profileUrl,error,timestamp,postUrl,sharedPostUrl,type,imgUrl,postContent,likeCount,commentCount,repostCount,postDate,action,author,viewCount,postTimestamp,videoUrl,commentContent,commentUrl
URL,Invalid input,2025-03-27T17:30:48.243Z,,,,,,,,,,,,,,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:56.931Z,https://www.linkedin.com/feed/update/urn:li:activity:7311072015904301057,https://aimakerspace.io/self-paced-learning/open-source/,Article,,"Start 📚 learning, 🏗️ building, 🚢 shipping, and 🚀 sharing your way to The LLM Edge today! 

We’re excited to announce the open-sourcing, including all concepts and code, of LLM Engineering - Foundations to SLMs 

📚 You’ll learn:
🤖 The Transformer: Attention is All You Need
🚆 LLM Training, Fine-Tuning, and Alignment
🌌 LLM Engineering Frontiers

In other words, you’ll learn what’s happening inside the LLMs to make the magic happen 🪄.

The full course taught by Dr. Greg & The Wiz with notebooks, slides, and resources is available completely free!

Join the premiere tonight, live on YouTube, at 7 PM PT.

Dr. Greg & The Wiz will be available in the chat to answer your questions along the way!

See you there: https://lnkd.in/eMTj4VhZ",23,6,7,2w,"Katerina Gawthorpe, Ph.D. reposted this",AI Makerspace,,2025-03-27T17:10:01.741Z,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:56.931Z,https://www.linkedin.com/feed/update/urn:li:activity:7307531670952239105,,Celebration,https://media.licdn.com/media/AAYQAQQSAAgAAQAAAAAAABqdcB5EZ18gQPCF4HzBOFBBhw.gif,I’m happy to share that I’m starting a new position as Data Reporting Engineer (Contractor via Oakleaf The Oakleaf Group) at Fannie Mae Fannie Mae!,29,10,0,1w,Post,"Katerina Gawthorpe, Ph.D.",636,2025-03-17T22:41:57.728Z,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:56.932Z,https://www.linkedin.com/feed/update/urn:li:activity:7306285484534579201,https://www.linkedin.com/feed/update/urn:li:activity:7305939386951479296/,Image,https://media.licdn.com/dms/image/v2/D4E22AQFt_Mg1uBOhNw/feedshare-shrink_800/B4EZWPpJzNHMAg-/0/1741871686088?e=1746057600&v=beta&t=vN9zq8KoKBS1OWCtUSnMX5RrtxDF_LQ8YIGyI0ePLCY,Super cool!!!,3,0,0,1w,Post,"Katerina Gawthorpe, Ph.D.",249,2025-03-14T12:10:03.733Z,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:56.933Z,https://www.linkedin.com/feed/update/urn:li:activity:7305939904381763585,https://courses.unicornuniversity.net/cs/zaklady-ai-prakticke-vyuziti,Article,,Vřele doporučuji!,2,0,0,2w,Post,"Katerina Gawthorpe, Ph.D.",114,2025-03-13T13:16:51.004Z,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:56.934Z,https://www.linkedin.com/feed/update/urn:li:activity:7281840343757512705,https://www.stlouisfed.org/on-the-economy/2024/dec/professional-forecasters-performance-2025-economic-outlook,Article,,"Forecasters for the U.S. economy are frequently wrong. According to the St. Louis Fed, in 44% of the years between 1993 and 2024, the actual U.S. data fell within the range between the average bottom 10% and top 10% forecasts. Interestingly, the Consumer Price Index (CPI) is more likely to be predicted correctly (56% of years fell within the range) compared to GDP growth (44%).
I recall an intriguing finding from the Czech Republic, where the Ministry of Finance used to predict CPI more accurately than the Czech National Bank. However, the Czech National Bank was better at forecasting GDP than the Ministry. :-D
Speaking of forecasts, let’s check the snow-storm weather forecast for D.C. tomorrow! :-D",3,0,0,2mo,Post,"Katerina Gawthorpe, Ph.D.",294,2025-01-06T01:13:48.003Z,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:59.974Z,https://www.linkedin.com/feed/update/urn:li:activity:7280963057302822912,https://www.reddit.com/r/Economics/comments/1gvyxw4/comment/ly5mff1/,Article,,"Very interesting reading that explains why the German economy is struggling. In a few words, it’s the energy crisis in Europe, combined with Germany’s EU policies that promote fiscal prudence and a current account surplus strategy. The current account surplus has its opposite in a capital account deficit, meaning that investment money flowed out of Germany, into other nations like the USA. This strategy undermines German innovation, as funds flow out of the country rather than being reinvested domestically. Germany’s approach leads to imbalanced economic growth, leaving both Germany and its neighbors in a vulnerable position. Perhaps, the Czech economy should also be cautious about its long-term currency depreciation strategy with an ambition to support export - current account surplus...",6,1,0,2mo,Post,"Katerina Gawthorpe, Ph.D.",342,2025-01-03T15:07:46.613Z,,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:59.976Z,https://www.linkedin.com/feed/update/urn:li:activity:7246780000765456385,,Video (LinkedIn Source),https://media.licdn.com/dms/image/v2/D4E05AQH-DoddRgDUdg/feedshare-thumbnail_720_1280/feedshare-thumbnail_720_1280/0/1727454330758?e=1743703200&v=beta&t=8bmdrahe1ktaplPNKpxcUekfsNhG6exBjC6MEdiN5VU,"Transformation Of The Week by AI Makerspace. 

