reply if you have other ideas about how language models could currently be improved without any new developments and I'll add them to the thread
AI
-
Language Models Gaining Access to Tools and APIs
By
–
giving language models access to tools and api’s chatgpt has done this with plug-ins, no longer can you say chatgpt is bad at math or doesn't have access to current info soon every api on the internet will be forced to have language model interaction documentation
-
Language Models with Long-Term Memory via Embeddings and Search
By
–
language models with long-term memories using embeddings and semantic search language models will be able to access some short convo you had a month prior
-
Chained prompt sequences as the future of AI APIs
By
–
using pre-created chained-prompt sequences instead of just one single prompt as api costs continue to decrease this will become more popular, the intermediate prompts will be hidden and only the output will be shown could imagine chatgpt integrating a "prompt marketplace"
-
GPT-4 Vision Capabilities: Multimodal Reasoning and Image Generation
By
–
I haven't even touched on the vision capabilities possible in gpt-4 some obvious ones:
-multi-modal reasoning and action (e.g. MM-REACT paper)
-better image gen (imagine gpt-4 iterates on your Midjourney prompts) -few-shot prompting and learning through images/videos -
Parallel Language Models Orchestrated by Conductor LLM
By
–
running language models in parallel with each one focused on a sub-task, all orchestrated by a conductor language model picture something like a massive tree of GPT models working on answering a single complex prompt
-
Language Models with Dynamic Memory and Reflection Capabilities
By
–
language models equipped with dynamic memory and the ability to reflect (e.g. Reflexion paper)
-
Local Language Models Using Cloud Models as Plugins
By
–
language models running locally on your machine that treat more complex language models in the cloud like plug-ins
-
Using Language Models to Pre-process and Optimize Your Prompts
By
–
using another language model for pre-processing your prompts to transform them into something that gives better responses if you want a prompt that you can use to already do this, read the prompt tip section here: https://
thepromptreport.com/p/report-5-eve
ryone-write-jailbreaks
… -
Unlocking Language Models’ Potential Without New Development
By
–
gpt-5 is not needed to 100x the potential these models have we could stop all language model development today and we still haven’t scratched the surface of their capabilities here are a few non-obvious ways language models can be improved without creating any new models: