1 GPT call is a bit like 1 thought. Stringing them together in loops creates agents that can perceive, think, and act, their goals defined in English in prompts. For feedback / learning, one path is to have a "reflect" phase that evaluates outcomes, saves rollouts to memory,
LLMS
-
AI System Generates Wisdom From Input Data
By
–
Turns out if you insert food into that guy, he spits back wisdom!
-
GPT Models Lack Awareness of Their Own Strengths and Limitations
By
–
Interesting non-obvious note on GPT psychology is that unlike people they are completely unaware of their own strengths and limitations. E.g. that they have finite context window. That they can just barely do mental math. That samples can get unlucky and go off the rails. Etc.
-
AutoGPTs: The Next Frontier of Prompt Engineering and AI Systems
By
–
Next frontier of prompt engineering imo: "AutoGPTs" . 1 GPT call is just like 1 instruction on a computer. They can be strung together into programs. Use prompt to define I/O device and tool specs, define the cognitive loop, page data in and out of context window, .run(). https://t.co/EKy84pa5bB
— Andrej Karpathy (@karpathy) 2 avril 2023Next frontier of prompt engineering imo: "AutoGPTs" . 1 GPT call is just like 1 instruction on a computer. They can be strung together into programs. Use prompt to define I/O device and tool specs, define the cognitive loop, page data in and out of context window, .run().
-
GoodAI Autonomous Agents Design Test Programs Autonomously
By
–
In GoodAI, we are working on autonomous agents powered by LLMs. This is one of those projects where the agent gets an initial goal (to build a program) and then autonomously designs tests, runs them, and proposes corrections until the program works. Another use case for
-
Training LLMs to Use Multiple LLMs as Tools
By
–
– Toolformer showed you can train LLMs to use tools. – HuggingGPT showed that you can use LLMs as tools. Who wants to train an LLM that uses other LLMs that can use tools? Can 100+ small LLMs outperform #GPT4?
-
Chat-fished: Mistaking AI-Generated Text for Human Writing
By
–
Getting “Chat-fished”: when you read something assuming it was written by a human but…
-
GPT-4 Limitations: Long-term Planning and In-context Learning
By
–
For example, GPT-4 still doesn’t seem to consistently follow long term plans (if you are developing agents with LLM in the loop), or emulate idiosyncratic personalities, or learn in-context new facts that contradict facts that are already in weights. But my guess is that OpenAI
-
Fine-tuned LLMs vs General Models: When Customization Matters
By
–
Yeah, the need for customized find-tuned LLMs decreases when the general LLM can do what you need out of the box. The only use I see for fine-tuned LLMs is when GPT-4 doesn’t want to do what you need, and prompt engineering doesn’t help.
-
Dialog-Enabled Resolving Agents Enhance LLM Clinical Task Performance
By
–
8/ Dialog-Enabled Resolving Agents (DERA) – a paradigm to enhance large language model completions by allowing models to communicate feedback and iteratively improve output; DERA outperforms base GPT-4 on clinically-focused tasks.