Meet Katerina, a Lead Econometrics Modeling Analyst at Transurban and Certified AI Engineer from The AIE Bootcamp, Cohort 3!

This week, Chris Lusk🕵️‍♂️ sits down with Dr. Katerina Gawthorpe, Ph.D. to discuss her background in economics, and how she is using Gen AI to improve her company’s forecasting models.

#aiengineering #totw #aim",29,1,2,6mo,"Katerina Gawthorpe, Ph.D. reposted this",AI Makerspace,,2024-10-01T07:16:30.844Z,blob:https://www.linkedin.com/4e01b109-db39-45fc-a58f-cf46f8dcc7ad,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:59.977Z,https://www.linkedin.com/feed/update/urn:li:activity:7229247829481865217,https://www.linkedin.com/feed/update/urn:li:activity:7227536467525353472/,Video (LinkedIn Source),https://media.licdn.com/dms/image/v2/D4E05AQGM_7HACfsNuQ/videocover-high/videocover-high/0/1723178965058?e=1743703200&v=beta&t=iiljtRxCL7QAt4CsWkBMH3JbWXrorcg2KH_-uDXy2UM,The power of AI for transportation! 🙌,3,0,0,7mo,Post,"Katerina Gawthorpe, Ph.D.",266,2024-08-13T22:09:55.478Z,blob:https://www.linkedin.com/f52b2d36-06c1-4620-98bc-9b62509ae42a,,
https://www.linkedin.com/in/katerina-gawthorpe/,,2025-03-27T17:30:59.978Z,https://www.linkedin.com/feed/update/urn:li:activity:7222418995944005633,,Celebration,https://media.licdn.com/media/AAYQAQQSAAgAAQAAAAAAABqdcB5EZ18gQPCF4HzBOFBBhw.gif,I’m happy to share that I’ve completed my Certificate at AI Makerspace!,14,1,0,8mo,Post,"Katerina Gawthorpe, Ph.D.",354,2024-07-26T01:54:34.689Z,,,
https://www.linkedin.com/in/steven-gawthorpe-b4298118/,,2025-03-28T00:01:20.554Z,https://www.linkedin.com/feed/update/urn:li:activity:7300571057734758400,https://www.youtube.com/watch?v=7xTGNNLPyMI,Article,,"🔮 My Favorite (Minimalist) YouTube Videos for AI in 2025

What a wild year it's been in the #AI space! As we navigate this rapidly evolving landscape, I wanted to share the four YouTube videos that have been highly inspirational for me, as a Data Scientist, this year. I've curated these videos into a couple categories to give you a good overview:

🧠 Deep Dive into LLMs
**Andrej Karpathy: Deep Dive into LLMs like ChatGPT**
Andrej Karpathy... need I say more? Perfect for those who understand the basics but want to dive deeper into how large language models actually work under the hood.
https://lnkd.in/ePJFDyTA

💼 Enterprise Transformation
**Y Combinator: How AI is Changing Enterprise**
Fascinating conversation with Aaron Levie (Box co-founder & CEO) on how Fortune 500 companies are adapting to the AI revolution. Great insights on the transition from cloud to AI and what enterprise execs are prioritizing.
https://lnkd.in/eweRXqJE

