The aim of my series on “GenAI Demystified” is to help a wide range of readers fill their AI tool boxes with not only tools but knowledge, approaches and a healthy dose of skepticism. Equipped with such a “tool-kit” [“tool-kit” being a metaphor for a set of skills and knowledge], one can master the complex web of AI technologies. To that end, please be reminded that Oracle’s OCI Generative AI Professional certification is free until the end of the month 31-Jul-24. https://blogs.oracle.com/oracleuniversity/post/announcing-oci-2024-generative-ai-professional-certification-and-course
One of the first things I did was to begin building a glossary of terms used in the training material. Being very analytical in nature, I constantly revisited this and evolved the glossary into a categorized taxonomy. This greatly helped me organize the concepts into a structured framework. You can read about it here: Studying for the OCI GenAI Certification? Consider The Power of a Taxonomy of Terms — Roger Cornejo.
Secondly, I would recommend making heavy use of the pause capability in the training material and take notes as you encounter points. I can tell you that there will be questions on the final test that cover details that perhaps at first blush would seem less important; quality notes were invaluable to me in reviewing the material before the final, and my notes helped me call to mind many points I would otherwise have glossed over.
The third thing I would mention is that I also made heavy use of the transcript which follows along as the trainer is speaking. This helped me in several ways: I could scan back (and forward) in the transcript to reenforce some points and to see where the conversation is heading.
Take the quizzes and tests. In my opinion the final test was harder than the sample quizzes and test, but still they were helpful in letting you know the kind of questions you’ll get.
As others have noted, when you take the tests, some of answers choices can be discounted right away because they are obviously wrong. In many cases two answer choices were real close to each other, and those were harder to distinguish the right one of the two. For example, there will be questions that want you to distinguish and understand the trade-offs between say 1) soft prompting, 2) continuous pre-training, 3) parameter efficient fine-tuning, and 4) t-few fine-tuning. To help with this I actually built a truth table for statements like:
a) Modifies all parameters of the model?
b) Modifies all parameters with new data?
c) Modifies all parameters with unlabeled data?
d) Modifies all parameters with labeled data?
e) Modifies only a few parameters of the model?
f) Modified the prompt only?
Those questions were the rows in the truth table and the columns were the concepts of: 1) soft prompting, 2) continuous pre-training, 3) parameter efficient fine-tuning, and 4) t-few fine-tuning. Then I just put a T or F in the “cell” for that question and concept.
Another thing I did was to build a table of the various models discussed in the training and record some of the key features such as the use case they are optimized for, the number of parameters, input and output lengths, and vector dimensions and run limits. This helped me to have a quick comparison overview the various models.
A few other things, make sure you understand the following (see my full list of terms in my “taxonomy” post):
Ø the difference between “accuracy” and “loss”
Ø what is greedy decoding
Ø what “cosine distance” means
Ø what the “Temperature” parameter is and how to use it
Ø what the “Extractiveness” parameter is and how to use it
Ø what is “groundedness”
Ø the pieces and parts of LangChain
Ø the different kinds of prompting techniques
Ø the vector databases and their capabilities
Ø details on RAG such as answer relevance, Sequence Model and Token Model
Ø OCI’s sizing rules for hosting both inference and fine-tuning
Well before taking the test, be sure to test your lock-down browser and the User Access Settings. I actually missed a test [you have 2 free tries] because I couldn’t get Admin rights to my machine.
Wishing you the best. Happy testing!