Tech Break by Friday

I don’t know if all of you know how much I love to talk about business or side hustles since this podcast does not focus on these topics.

I would love to talk to you today about how you can use AI to create a logo. I am not good with design, but I tend to have a good eye for aesthetics. Since the advancement of AI in our era, there are multiple tools infused with AI to create cool and creative logos with your prompts.

The tool I have used for my own logos until now is Logo.com. The process is straightforward:

First, you provide your business name

Then, you can choose your color, theme, or style.

In a couple of minutes, you have a design for your logo.

Now, if you want, you can either edit the design or let the AI edit it with prompting.

Besides Logo.com, there are various tools on the market, such as Looka.com, Brandcrowd, and Logo.com.

There are pros and cons for all the tools. Some of them have paid plans, and some of them are free.

This is the interface of Logo.com:

(Source: https://logo.com/)

The blog post contains more tools like this, along with their price ranges and services. I used this to create a new logo for the podcast, and I have to admit that I really liked the result.

Also, if you would like more autonomy in designing your own logo but would like some AI features to help, you can try Canva. Canva is a great design tool, and you now have the option for a free trial to use the AI features.

This is only one of the tasks that AI has proven so helpful. Now, you can create your logo with minimal costs. Especially before your business makes revenue, you don’t have to spend a high amount of money to design your brand.

I hope this episode was interesting to you today. Thank you for listening. If anyone else is interested in this, please share it with them.

If you have any podcast ideas that you would like to see, please use my Substack chat to share them with me.

Have a great time, everyone, until the next one. <3

References:

* Logo.com: https://logo.com/

* Superside AI Logo Generators: https://www.superside.com/blog/ai-logo-generators

* Looka.com: https://looka.com/

* Canva: https://www.canva.com/

* Canva Call Members Link: https://www.canva.com/join/fcx-xgz-sgy

Hi everyone, this is another episode of Tech Break by Friday. Where does the term Agentic AI come from? What are AI agents, and why use them instead of foundation models?

Psychologist Albert Bandura introduced the concept of agency. Agency in individuals is described as the ability to act in full or partial autonomy. So, AI agents are AI systems that can perform a certain number of tasks autonomously.

How does an AI Agent System problem solve:

* Perceive: The system gathers information from its environment. In this stage, it will try to find key elements in the data, depict categories, or determine the data's relevance.

* Reason: An LLM acts as the operator or reasoning engine. It understands the tasks given, allowing tools to be utilized to produce results and reach the final solution. One use case would be to use RAG to determine how to solve a specific problem from the data.

* Act: It makes decisions and takes appropriate actions. By integrating custom implemented or external tools, agents can act with them to get a result for the problem presented. If the agent receives a prompt or an instruction on how to solve the specific issue, it will act on the instructions. This removes the manual work of maybe calling different tools and collecting their results.

* Learn: Usually, there is also a feedback loop. This loop allows the agent to improve on the tasks for next time. It could be like a flywheel since the data from the agent’s iterations is fed back to the system to enhance the agent.

(Source: Pounds, E. ,2024)

A powerful solution for businesses to drive faster decision-making and operate more efficiently.

As humans, we use our short-term and long-term memory to remember things more easily. Short-term memory helps us remember the context of a conversation with another human. In agents, short-term memory can be used to keep the conversation context in the same thread as customer bots. I want you to know that AI agents remember the information that you have provided in the conversation previously. But we also have long-term memory. A direct equivalent to long-term memory for AI agents is a vector store that contains documents transformed into vectors. The agent has access to this data store to use at all times, and they can act upon them.

AI agents can act autonomously, but they need an LLM as their main component. Their ability to reason depends on the large language model used for this specific task. Agents have a memory to act on the context provided and/or the external data. Their ability to solve problems is simulated from how we humans observe, reflect on the world, and decide if a certain action will solve the problem. In case of success, we learned how to solve the issues, and in case of failures, we knew what wouldn’t so we could try something else.

References:

* Pounds, E. (2024, October 22). What Is Agentic AI? NVIDIA Blog. https://blogs.nvidia.com/blog/what-is-agentic-ai/

* Rogerson, K. (2024, October 15). Agentic AI Explained: Definitions, Benefits & Examples of AI Agents. Interface.Ai. https://interface.ai/blog/what-is-agentic-ai/

* (24) Understanding Agentic AI: Origins, Applications, and Its Near-term Impact on Human Resources | LinkedIn. (n.d.). Retrieved March 26, 2025, from https://www.linkedin.com/pulse/understanding-agentic-ai-origins-applications-its-near-term-hill-ir6fc/

* Bandura, A. (1997). Self-Efficacy: The Exercise of Control. W.H. Freeman.

* Skinner, B.F. (1971). Beyond Freedom and Dignity. Knopf.

* Marcus, G. (2019). Rebooting AI: Building Artificial Intelligence We Can Trust. Pantheon.