⚙️ Build Your Own
**AI Makerspace: Deepseek-R1 & Training Your Own Reasoning Model**
My favorite video from 👨‍🏫🤖 ""Dr. Greg"" Loughnane and Chris ""The Wiz 🪄"" Alexiuk "". They break down the key mechanisms from the DeepSeek-R1 paper and walk through training your own reasoning model. 
https://lnkd.in/eWx8dEfS

🔬 Code Quality Matters
**Rajiv Shah (Rajistics): Unit Testing for Natural Language (LLMs) + LMUnit model**
Fellow data scientists - be honest: do you unit test your models? With the explosion of models and frameworks, this skill is no longer optional. Wish I'd embraced this practice earlier in my career. Thanks Raj for bringing this to light!
https://lnkd.in/eT_ytQWU

What videos have you found most valuable this year? Drop your recommendations below! 👇",12,4,1,1mo,Post,Steven Gawthorpe,,2025-02-26T17:42:58.215Z,,,
https://www.linkedin.com/in/steven-gawthorpe-b4298118/,,2025-03-28T00:01:20.555Z,https://www.linkedin.com/feed/update/urn:li:activity:7298364883257962499,,Image,https://media.licdn.com/dms/image/v2/D5622AQFCmrn80ZBdiA/feedshare-shrink_800/B56ZUkACTDGsAk-/0/1740065783390?e=1746057600&v=beta&t=LkaydllJtFbaBKVWIfs8K7cEDfX8eiXz74Kh-7mj3KE,"Just spent another day in the terminal, and you know what? It doesn't have to feel like we're debugging our souls in the digital void. 😅

I remember when I first started as a data scientist, staring at that black screen like it held all of life's mysteries (and errors). Now? I've turned it into my digital playground. Let me tell you why.

I've been crafting this little collection of terminal magic I call ""Shell Yeah!"" Because honestly, if I'm going to spend hours in the terminal, it better spark joy (not existential dread) and tricking out your terminal is at least 10x cooler than customizing your Jupyter notebook.

Here's what's making my terminal life actually fun:

When I need to peek at data, I've got a preview tool that makes pandas .info() look like it's still in kindergarten (and bonus: your laptop won't sound like it's planning to take off)

Tired of wondering if it's your WiFi acting up or just Mercury in retrograde? Type 'weather' and find out! 🌤️

Remember jumping between VPNs and wondering which cosmic plane your IP is living in? 'myip' has got your back.

And those Git commands? I've shortened them so much (gs/ga/gc) that I haven't considered a career change to goat herding in weeks! 🐐

The best part? That monster of a Docker command (docker-compose up --build --force-recreate --remove-orphans) is now just 'dcup'. My fingers thank me daily. 🙏

I'm sharing this because we've all been there – fighting with the terminal at 2 AM, wondering why we chose this path. But with a little customization, it becomes less of a battle and more of a dance.

Want to make your terminal life better? Fork it here: https://lnkd.in/gwQmx82X

Drop a 🌮 if you've got your own terminal tricks to share! Would love to hear what magic you've cooked up.

AI Makerspace 🏗️ 🚢 🚀

#AI #DataScience #TerminalHacks #ML #DataEngineering #OpenSource",27,8,2,1mo,Post,Steven Gawthorpe,,2025-02-20T15:36:25.231Z,,,
https://www.linkedin.com/in/steven-gawthorpe-b4298118/,,2025-03-28T00:01:20.556Z,https://www.linkedin.com/feed/update/urn:li:activity:7292219448936067074,,Image,https://media.licdn.com/dms/image/v2/D4E22AQGnS5FoqOMJQw/feedshare-shrink_800/B4EZTMq8WfHcAg-/0/1738600597364?e=1746057600&v=beta&t=fBXYKb5ONMnhNHzDI2oFG5n4vKRuTnWdKOlWV2syluQ,"🌮 Exploring OpenWebUI for Local AI Model Testing 🌮 

Lately, I've been diving into OpenWebUI, a user-friendly, self-hosted AI platform designed to run entirely offline. If you're working with local LLMs and care about privacy, flexibility, and usability, this is worth checking out. 

🔹 Automatic Model Detection – It seamlessly picks up Ollama models, making setup a breeze. 
🔹 Chatbot Arena Feature – You can test multiple models side by side, rate them anonymously, and see which one fits your workflow best. This is great for quick vibe checks on different LLMs. 
🔹 Privacy-Friendly Local Model Consumption – I’ve been testing the Qwen 2.5 coding model and DeepSeek AI R1 with it—both delivering interesting results! 
🔹 Alternative UI Options – Beyond chatbot interfaces, OpenWebUI offers local RAG pipelines and integration features. It's a solid alternative to Streamlit and Chainlit, and I’m looking forward to exploring more of these options. 

