[Lex Computer & Tech Group/LCTG] Prompt Engineering

Tue Aug 27 06:39:59 PDT 2024

Bob,

You are right:  I agree with your concern with equating Large Language
Models (LLMs) with Artificial Intelligence (AI).  Doing so succumbs to the
hype of Wall Street.

Unfortunately to the uninformed, introducing this TLA into the conversation
makes the impressive results of LLMs even more mysterious and just for
techies.

-- Harry

On Mon, Aug 26, 2024 at 1:58 PM Robert Primak <bobprimak at yahoo.com> wrote:

> Very good finds, Harry.
>
> Within the context of reading people's thoughts, a lot of what prompt
> engineering can and cannot do depends on the size of the data set it is
> trained on, and the variety of test subjects from whom the training data is
> collected. Too small a sample size in either regard, and there is a real
> chance of researcher bias (conscious or unconscious) entering into the
> picture and affecting the output from the LLM.
>
> I do not equate LLMs with AI, so I restrict my terms to what we are
> talking about -- LLMS only, not more general AI.
>
> -- Bob Primak
>
>
> On Monday, August 26, 2024 at 11:35:42 AM EDT, Harry Forsdick via LCTG <
> lctg at lists.toku.us> wrote:
>
>
> Folks,
>
> Remember when Conor O'Mahony gave his three excellent talks about AI / ML
> systems?  [see
> https://docs.google.com/document/d/1fj8FcfD_e-NEi20O3AL_8fg45DdB136fvyCNl7V2gaM/edit?usp=sharing
> ]
>
> One of the topics he talked about was "prompt engineering".
>
> I interpreted what was said about prompt engineering to refer to the kinds
> of things we all did with plain old Google to get it to return the kind of
> results we were after.
>
> I recently asked https://perplexity.ai the question "*What does "prompt
> engineering" mean in the context of AI search systems?*".  I learned that
> my interpretation was only partially correct.  Instead of being just things
> an end-user does, prompt engineering is also something that a developer of
> an AI question/answering system must do to take the query submitted by
> end-users, and add additional controls to get the AI answer engine to
> return a reasonable, readable answer.
>
> One of the references in this response,
> https://mitsloanedtech.mit.edu/ai/basics/effective-prompts/, discusses
> what an end user should learn to do.  It basically says that this is like
> learning how to ask questions of an expert who knows a lot about the
> subject of your query: you can steer the answer to the one you are after if
> you just ask the question correctly.  To quote from the above reference:
>
> Prompts are your input into the AI system to obtain specific results. In
> other words, prompts are conversation starters: what and how you tell
> something to the AI for it to respond in a way that generates useful
> responses for you. After that, you can build a continuing prompt, and the
> AI will produce another response accordingly. It’s like having a
> conversation with another person, only in this case the conversation is
> text-based, and your interlocutor is AI.
>
>
> Below is the answer I got back from Perplexity.  If you want to dig deeper
> than the Perplexity answer, there are references to source material used in
> the response.
>
> Regards,
>
> -- Harry
>
>
> *What does "prompt engineering" mean in the context of AI search systems?*
> Prompt engineering is a crucial process in the context of AI search
> systems, particularly for large language models (LLMs) and generative AI
> tools. It involves crafting well-structured and effective input queries or
> instructions to guide AI models in producing desired outputs or
> responses[1].
>
> *Definition and Purpose*
> Prompt engineering refers to the art and science of designing and
> optimizing prompts to elicit specific behaviors from AI models[4]. Its
> primary goal is to bridge the gap between human intention and machine
> understanding, enabling AI systems to generate accurate, relevant, and
> contextually appropriate responses[2].
>
> *Components of Effective Prompts*
>
> 1. *Instruction*: Prompts often require specific and clear requests to
