How I work more on LLMs

I want to work on LLMs but unfortunately sometimes I feel I don't have the time. Day job, kids, watching news + podcasts etc.

We follow some people, sometimes there is a tweet we bookmark but we don't really follow up or we don't really have the time or spend enough time in order to... and there is because there is also a lot of content so we kind of give up. It's not like..

That is too much stuff. But you need to do more. More more more. I think we need to be more strategic about our days and at least me and how we can do more AI.

Adding LLMs into Your Day Job

You can try migrating your day job time into LLM time. Here's what can be done:

1. Your current situation:

   • You've got a 9 to 5 that's not LLM-related
   • Maybe you're in software engineering - front-end, back-end, whatever
   • Could even be an ML engineer not touching LLM stuff (because let's face it, most projects aren't there yet)
   • And yeah, you need to get paid, so the company needs to be on board with your decisions.

2. Steps to take:

   • Schedule a meeting with your manager or a senior team member
   • Clearly express your interest in transitioning to LLM-related work
   • Propose a project and request time allocation:
     • "I would like to dedicate a portion of my time to LLM projects"
     • "I assure you that my current responsibilities will not be neglected"
     • "Perhaps we could start with a modest allocation, such as 10% to 20% of my time"
   • Emphasize that the project will be beneficial to the company, you just need to invest your time first.

3. The pitch: This might work better if you've been there a while, but I think it's fair game to go to your manager and say:

• "Look, I want to be part of the LLM revolution"
• "It's a win-win, I learn, the company benefits"
• Start small, like an afternoon, a week or 1 quarter goal.

It's important to emphasize that you're not requesting time to learn a recreational sport; your goal is to delve into LLMs. This proposition presents minimal risk for the company while offering significant potential benefits. If they grant you this opportunity, you'll likely be highly motivated, creating a mutually beneficial situation. I believe there's a strong possibility of successfully implementing this arrangement.

Which project you can take

LLM Projects in Marketing and Sales

When considering which projects to undertake, marketing, SEO, and sales teams are often prime candidates for LLM integration. Here are some ideas:

1. Improved Call Summarization:

   • If your company records sales calls, you can leverage LLMs to provide more accurate and insightful summaries.
   • For example, if you're using a tool like Kong, extract the calls and re-summarize them using a model fine-tuned for your company's specific needs.
   • This project not only adds value to your company but also provides an excellent learning opportunity.

2. Marketing Content Analysis:

   • Review existing marketing materials that your company has deployed.
   • Use LLMs to analyze what constitutes effective and ineffective marketing for your specific audience.
   • Develop an AI-powered filter or checker for marketing content.
   • Create an internal website where marketing materials can be validated before distribution.

3. AI-Assisted SEO Optimization:

   • Implement LLMs to analyze and optimize website content for search engines.
   • Generate SEO-friendly content suggestions based on current trends and company keywords.

These internal projects offer a great starting point for integrating LLMs into your company's workflow while providing tangible benefits and learning opportunities.

LLM Projects in Software Engineering / ML Engineering

1. Code Review Automation:
   • create a bot that is reviewing PRs, validating the style, the format, the structure, the content, the intent of the PR.
   • you can have a crude version, then iterate later on.
2. Build weely reports / summaries
   • people up the ladder love summaries. Just have an internal blog / confluence that you update weekly with things done by your team

Optimizing Your Time Outside of Work

Alright, let's talk about optimizing your time outside of work. You're probably consuming content already - podcasts, Twitter, all that jazz. What you need to do is curate a list of five to ten go-to resources or people in the LLM industry. Don't have any favorites? No worries, I've got a link with some suggestions of folks I follow. You can check out my personal recommendations for more info on this.

Once you've got your list, stick to it. Make it part of your daily routine. Dedicate an hour or two to diving into their content. But here's the kicker - you want to make this learning process as frictionless as possible. Whether you're in the bathroom, waiting for the bus, or whatever, you should be able to easily absorb information.

Here's a neat trick: when you're driving, you can learn about large language models through audio. Create an MP3 file that's essentially an audiobook. I can show you how to make these with chapters. Each chapter could be a summary of the latest blog posts, explanations of key concepts, or breakdowns of new terms people are using. It's like having a personalized LLM audiobook that you update weekly.

Pop that in your AirPods, and boom - every walk, bus ride, or spare moment becomes a learning opportunity. You've got content at your fingertips, ready to go. Pretty sweet, right?

And the third thing you can do is you have three places where you can have time. Your daily job, the time outside, and now when you want to work. So when you want to work on your side project, I really think you need to have really small side projects because each small project will touch one to ten LLM concept of ideas and you don't have time to explore all of them. So your project needs to be small. Like one weekend, even one day should be enough for you. Don't spend any time on the front end, back end. You should just have like one, even stay with notebook for now and only work one notebook, do something with one notebook. Don't spend time connecting things. Don't not spend time connecting things at this time because when you connect two things, time is more spent into the connections and something else. So just try something, write a verse, write a verse.

When you meet people you should, you cannot meet people doing LLM because you are in an island, people around you doing LLM are busy, it's not really productive to surround yourself with people doing LLM. You can just go on Twitter and follow them, but I'm not sure it's really productive to spend physical time with them. I think it's more productive to just follow them on Twitter.

You'll inevitably spend time with family, friends, and other people not involved in LLMs. Here's where you can leverage your knowledge: figure out how LLMs can assist them. For instance, if you have a friend running a business, take the time to understand their operations and identify pain points. Position yourself as a pro bono consultant.

Start by asking questions like, "Are you recording your calls? If not, perhaps you should consider it." Explain the rationale and the process. Once they've accumulated some recordings or even just a collection of emails, you'll have data to work with.

It's likely that this business owner isn't operating at peak efficiency, particularly when it comes to calls. This isn't a criticism; inefficiency is our default state. It's akin to baking bread - there's always room for improvement in terms of equipment or methodology. This principle is especially applicable to communication and business processes. Even in seemingly straightforward professions like hairdressing, there may be overlooked opportunities for enhancement.

Your objective is to apply LLMs to address these inefficiencies, or at least attempt to do so. While we don't typically seek out problems to solve, in this context, that's precisely what you're doing. You're searching for a set of problems. Given the vast capabilities of LLMs, it's a reasonable assumption that if you identify around five problems, you'll be able to apply LLMs to at least one of them. There might be a 20% chance for each problem, so aim to find five to ten. If you're unsuccessful across the board, chalk it up to an unusually unlucky streak and consider moving on.

When you do find a way to assist someone, aim to demonstrate results quickly - ideally within a day. For example, I once provided daily call summaries for a friend. Each day, I'd send an email recapping the key points. This regular update keeps the AI application fresh in their mind, potentially sparking ideas for further implementations. You can discuss these possibilities later or simply progress to the next project. The crucial part is taking that initial step. With consistent reminders, you'll naturally be inclined to refine your approach, and new ideas will likely emerge over time.