Fadel Megahed

@lesprgm @dpatil Hi Leslie, they open multiple windows. I believe that the summer window closed but the fall window should open soon. I had students to “co-op” virtually during the semester.

@FadelMegahed @dpatil I currently go to school in Ohio and just found out about this does this mean I am not eligible since the window has closed now that we are in April? I think it ended in March

I've been spending time across the US with college students & faculty. There's a theme: students can't get internships this summer. Not just a few students. LOTS of students. Doesn't matter what kind of school you're at. The pipeline that used to work — apply, interview, get placed — feels broken in a way I haven't seen since 2008. Back then I wrote something called "Hack Your Own Summer" — the idea that if nobody's going to hand you an internship, you should build your own. Make something. HAVE SOMETHING TO SHOW FOR IT!!!! I think it's time to do that again, but bigger. Here's what I'm thinking: What if we stood up a program — free, open, student-led — where students who don't have internships this summer team up and build real projects? Real enough to put on a resume. Real enough to demo. Think of it like a summer-long hackathon or a YC-esque kind of thing with structure: teams, mentors, weekly speakers, and a demo day at the end. I don't have it all figured out. That's the point of this post. I NEED HELP!! If you're a student without an internship this summer — what would make something like this worth your time? I think this would be best if run by a bunch of college kids. Would you want to do that? If you're a professional who'd volunteer a few hours to mentor or coach — what would you need to say yes? If you've run something like this before — what worked and what didn't? Drop your ideas below. I'm reading everything. And if you're a student who wants to help build this from day one — LMK!

I (finally) put together a new LLM Architecture Gallery that collects the architecture figures all in one place! sebastianraschka.com/llm-architectu…

New art project. Train and inference GPT in 243 lines of pure, dependency-free Python. This is the *full* algorithmic content of what is needed. Everything else is just for efficiency. I cannot simplify this any further. gist.github.com/karpathy/8627f…

@justrami To be fair, this is just a terrible starting eleven. No Salah, Raphinha, Mbappe. Arguably, I would put McTominay over Jude too; Jude was outstanding before Mbappe joined!! I know his role is different now, but I just don’t think he is as influential as he was to Madrid.

A very serious question: What does Mohamed Salah need to do to be at Palmer’s level and earn a place in the lineup in the future?! Does he need to achieve fewer numbers and get injured more often?!

Google decided to go absolutely ham with the product velocity. This seems like it lets you build AI agents and automations onto of your google suite. Actually love this.

alright, MCP megathread 🧵 you should probably bookmark this ↓

📊 #rstats #fpl #analytics #or I just published my FPL optimization model using @FPLReview predictions! 🚀 💻 R script finds optimal lineups using a mixed integer programming approach that you can easily update for your own strategy. 🔄 Can force bench/include specific players. I had a specific bench that I wanted to include (Fabianski, Barco, Faes, and Rogers). 🤯 The top 20 lineups are very close in expected value Interesting finding #1: OR model prefers Haaland over Salah! That being said forcing Salah in only drops 3 pts over 3 GWs (so that is definitely within the margin of error). Interesting finding #2: In many of the top lineups, it doubled down on @Arsenal defenders either gabriel and Saliba or Gabriel + Raya. Caveat: Model only as good as predictions. Shows uncertainty in FPL strategy! Analysis open-sourced at: fmegahed.github.io/portfolio/proj…. P.S.: I do not show any of the expected points to preserve the IP of the amazing @FPLReview. You can either subscribe to them, use your own prediction model, or use other sites that make predictions. I believe that the @OfficialFPL 's API also provides expected points via their API. P.P.S: @fplreview also has their own MIP model, but since I did not know its assumptions, I built my own. Have fun with your fantasy predictions!!

I put together a quick colab notebook using Llama-3-8B-Instruct for chatting with a PDF. Unstructured API for partitioning and chunking a large PDF file, FAISS for vector storage, @langchain for RAG, and quantized Llama-3-8B-Instruct (so that it fits on free Colab's GPU). This was fun! #scrollTo=bKZWdc1_J5hm" target="_blank" rel="nofollow noopener">colab.research.google.com/drive/1BJYYyrP…

As we increasingly rely on #LLMs for product recommendations and searches, can companies game these models to enhance the visibility of their products? Our latest work provides answers to this question & demonstrates that LLMs can be manipulated to boost product visibility! Joint work with @AounonK. More details👇 [1/N]

Introducing Command R Plus ⚡ > Beats claude-3, mistral-large, gpt-4 turbo. > 104 Billion parameters. > Built with multi-step tool use and RAG. > Supports 10 languages. > Context length of 128K. > Trained with grounded generation capabilities - citations and responses based on context. > Optimised for code - rewrites, snippets and explanations. > Available in 4-bit and 8-bit too. > Directly available in Transformers to play around. 🤗 huggingface.co/CohereForAI/c4…

Today, we're announcing Claude 3, our next generation of AI models. The three state-of-the-art models—Claude 3 Opus, Claude 3 Sonnet, and Claude 3 Haiku—set new industry benchmarks across reasoning, math, coding, multilingual understanding, and vision.

@jennifer_lin16 @prisonrodeo Are the limits the same with osm and arcgis?

@prisonrodeo use tidygeocoder R package but do it in smaller batches (~1000 a time) -- I've had success with this with a dataset much much larger.

Copilot is live in the most recent normal release of RStudio, btw. Tools -> Global Options -> Copilot

@MineDogucu @polleverywhere Instead of polleverywhere, I use mentimeter and slido. I think they are both “simple” and can be embedded in RMarkdown via iframe.

@ThorstenWuest Sounds good.

@FadelMegahed Thanks Fadel, yes, let’s catch up next week?

This directly impacted our large NSF funded research project as one of the colleagues fired (sorry “RIFed”) at Chambers was working on the econ part of the project… this is impacting every part of the university incl the research enterprise!

@janetfwv @DanielBurbankTV So sorry to hear that. If you have a teaching portfolio that you can share via a DM, I would be happy to share it with our department chair as we are likely to be recruiting for at least one position in the near future.

Story by @DanielBurbankTV that ran last night. Multiple sources, including an interview with me, were used to describe the situation at WVU and the #Chambers8. wchstv.com/news/local/wvu…

It looks like @johnowhitaker & I may have found something crazy: LLMs can nearly perfectly memorise from just 1-2 examples! We're written up a post explaining what we've seen, and why we think rapid memorization fits the pattern. Summary 🧵 follows. fast.ai/posts/2023-09-…

One issue of using embeddings to retrieve relevant documents is that the results might vary with the slightest change in the wording of the query. @langchain MultiQueryRetriever tries to address this issue with the help of LLMs. Let's see how to use it 👇🧵