* Boden, M. (2018). Artificial Intelligence: A Very Short Introduction. Oxford University Press.

I came around the term vibe-coding by a post from Andrej Karpathy on X:

There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g., Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper so I barely even touch the keyboard. I ask for the dumbest things like "decrease the padding on the sidebar by half" because I'm too lazy to find it. I "Accept All" always, I don't read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I'd have to really read through it for a while. Sometimes the LLMs can't fix a bug so I just work around it or ask for random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.

-Andrej Karpathy on X

In my opinion, vibe coding is a new way for people to interact with LLMs. They can create initial features and apps based on their ideas. Highly technical people might be better at explaining what they want to get the desired result. Anyone can use it. I have used Cursor IDE with Claude 3.7 Sonnet, and I really enjoyed the process of building an app in a couple of hours. I also built a personal website in just an hour. I have constantly reviewed the code that the LLM suggested, and this is the responsible way to do it. If you are a beginner in coding or app development, you can use these tools to start learning. For example, you can ask how you can build a navigation bar for a website and how you can change the styling. For me, I ask questions in various LLMs to learn more and mostly double-check my answers using perplexity.

If you would like me to make a podcast comparing the pros and cons of AI coding generation tools, let me know in the comments.

So, what kind of vibes are you getting? Are you going to try to build something vibe-coding?

Tools mentioned:

* Cursor: https://www.cursor.com/pricing

* Windsurf : https://codeium.com/pricing

I would really appreciate it if you liked or commented on any podcast platform that you are listening to. I make this content free for everyone because I want to provide value to the world.

References:

* https://x.com/karpathy/status/1886192184808149383?s=54

* Cursor: https://www.cursor.com/pricing

* Windsurf : https://codeium.com/pricing

* Music generated by Mubert https://mubert.com/render

Have you heard people talk about AI, LLMs, and LORA? Now you can find out what it is.

LoRA stands out for Low-Rank Adaptation, and it is a clever technique that makes training AI models much more accessible and efficient.

Imagine you have a massive, pre-trained AI model that's already good at generating images or text. Instead of trying to modify the entire model (which would be like repainting a whole house), LoRA lets you make minor, targeted adjustments (like just touching up a few walls).

Why LoRA is Amazing for Beginners

LoRA makes AI training accessible because:

* Much less computing power is needed: Instead of training billions of parameters, you might only train a few million.

* Faster results: Training can take minutes or hours instead of days or weeks.

* Less technical knowledge required: Tools like replicate make the process more user-friendly.

* Smaller datasets work: You can achieve good results with just 20-100 quality images.

What You Can Create

With LoRA training, you can teach AI to generate images in specific styles, create consistent characters, or produce specialized content that the original model wasn't capable of.

The process is iterative. You can test your results, make adjustments, and improve your model over time as you learn what works best.

Final Thoughts

LoRA has democratized AI training, making it accessible to hobbyists, artists, and small businesses who previously couldn't afford the computing resources needed for full model training. It's an exciting entry point into the world of AI customization that allows for creativity without the traditional barriers to entry.

Music generated by Mubert https://mubert.com/render

Tools Mentioned (I am not an affiliate of any of these; I just genuinely use their services)

* Google Colab

* Replicate - Training/Fine-tuning using LORA.

* My Hugging Face Account

Thank you! If you enjoyed the content and want to get more, please subscribe. I make all of these episodes for free to share my knowledge with the world.

Citations:

* https://myaiforce.com/real-life-lora-training/

* https://www.datacamp.com/tutorial/mastering-low-rank-adaptation-lora-enhancing-large-language-models-for-efficient-adaptation

* https://www.mimicpc.com/learn/kohya-ss-lora-training-guide

* https://blogs.rstudio.com/tensorflow/posts/2023-06-22-understanding-lora/

* https://turboflip.de/how-to-train-your-own-lora-network-step-by-step-guide-with-kohya/

* https://learn.rundiffusion.com/basic-lora-training-with-kohya/

* https://www.reddit.com/r/StableDiffusion/comments/11vw5k3/lora_training_guide_version_3_i_go_more_indepth/

* https://civitai.com/articles/9340/non-technical-on-site-lora-training-guide-focusing-on-dataset-contents-for-pony-and-illustrious

* https://www.reddit.com/r/StableDiffusion/comments/170f6xx/training_a_lora_on_a_concept/

* https://civitai.com/articles/9005/a-detailed-beginners-guide-to-lora-training-on-civitais-trainer

* https://vancurious.ca/generative-AI-Kohya

* https://datascience.stackexchange.com/questions/130798/why-lora-is-for-fine-tuning-but-not-for-training-too

* https://stable-diffusion-art.com/train-lora/

* https://huggingface.co/docs/diffusers/en/training/lora

* https://civitai.com/articles/3105/essential-to-advanced-guide-to-training-a-lora

* https://datascience.stackexchange.com/questions/123229/understanding-alpha-parameter-tuning-in-lora-paper

Welcome to another episode of Tech Break by Friday. I am Paraskevi, and I am a Software Engineer with expertise in AI. This week, I wanted to discuss what an LLM, or Large Language model, is.

