I love @Aella_Girl
’s bold, empirically-driven Kinsey surveys for the modern age. But by same token I am worried about people starting to take advice from a chatbot with a very superficial notion of what it is taking about. +1 to still asking hive mind on a question like this.
LLMS
-
Chatbot Advice Risks: Beyond Superficial AI Understanding
By
–
-
Interactive LLM Research Transforms Information Discovery Workflow
By
–
+1, it's so fun to conversationally/interactively research a space with an LLM – get summaries and insights distilled into bullet points and table format. hard to go back to my previous method of chaotic refinement of a bazillion chrome tabs
-
Langchain Makes Math News Easy With Referencing Modules
By
–
A lot of this is already possible, but not super reliable yet. @hwchase17
’s Langchain makes math and news easy. I’ve built some referencing modules on top of it as well… might share more on this soon. -
LangChain Adds Custom Prompt Support for LLMBash API QAEval
By
–
Custom prompts for LLMBash, API, and QAEval chains – @nickscamara_ Francisco Ingham Continuing the trend from yesterday, more chains now have the ability to pass in custom prompts! LLMBash: https://
langchain.readthedocs.io/en/latest/modu
les/chains/examples/llm_bash.html#customize-prompt
… QAEval: https://
langchain.readthedocs.io/en/latest/use_
cases/evaluation/question_answering.html#customize-prompt
… -
LangChain v0.0.63 Releases Enhanced Parameters and Custom Prompts
By
–
v0.0.63 Expose more @serp_api parameters – h/t @kweyamba_ai
Custom prompts for LLMBash, API, and QAEval chains – @nickscamara_ Francisco Ingham
Improved css on docs site – h/t Albert Avetisian -
Fine-tuning diverse high-quality data for better AI generalization
By
–
5. Finetune Finetune with diverse, high-quality, and task-specific to help the model generalize better and avoid hallucination.
-
External Filters Add Constraints to Improve AI Text Generation
By
–
4. External Filters Add specific constraints or thresholds on generated text to filter out the unrealistic or non-sensical generated output.
-
Limit predicted tokens to reduce AI hallucination risks
By
–
3. Limit the number of predicted tokens When generating text, limit the number of tokens the model is allowed to generate. The longer the text generation process, the higher the risk of hallucination.
-
Smaller Models Reduce AI Hallucination Tendencies
By
–
2. Use a smaller model The size of the model is directly proportional to its ability to remember and generate realistic text but it also increases the tendency to hallucinate. Try using a smaller model that is better suited to the task at hand to avoid model hallucination.
-
Five Practical Tips to Reduce LLM Hallucination
By
–
Large language models like GPT-3 have a tendency to "hallucinate" or generate text that is not based on the input data and is not accurate. Here are five practical tips to reduce hallucination in LLMs. (A thread)