–
Big context windows aren’t the panacea yet – content in the middle of the input prompt is less likely to be used than if it’s on the ends. New work by team @samaya_AI “current language models do not effectively use their entire context, and that retrieval is still a crucial
-
Tuning LLM Personality: Agreeableness Control Coming Soon
By
–
I’m way too agreeable. I want LLMs to model my personality, and allow me to tune my agreeableness like the volume on my phone. I think that’s coming soon:
-
Shaping LLM Capabilities Through Guidance and Alignment
By
–
i keep hearing people say "we are discovering new capabilities" of LLMs, or "new capabilities emerge". that's the wrong metaphor. we are guiding, shaping, and aligning these models. by showing positive examples of desired outputs we aim to increase the likelihood that an LLM
-
NVIDIA Guide Secures LLM Systems Against Prompt Injection
By
–
A great (new) guide and overview on securing LLM systems against prompt injection by @nvidia We did a webinar on prompt injection a few months and the main takeaway was more awareness was needed around this. Great to see posts like this doing that
-
GPT-4 Performance: Reliability and Future Model Improvements
By
–
with gpt4 its fairly doable. a bit hit and miss with other models tho. i expect this to only improve