LLMs are artificial intelligence systems able to process and generate text in human language. You might have seen chatbots or language translation applications that use these models. A very famous model is ChatGPT from OpenAI.

You can use LLMs from applications like perplexity. We discussed it in episode 2.

There are some new LLMs released these couple of weeks, and you may find it interesting.

* x.ai released Grok 3 LLM and voice mode. - I really like this because it gives me concise, quality answers.

* Open AI released GPT 4.5. According to their announcement, this is their largest model with a better understanding of the world.

* Lastly, one of my favorite models is the Claude 3.7 Sonnet by Antropic. I have used it intensively, and it has shown significant improvements in coding and agentic use tasks.

As LLMs continue to evolve, they are reshaping how we interact with technology and access information, making them a crucial component of the modern world.

Join the chat to discuss more about interesting tech and AI.

Links below to test the models:

* https://claude.ai/

* https://openai.com/

* https://grok.com/?referrer=website

References:

* https://www.anthropic.com/news/claude-3-7-sonnet

* https://openai.com/index/introducing-gpt-4-5/

* https://x.com/bycloudai/status/1895233722309599364

* https://x.ai/

* https://youtu.be/BDseU-kmDYY?si=Xmu_p8mqoSZ3fhkZ

Music generated by Mubert https://mubert.com/render

Instagram: https://www.instagram.com/tech.break.by.friday/

Hey there! So I am not sure how many of you are still using Google. I am not. Perplexity.ai is the new AI search engine tool I am using in my everyday life and it has saved me a huge amount of time.

Perplexity AI is a conversational search engine launched in 2022 that combines large language models (LLMs) with real-time web searching capabilities 1 5. Founded by former engineers from OpenAI, Databricks, Meta, and Quora, it differentiates itself by providing cited sources within its responses 5.

Key Features

Search Capabilities

* Uses multiple AI models including GPT-3.5, GPT-4, Claude, and Google DeepMind's Gemini 1

* Provides real-time information from across the internet with cited sources 6

* Maintains contextual memory for natural follow-up questions 2

Focus Modes

* Web Focus: Searches across the entire internet

* Academic Focus: Searches scholarly resources

* Social Focus: Analyzes social media content

* Video Focus: Searches video platforms

* Writing Focus: Optimized for content creation

* Math Focus: Powered by Wolfram Alpha 2

Available Versions

Free Version

* Access to GPT-3.5

* Deep Research and DeepSeek R1

* Limited to five Pro searches daily 3

Pro Version

* Access to GPT-4 and Claude

* Experimental Perplexity model

* Additional features like image generation and document upload capabilities 4

As of January 2024, Perplexity AI serves approximately 10 million regular users, with its user base continuing to grow rapidly 3.

Citations:

* https://www.descript.com/blog/article/what-is-perplexity-ai

* https://learnprompting.org/blog/guide-perplexity

* https://de.wikipedia.org/wiki/Perplexity.ai

* https://www.reddit.com/r/perplexity_ai/comments/1dmcf20/what_are_some_use_cases_for_perplexity_ai/

* https://en.wikipedia.org/wiki/Perplexity_AI

* https://www.perplexity.ai/hub/faq/how-does-perplexity-work

* https://www.perplexity.ai/hub/getting-started

I have a referral link from my Pro version. If you like to save 10$ from the Pro subscription I leave the link here: https://perplexity.ai/pro?referral_code=GCJACX7C

PS: Loom cut my demo in 2 parts. I hope you still find these useful. Please let me know in the comments if you liked the episode and what you would like to see in the next episodes. 🫶

Loom Demo Part 1

Loom Demo Part 2

Music generated by Mubert https://mubert.com/render

Thanks for reading Paraskevi’s Substack! Subscribe for free to receive new posts and support my work.

Instagram: https://www.instagram.com/tech.break.by.friday/

Episode 1: Welcome to Tech Break by Friday

In this debut episode, Paraskevi (meaning "Friday" in Greek) introduces her new tech and AI-focused podcast aimed at making technology more accessible and human-centric.

Episode Highlights

Introduction & Mission

Welcome to a space where technology meets accessibility

Breaking down the barriers in the tech skills gap

Making complex tech concepts understandable for everyone

What to Expect from Future Episodes

Practical technology tips and solutions

AI concepts explained in simple terms

Productivity enhancement strategies

Maintaining human connection in the digital age

Topics We'll Explore

Artificial Intelligence developments

Software and productivity tools

Tech trends and their practical applications

Ways to use technology while preserving joy in life

Connect With Us

Instagram: https://www.instagram.com/tech.break.by.friday/

Attributions:

Music generated by Mubert https://mubert.com/render