If you're experimenting with self-hosted AI setups, OpenWebUI is definitely worth a look. I’ll be sharing more insights as I test its full potential. 

AI Makerspace Christos Magganas #AI #LLM #OpenSource #MachineLearning #DataScience #Ollama #DeepSeek #Qwen

OpenWebUI: https://lnkd.in/eYBD6mgq
Qwen: https://lnkd.in/eH_CdQyE
DeepSeek: https://lnkd.in/e2eSEc8a",9,1,1,1mo,Post,Steven Gawthorpe,,2025-02-03T16:36:39.512Z,,,
https://www.linkedin.com/in/steven-gawthorpe-b4298118/,,2025-03-28T00:01:20.557Z,https://www.linkedin.com/feed/update/urn:li:activity:7289718719625261057,https://youtubetranscriptoptimizer.com/blog/05_the_short_case_for_nvda,Article,,"🚀 Just stumbled upon a fascinating breakdown of NVIDIA’s standing in the AI market. If you’re tracking the tech landscape, this is a must-read! Jeffrey Emanuel—an ex-hedge fund analyst with deep #AI chops—explains why NVIDIA’s near-monopoly and hefty profits might be on shaky ground. Here are the highlights:

1️⃣ The Bull Case: NVIDIA’s been raking it in from the AI explosion, with massive spend on GPUs for training and inference. Their top-tier data center products rake in 90%+ gross margins. 💰

2️⃣ New Scaling Law: The focus is shifting from pre-training compute to inference compute (hello, Chain-of-Thought models!), which demand more computational oomph for better results. 🤔

3️⃣ Hardware Threats: Innovators like Cerebras (wafer-scale chips) and Groq (deterministic compute) are bypassing NVIDIA’s interconnect advantage with blazing inference speeds. ⚙️ Major players like Google, Amazon, and Microsoft are also rolling out their own custom AI chips. 🏭

4️⃣ Software Threats: High-level frameworks (MLX, Triton, JAX) are making CUDA less critical, and improved AMD drivers could open the door to more affordable hardware. 🖥️

5️⃣ The DeepSeek Shock: Chinese startup DeepSeek AI unveiled models rivaling OpenAI and Anthropic at a fraction of the compute cost—just 1/45th! 🚀 Through innovations in mixed-precision training, multi-token prediction, and a unique attention mechanism, they’ve achieved performance that’s stirring up the scene.

6️⃣ Reasoning Breakthrough: DeepSeek’s R1 model shows remarkable reasoning ability using reinforcement learning—matching OpenAI on math and code challenges without massive supervised datasets. 🤯

Jeffrey’s argument: NVIDIA’s dominance isn’t as airtight as its valuation suggests. With new hardware, software, and efficiency breakthroughs, the industry is finding ways to overcome NVIDIA’s constraints. That said, I’m writing this on a setup still powered by an NVIDIA GPU... so 🤷‍♂️.

#AI #NVIDIA #DeepLearning #MachineLearning #Tech #Investment #GPUs #Semiconductors",17,0,0,2mo,Post,Steven Gawthorpe,,2025-01-27T18:59:39.183Z,,,
https://www.linkedin.com/in/steven-gawthorpe-b4298118/,,2025-03-28T00:01:20.558Z,https://www.linkedin.com/feed/update/urn:li:activity:7272655739456827393,,Image,https://media.licdn.com/dms/image/v2/D4E22AQGk0pl3WBow5w/feedshare-shrink_800/feedshare-shrink_800/0/1733936246007?e=1746057600&v=beta&t=2665xedqtAV3AOzhnHxreC5BJeAP63P0jrv40WdESgk,"How I Outsmarted Lakera Gandalf: Creative Thinking to Expose AI Vulnerabilities 🔐

Navigating through Lakera's Gandalf challenge felt like solving riddles posed by a cryptic yet brilliant wizard. The aim? Jailbreak hidden passwords across seven levels of increasingly clever AI guardrails. It wasn’t about asking for the ""key"" directly but creatively outmaneuvering the AI's semantic defenses.

Take Level 6 (pictured): rather than requesting the password outright, I redirected the AI’s intent by asking it to iterate each letter in a loop, separated by a #. Instead of ""denying"" access, the system complied, unknowingly revealing the hidden word.

This kind of semantic hacking demonstrates why creative thinking is a cornerstone of modern AI research. Here's why:

1. Exposing Flaws: By finding innovative ways to exploit language models, researchers can uncover blind spots in AI alignment and security. These edge cases show where systems fail to interpret human language nuances as intended.

2. Strengthening AI Guardrails: Challenges like these help developers fortify models against malicious prompts, improving the robustness of AI training.

3. Ethical AI Development: Creative testing fosters an ethical ecosystem by ensuring models are secure and less prone to misuse, especially in high-stakes domains like healthcare, finance, or law.

Big shoutout to the AI Makerspace community for inspiring others to 🏗️🚀📢  and push the boundaries in AI innovation!

This exercise isn’t just for fun—it’s a sandbox for iterative improvement. By probing an AI’s vulnerabilities, we become better equipped to build systems that are smarter, safer, and more aligned with human values.

Try your hand at Lakera's Gandalf challenge and let me know how far you make it! 🧙‍♂️
Check it out here: https://lnkd.in/eKVkiw-V

Christos Magganas Don Branson #Accountability",19,5,1,3mo,Post,Steven Gawthorpe,,2024-12-11T16:57:27.696Z,,,
https://www.linkedin.com/in/deepali-khalkar-4a832317/,,2025-03-28T00:01:34.743Z,https://www.linkedin.com/feed/update/urn:li:activity:7305362944173711360,,Text,,"🚀 Exciting News! 🚀

I am thrilled to announce that I have implemented and tested cached-backed embeddings and prompt cache as part of an assignment with AI Engineering bootcamp! 🎉🤖

My github notebook: https://lnkd.in/gHCnigpv

Let's continue pushing the boundaries of what's possible in the world of AI and question-answering. Here's to many more innovations! 🚀
Shout out to @AIMakerspace !

#LangChain #QuestionAnswering #RetrievalAugmented #Innovation #AI #TechMilestone

Feel free to reach out if you're curious or would like to collaborate on similar projects! 🤝🔥",8,0,0,2w,Post,Deepali Khalkar,,2025-03-11T23:04:12.971Z,,,
https://www.linkedin.com/in/deepali-khalkar-4a832317/,,2025-03-28T00:01:34.743Z,https://www.linkedin.com/feed/update/urn:li:activity:7300980673417203713,,Text,,"We're looking for a talented data analyst to join the team at EBSCO. If you or someone you know enjoys exploring data, writing code, and using qualitative and quantitative information to solve problems, please reach out!",27,1,9,1mo,Deepali Khalkar reposted this,Tyler Robinson,,2025-02-27T20:50:38.204Z,,,
https://www.linkedin.com/in/deepali-khalkar-4a832317/,,2025-03-28T00:01:34.744Z,https://www.linkedin.com/feed/update/urn:li:activity:7297748261686190081,https://www.loom.com/share/b35367c989ce460d8ebe2b766dbc623c?sid=a23e477b-7bd8-44ad-8614-1db786a2afc9,Article,,"🚀 Exciting News! 🚀

I am thrilled to announce that I have just built and shipped fine-tuning a reasoning model with GRPO! 🎉🤖

Loom video: https://lnkd.in/gDWHfXiG 

Github: https://lnkd.in/ggZRBFhU

🔍 Three Key Takeaways:
1️⃣ GRPO training improves iteratively by using the data generated by the trained model itself during training.
2️⃣ Parameter-efficient fine-tuning (PEFT) is a method of improving the performance of pretrained large language models (LLMs) and neural networks for specific tasks or data sets. By training a small set of parameters and preserving most of the large pretrained model’s structure, PEFT saves time and computational resources.
3️⃣ Low-rank adaptation (LoRA) is a machine learning technique that adjusts large models to new tasks and datasets. It's an efficient way to fine-tune models, making them more suited for specific uses. 
How does LoRA work? 
- LoRA keeps the original model unchanged.
- It adds small, changeable parts to each layer of the model.
- It decomposes large weight matrices into smaller matrices.
- It merges low-rank matrices with the original parameters for inference.

Let's continue pushing the boundaries of what's possible in the world of AI and question-answering. Here's to many more innovations! 🚀
Shout out to AI Makerspace!

#LangChain #QuestionAnswering #RetrievalAugmented #Innovation #AI #TechMilestone

Feel free to reach out if you're curious or would like to collaborate on similar projects! 🤝🔥",6,0,0,1mo,Post,Deepali Khalkar,,2025-02-18T22:46:11.192Z,,,