TL;DR Version

Jev is a new AI model which doesn’t generate text but instead does classification tasks.

So I used it to see if I should switch careers and become a plumber.

Before starting I did some preparations:

• Updated my resume to use it as context.
• Came up with some existential questions like: ‘based on my resume, should I switch careers and become a plumber?’.

Here is the result:

So Jev thinks that with my professional experience I would make a lousy plumber (0% fit)!

Jev doesn’t actually predict if I should become a plumber or not, because it doesn’t have reasoning.

noul_1.pngIt takes the input (my resume) and compares it against all descriptions (plumber, AI Engineer, etc) and sees how well they match.

This result just shows that my resume is an AI Engineer’s resume 🤓.

Bonus: At the end of the article I show you how to use Jev to persuade your way out of boring tasks.

What is Jev?

Now that we are all here, let’s understand what Jev is and how it compares to LLMs.

Let’s go with the simplest example for how an LLM works: it takes text as input and it generates an answer as text.

Jev on the other hand does not generate text but a structured output. But what does this mean?

Jev’s Primitives

Jev uses 3 primitives:

1. Choice: hard categories (e.g. ‘Is an email spam?’)
2. Score: a gradient (e.g. ‘What is the seniority level of this candidate - junior, mid, senior?’)
3. Noul: yes or no questions (e.g. ‘Looking at my blood work, should I down a fourth cheeseburger?’)

1. Choice

Here are some good old categories which fit our career path scenario:

1. plumbing: should I become a plumber because all knowledge work will be entirely automated?
2. computational_engineer: I have lots of experience here, should I maybe just go back doing mechanical engineering?
3. ai_sales_job: if most software will be written by AI, should I pivot to selling AI applications?
4. ai_engineer: Should I keep working as an AI Engineer, because AI will be integrated in most products?
5. none_of_the_above: do something else entirely. Maybe becoming a Youtuber.

"right_career_path": {
  "type": "choice",
  "instructions": "What would be a good fit for me as a next career move?",
  "criteria": {
    "plumbing": "Become a plumber. Leaks don't get fixed by agents, and nobody asks a plumber for an architecture diagram.",
    "ai_engineer": "Keep working as an AI Engineer and hope every company still needs someone to plug AI into things.",
    "ai_sales_job": "Pivot to software sales. The products write themselves now, somebody still has to sell them.",
    "computational_engineer": "Go back to mechanical engineering, because that's what I studied at university.",
    "none_of_the_above": "Something else entirely. I'm on the wrong path."
  }
}

This is where my dream was shattered, because based on my professional experience becoming a plumber might not be a good fit after all.

Probabilities:

• The result doesn’t mean that the category ai_engineer is correct or not, but how well each option fits relative to each other based on my resume.

Confidence:

• Always has a value between 0 and 1 and a high value (91%) shows that we have a peak in the data (ai_engineer 93%).

Just for fun, let’s reduce the confidence by using more nuanced categories and see what kind of ai_engineer work I do?

1. ai_application_engineer
2. ai_solution_architect
3. ai_forward_deployed_engineer
4. machine_learning_engineer
5. none_of_the_above

  "right_ai_career_path": {
    "type": "choice",
    "instructions": "What would be a good next role for me as an AI Engineer?",
    "criteria": {
      "ai_solution_architect": "Designs how AI fits into a company's systems, chooses the right approach and technology, and guides teams through the build without writing most of the code",
      "ai_application_engineer": "Builds AI features into products: new AI applications or AI added to existing software, such as chatbots, RAG, and agents",
      "ai_forward_deployed_engineer": "Works on site with enterprise customers, building and deploying AI systems inside the customer's own infrastructure",
      "machine_learning_engineer": "Trains, evaluates, and deploys machine learning models, working with data pipelines and model performance",
      "none_of_the_above": "Something else entirely. I'm on the wrong path."
    }
  }

This is actually more interesting, because the confidence dropped to 44%. Not because the answer is wrong, but because the spread is mostly between ai_application_engineer (56%) and ai_solution_architect (31%) and Jev is not sure which one fits best.

2. Score

Score categories need to be defined as a gradient (e.g. the seniority level of a candidate based on its resume: junior, mid, senior)

Let’s keep on going with my career path example.

The gradient we define is from full AI Engineer to Plumber: AI Engineer -> AI Sales (further away from AI) -> Mechanical Engineering (still further) -> Plumber (nothing to do with AI).

  "escape_ai_wave": {
    "type": "score",
    "instructions": "How far away from AI should my next career move be?",
    "criteria": [
      "Continue working as an AI Engineer and keep plugging AI into everything, including things that were fine without it",
      "Move to AI sales. Still surrounded by AI, but now I only have to create powerpoint presentation where I explain it",
      "Go back to mechanical engineering, because that's what I studied a long time ago",
      "Become a plumber. ChatGpt won't fix leaks any time soon."
    ]
  }

Score’s result has three parts: score, confidence, and probabilities.

Based on my resume: - on a scale from 0 (stay in AI) to 3 (become a plumber) - I should lean towards stay in AI (a 0.41 score with 78% probability) - with a tendency to go back to mechanical engineering (14% probability) - causing a lower confidence (59%) value because of the split between AI (78%) and mechanical engineering (14%).

3. Noul

This one is simple: a yes or no question (e.g. ‘Looking at my resume should I become a plumber or not?’)

"become_a_plumber": {
    "type": "noul",
    "instructions": "Looking at my resume, should I become a plumber instead of woking as AI Engineer because AI might write all code in the future?"
  }

Similar to the Choice result, we see that my profile is made for AI Engineering and not plumbing.

Bonus - Persuade your way out of boring tasks

Let’s say you received a boring task (e.g. doing PowerPoint presentations), and you want to persuade your boss that you are not a good fit.

Jev can help you!

Jokes aside, it’s always a good idea to learn new skills, and work on different topics even if it includes Microsoft Office work.

Here is the truth why I got the task in the first place using a Noul. My profile does fit a person who could do the PowerPoint presentations.

← Back to home