yep and we can’t undo the softmax
LLMS
-
OpenAI Tokenizer: Understanding Token-Logprob Key-Value Pairs
By
–
not sure what your first question means, but in this case we rely on knowing the tokens of the tokenizer; openAI gives us back key-value pairs of token-logprob
-
LLMs Predict Human Behavior Complexity for Policy Research
By
–
Can large language models predict the complexity of human behavior in experiments? An interdisciplinary team seeks to push the boundaries of AI to help social scientists identify effective policy or public health interventions. https://
stanford.io/3HdVf11 -
Supervised Fine-Tuning and Prompt Engineering as Forms of AI Model Alignment
By
–
Yes. Arguably supervised fine-tuning and even prompt engineering are forms of model alignment even if nobody calls them that.
-
AI Alignment and Instruction Tuning Discussed in InstructGPT Context
By
–
The sense used in the InstructGPT paper is good — a model is aligned when it does what its designers want. Instruction tuning is the canonical form of LLM alignment, but earlier methods like filtering pre-train data of undesired content count too.
-
Inverting Prompts, Training Data Detection, and Distillation Techniques
By
–
– inverting prompts from logits
– training data detection (
@WeijiaShi2 worked on this)
– distillation -
LLMs Logical Error Identification and Self-Correction Benchmark
By
–
Outside of the mathematical setting, large language models can be prone to making logical mistakes. Today we present an evaluation benchmark for mistake identification across settings and examine how LLMs might learn to correct their own logical errors. →
https://
goo.gle/48Ox58T -
Differentiating AI Jailbreaks from Prompt Injection Attacks
By
–
What you’re describing is a jailbreak, not a prompt injection. Prompt injection is when data is misinterpreted as part of the prompt instructions (against the intention of the prompter). A jailbreak is when the prompter bypasses safety policies of the model.
-
Researchers Jailbreak ChatGPT API by Accessing Hidden Token Probabilities
By
–
fun research story about how we jailbroke the the chatGPT API: so every time you run inference with a language model like GPT-whatever, the model outputs a full probabilities over its entire vocabulary (~50,000 tokens) but when you use their API, OpenAI hides all this info from
-
LangChain v0.1.0 Introduces Seven Agent Types Documentation
By
–
LangChain v0.1.0 Highlights Agents As part of v0.1.0, we: Added documentation on the seven different agent types (and when to use each one)
Rewrote agents in LCEL for improved ability to customize Agent Types: https://
python.langchain.com/docs/modules/a
gents/agent_types/
… YouTube: