Just a couple of days ago I was in a room with data scientists (my background in electrical/computer engineering with a DSP/ML specialization and I have been a software engineer for the last ten years). So I was showing the architecture of the RAG system I implemented (with a REST API) and how I got to deploy a Llama-2 model in AWS infra. I could tell all the engineering stuff was way over their heads. I just kept thinking, how can they help me as we move forward to production (deploying to Kubernetes and hardening the code). Software engineering skills are becoming more important to make it in the ML world as now systems need to interact continuously, almost in real-time, with company data rather than taking the data and training a model for later deployment.
@@melski9205 what do you mean by "they wrote ML"? You have no freaking idea of what you are talking about! It just shows you are some type of wannabe who believes he knows everything because he took a course in Udemy. You clearly have never worked in the field. This is one of the most idiotic comments I've ever seen.
Gerardo, i feel you in that situation where you're like "oh, I'm not getting thru here"! I'd suggest to if you have time, run your presentation with one representative of the audience, or even just comment the slides. That way next time you'll hopefully know in advance! :-) But I guess your main point is why the data and ML engineer roles came into being in the first place? Some DS ppl will happily learn engineering skills and mindset, those you can take along on the journey. Maybe for the others it's better to have them in more classic roles in predictive modeling, data viz etc? Or isn't there enough work like that in your context?
I am a software engineer and this is exactly what I have been doing at work for the past few months, using GenAI models and vector databases for RAG to build a chatbot UI. Didn't knew it's called "new age machine learning engineer" now haha. Thought we don't use langchain. We are doing everything from scratch, all the code, be it for vector databases or interacting with ML models. I think this is definitely going to become very popular because of it's usefulness, though from a job market perspective, data scientists would probably still be paid higher. Hope that changes though.
Very well put together. I am an old school software engineer with a strong background in backend development who has, I think according to your definition, become a Machine Learning Engineer. I compensate for the areas in which I am lacking (front end UI development) by a) developing within the Drupal content management system and b) using an AI assistant for a good deal of coding. I have developed, over the past few months, 3 R.A.G. applications which are fairly robust and work very well. I did all of this using PHP as my primary language (people nowadays seem to forget that APIs, by definition, are, or should be, language agnostic). I tell you all of this to say that your analysis is spot on. If you can do this sort of work with a team, that's the best. But, I'm living proof that if you absolutely have to, you can also do it on your own -- thanks in no small part to Generative AI.
Mate, what you described as "new age" is just a regular MLE in any serious company. It has always been about developing applications, deploying models etc. It took me like 5 minutes to understand a RAG architecture, because most of it is literally just calling an API and the rest is very very basic data science by computing vector similarities... you also don't need a vector database to achieve this lol, you can store vectors in-memory a lot of the time queried from a regular Postgres database. I do agree though that the data science space will have higher demands for SWE skills in the future. and I have been anticipating that for the past 3 years with the projects I've undertaken.
My thoughts also. People just want to chase new ML trends so they don't have to stick to boring ML engineering. There is always possible easier solution behind the corner
🎯 Key Takeaways for quick navigation: 00:00 🤖 The "New-Age" Machine Learning Engineer works with generative AI and differs from traditional machine learning engineers. 01:10 🧠 New-age ML engineers need more software engineering skills, focusing on building applications that connect various services. 04:19 🌐 These engineers must work with pre-trained language models, which require a non-deterministic mindset and context customization. 07:44 🔍 Debugging and experimentation are crucial as these models can provide incorrect outputs; you need the mindset of a data scientist. 11:10 🧰 Gen AI tools complement traditional ML models, and new-age ML engineers have an additional set of models in their toolkit. 15:04 💻 New-age ML engineers should be proficient in Python, APIs, Factor databases, data processing, and building web applications. 16:42 📈 Generative AI is trending, but traditional data science and ML skills are still highly valuable in the digital transformation of organizations. 19:00 🌟 Gen AI is just one tool among many, and it's essential to experiment with it while keeping traditional skills in mind for turning raw data into valuable information.
Great video! I like when you implicitly say that this new role is about SW with a probabilistic mindset (ps: i see probabilistic as one form of non-determinism), it makes total sense to me. By the way, I have been people describing this new role as AI Engineer, where the main goal is to develop reasoning systems instead of learning system as ML Engineers would do
I came to the same conclusion. Less about training models mode about Vector DB and building pipelines, using some json to pass around data and present it to the user in a web UI.
I 100% agree. I am a financial analysis major and have realized I can essentially automate all of my research, visualize it and make it a lil nice app hahahah. all with ML / AI , etc. But MAN as a nooby this shit is fucking crazy. I literally had to drop hobbies because their is so much learning it is insane on top of my full time job and school.
Hi, Dave! Can you please make a video on Full roadmap to start with machine learning followed by deep learning to transformers continued with generative AI...and if you can give topic wise roadmap then it will be very helpful for us to start... hope you will do 😊 it's a request from one of your subscriber 😊 Thank you in advance ❤.. Btw you are my favourite youtuber in machine learning and generative AI category .
Probably you should go to university or college because learning is about "grinding". If you don't have time or money, get really intimate with one machine learning book. For students in university, it is really common to get really attached to one book and you are going to understand everything through that book. For me it is George Casella's "Statistical Inference"
I've been doing exactly what you said in this video ever since Chat-GPT was released (back then we used the GPT-3 API instead of the GPT-3.5 and 4. I can confirm that everything said in this video is true.
Love this! I reallly like how you don't just say use Botpress or Voiceflow and you will have a full on valuable product for most businesses - you really state the hard facts; you need some software skills to truly adapt and customize down the line. Would love to see more vids on the end-to-end software engineering workflows and pipelines for these AI projects. Keep up the good work!
Thanks for the video Dave. Usefull and practical. I would appreciate (and others as well I think) a video about using GPU vs CPU or both at the same time. There is a lot which can be improved, as I notice unexplainable results in performance when running the same code and switch from GPU to CPU to compare speed. You would expect that a GPU is always x times faster but this is not always the case I noticed. Sometimes a GPU is only like 10-20% faster than, eg an i5 12th gen. A lot has to do with optimizing the usage of the CUDA toolkit, but even then, I feel this can be way more efficient and we currently 'waste' precious compute capacity because of that. 'Ga zo door Dave ! Je brengt soms als enige gerelateerde content naar buiten zoals deze video, zeer waardevol!'
🎯 Key Takeaways for quick navigation: 00:00 🤖 The New-Age Machine Learning Engineer - New-age machine learning engineers work with generative AI. - They require software engineering skills and a data scientist's mindset. - Projects involve connecting multiple services, APIs, and managing data flows. 04:19 🛠️ Working with Pre-trained Language Models - Pre-trained language models require making them specific for business problems. - Working with non-deterministic models and understanding user-dependent outputs. - Combining software engineering skills with data scientist's experimental and debugging mindset. 10:56 🧰 Role and Skills of New-Age Machine Learning Engineers - New-age ML engineers need a mix of software engineering and data science skills. - They play a crucial role in building end-to-end AI applications. - Skills required include Python, API usage, working with factor databases, and front-end development. 16:42 📊 The Place of Generative AI in Digital Transformation - Generative AI is a valuable tool, but traditional AI tools like data science and machine learning remain essential. - Companies should consider both generative AI and traditional AI tools for productivity and value. - Experimenting with generative AI can complement existing skills in the data intelligence field. Made with HARPA AI
The role of software dev and engineer is becoming more and more general. Basically, other domains are continuing to merge and it seems those who are software dev or engineer are either naturally picking up these skills in other field/domains and/or it’s becoming more of the ‘norm’ to know Just my opinion
00:04 The new age machine learning engineer has different skills compared to a traditional machine learning engineer 02:18 Building applications with language models 04:42 Software engineers need to adapt their mindset to work with non-deterministic generative AI applications. 07:07 Building classification algorithms requires a combination of software engineering skills and experimental mindset. 09:33 Large language models are a new toolkit for machine learning engineers with their own rules and strategies. 12:00 AI applications are customer-facing and require a backend AI engineer 14:14 Key roles in completing these projects are configuring data connections and logic setup, frontend development, and understanding of Python. 16:27 Generative AI is a big step forward, but traditional analytics and machine learning continue to account for the majority of tasks and have new applications. 18:44 Data intelligence is about turning raw data into valuable information. 20:42 Subscribe and stay updated on RU-vid
Hi Dave!! Great vid as usual. I tried to schedule a meeting with you, but no slots availble :(. When will you be availble again? I am interested in your program.
Thanks! We expect to have more room end of October again. I am sorry about the inconvenience. Make sure to subscribe to the newsletters on our website to be the first to be notified when we are opening up new slots for Data Freelancer again.
Sounds more like the data scientists will be unnecessary for most of these teams, just software developers with a bit of devops and machine learning skills.