Unlikely to happen with proprietary systems because they probably think of their prompts as trade secrets (though they'll leak easily enough to anyone who tries to get them), but open source AI assistants would have less incentive to try and hide their implementation details
@simonw
-
AI Assistant Prompt Transparency and Logging Portal Access
By
–
Something that could really help with devices like that would be prompt transparency: I would like to be able to sign into a portal somewhere and see detailed logs of exactly what prompts and responses my AI assistant used to answer my questions, RAG context included
-
Tech Layoffs: Cost-Cutting Moves Shift Engineering Work Cheaper
By
–
My guess is it's one of those things where every division was told to find 5% to cut, and some exec somewhere found a small team of expensive engineers (thanks to both experience and tenure) doing something they figured could be handed off to something smaller and cheaper
-
Evaluating Small Models Beyond MMLU: Tool Use and Summarization
By
–
MMLU tests mainly for knowledge, which doesn't make much sense for smaller models – I'm more interested in how good it is at tool use, summarization, fact extraction etc – I've not figured out the best commonly reported benchmark for that yet though
-
Trusting AI Agents: The Importance of Transparency
By
–
Honestly I haven't personally experienced any agents that fit that description – I only trust agents where I can see exactly what they are doing, and beyond Code Interpreter I haven't actually used many examples myself
-
Defining AI Agents: Code Interpreter as Goal-Achieving LLM
By
–
(depending of course on which definition of "agent" you like to use! I'm using Code Interpreter as an example of an LLM that runs in a loop trying to achieve a goal across several attempts, using an external tool)
-
ChatGPT Code Interpreter: Agent Behavior and Default Integration
By
–
That was the original release of ChatGPT Code Interpreter for me – it's usability dropped quite a bit when they baked it into default ChatGPT, because it's harder to understand when and why it's going to activate It's still an amazing example of useful agent-like behavior
-
Unreliable Computer Systems: Navigating Modern Computing Challenges
By
–
We somehow managed to invent the world's most unpredictably unreliable computer systems! They're still incredibly useful, but you have to know all sorts of tricks to stay out of the many traps they unwittingly set for you
-
Engineering LLM Systems with External Tools for Reliability
By
–
But in a well-engineered system that provides access to tools, runs searches against external systems, then feeds those results back into the LLM as part of the context supplied alongside the prompt: the reliability shoots up enormously… but there will still be weird edge-cases
-
Can You Trust LLMs to Tell the Truth?
By
–
"Can I trust LLM AI to tell me the truth?" is such an interesting question Short answer: no, but it varies depending on the context Expecting it to tell the truth based on its weird opaque blob of matrices derived from its original training data is very risky indeed