> receive the desired output[3].
>
> 2. *Context*: Providing relevant background information helps the AI
> model understand the task better[3].
>
> 3. *Input Data*: This includes any necessary information for the AI to
> process and generate a response[3].
>
> 4. *Output Format*: Specifying the desired format of the response can
> improve the accuracy and relevance of the AI's output[3].
>
> *Role of Prompt Engineers*
> AI prompt engineers serve as intermediaries between machine learning
> models and human users[3]. Their responsibilities include:
>
> 1. Developing sets of inputs to train models for optimal outputs
> 2. Writing text-based prompts for various tasks (e.g., essay writing, blog
> post generation)
> 3. Evaluating AI systems for idiosyncrasies
> 4. Training and fine-tuning emerging AI tools
>
> *What does "prompt engineering" mean in the context of AI search systems?*
> Prompt engineering plays a vital role in AI search systems by:
>
> 1. *Improving Accuracy*: Well-crafted prompts lead to more precise and
> relevant responses from AI models[4].
> 2. *Enhancing User Experience*: Clear and concise prompts make it easier
> for users to interact effectively with AI models[4].
> 3. *Mitigating Biases*: Careful prompt design helps minimize the risk of
> generating inappropriate or biased content[4].
> 4. *Ensuring Consistency*: Proper prompting techniques enable more
> predictable and reliable AI responses[4].
>
> *Skills Required for Prompt Engineering*
> To excel in prompt engineering, professionals should possess:
>
> 1. Understanding of AI, ML, and Natural Language Processing (NLP)
> 2. Programming skills, particularly in Python
> 3. Strong communication skills
> 4. Knowledge of cognitive psychology and linguistics
> 5. Experience with pre-trained AI models like GPT-3 or GPT-4[3]
>
> In conclusion, prompt engineering is a critical aspect of AI search
> systems, focusing on optimizing the interaction between humans and AI
> models to produce high-quality, relevant, and accurate responses.
>
> Citations:
> [1] https://en.wikipedia.org/wiki/Prompt_engineering
> [2] https://www.datastax.com/guides/what-is-prompt-engineering
> [3]
> https://www.techtarget.com/searchenterpriseai/definition/AI-prompt-engineer
> [4] https://cloud.google.com/discover/what-is-prompt-engineering
> [5] https://www.ibm.com/topics/prompt-engineering
> [6] https://www.akooda.co/blog/what-is-prompt-engineering
> [7] https://researchguides.mvc.edu/ai/prompts
> [8] https://mitsloanedtech.mit.edu/ai/basics/effective-prompts/
>
>
> Harry Forsdick <http://www.forsdick.com/resume/>
> Lexington Photo Scanning <http://lexingtonphotoscan.com/>
> Town Meeting Member Precinct 7 <http://lexingtontmma.org/>
> harry at forsdick.com
> <https://mail.google.com/mail/?view=cm&fs=1&tf=1&to=harry@forsdick.com>
> www.forsdick.com
>   46 Burlington St.
> Lexington, MA 02420 <https://goo.gl/xZXT2F>
> (781) 799-6002 (mobile) <callto:17817996002>
>
> Click <http://forsdick.weebly.com/home/my-websites> to see my websites
> *Lafayette Returns*
> 200th Anniversary Reenactment
> *---->* September 2, 2024 (Labor Day) 1pm-2pm Lexington Battle Green
>
> Procession, Lafayette Reenactor, Minute Men, School Children, Bell Ringing,Period
> Costumes, 🔥 Cannon Fire 🔥, and ... cookies shaped like Lafayette
> ===============================================
> ::The Lexington Computer and Technology Group Mailing List::
> Reply goes to sender only; Reply All to send to list.
> Send to the list: LCTG at lists.toku.us      Message archives:
> http://lists.toku.us/pipermail/lctg-toku.us/
> To subscribe: email lctg-subscribe at toku.us  To unsubscribe: email
> lctg-unsubscribe at toku.us
> Future and Past meeting information: http://LCTG.toku.us
> List information: http://lists.toku.us/listinfo.cgi/lctg-toku.us
> This message was sent to bobprimak at yahoo.com.
> Set your list options:
> http://lists.toku.us/options.cgi/lctg-toku.us/bobprimak@yahoo.com
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.toku.us/pipermail/lctg-toku.us/attachments/20240827/1c24314f/attachment.htm>