Johann Beukes

Presented at the Northwestern University’s School of Professional Studies this morning. Beyond the Hype: Real-World AI Implementation and Value Creation. Makes me want to go back to college myself. #AI #MachineLearning #DataScience

Reading while waiting for next flight, returning from South Africa (Stellenbosch) and it was amazing how this book from #PieterDuToit was recommended to me by both people who thought it was revealing a dark secret society and people who thought it was a badge of honor.

Generative AI has its challenges, but AI-augmented UI offers a promising near-term solution. Check out a prototype created with React and OpenAI from our own @rblalock 👇 www.agentuity.com/blog/ai-aug...

Yes. Can’t go wrong with these lads.

Am I one of the 2 monthly listeners? 🤷‍♂️ #AnalogSociety

Worth also pointing out that there are many "tests so easy no AI system can pass them". Moravec's paradox remains. E.g., arxiv.org/abs/2404.12390

Sunday in Florida. Ran into this good boy. So famous, people have tattoos of him.

This week, I saw the power of AI assisted coding in practice. And it was glorious. - 3 teams - 3 projects *given to them* - 3 meet AND exceed results - only 24 hrs (instead of the normal 48hrs + 4hrs for demos) Give smart people all the tools they need!

Had an awesome start of year kickoff with one of my clients this week, culminating in a 24hr hackathon between three teams. Seriously, I felt like a father, proud of every single one on the team. They exceeded expectations by a mile. The #whoop data supports the stress and effect on sleep

The #1 skill for delivering ML models at an eye watering pace is setting performance goals. The goal is not to build the best model. The goal is to build a model that beats the status quo, and get it in production as fast as you can. It works like this:

I had a great conversation tonight with a close friend and the topic of friendship came up. TLDR: A close fiend is not the one you call up when 💩 happened It’s the one you call to celebrate with you when something great happened, and there’s pure happiness coming from that friend with no jealousy

100% agree. This past year, I’ve engaged and hired folks that: 1) I’ve worked with before this knew what they’ve done firsthand or 2) Had incredible experience with the specific technologies needed for a project. What I’ve NOT hired for was some formal education.

Apparently, you can run DeepSeek-V3 locally, provided that you have 8 M4 Pro 64GB Mac minis. ~5 tok/sec.

Happy Festivus! (for the rest of us) “The Festivus pole symbolizes simplicity and humor, rejecting holiday excess with its plain, unadorned design. Its minimalist presence highlights the practical, parody-filled spirit of Festivus, offering both humor and reflection.”

How long is long enough to wait… Probably just skipping this update at this point.

New reading 📖📖📖📖📖 I try to keep a 2-1 ratio of reading DS/AI/ML and Brain/Neuro/Psy literature. #BrainScience #NeuroScience #Psychology @deanbuono.bsky.social

I was informed I should share the YouTube livestream link, and not the zoom link I had previously shared, so here's the updated info: I will be sharing my research on innovation at the intersection of UX, AI, and psychometrics. Friday, Jan 17th 2025, 2-3PM MT www.youtube.com/live/kCmC4i-...

I found out today from one of my clients that they no longer have access to the training code for a specific model, trained using PyTorch and NTLK. So, they are asking me to come up with a strategy to "reverse engineer" the original model...oh, and yeah, I also don't have the original labeled data

Sunday read. 📖📖📖📖📖 Biocosm by James Gardner is a wild take on how life and consciousness could be central to the universe’s purpose. It’s definitely speculative and up for debate (with plenty of counter-theories out there), but it’s worth a read to stretch your perspective on the big picture

Tree is up! 🎄🎄🎄🎄

So... what is the best model if you want to generate a synthetic data? They looked at new generation, response generation, and quality enhancement. - GPT-4o excels at generating new generation - GPT-4o excels at response generation - Claude-3.5-Sonnet is better at quality enhancement

We can only hope. 😁

A month ago I promised to share the recording of my talk at Penn, "Why AI Needs the Humanities as a Partner." If you're feeling down, try the part around 14:30 where I talk about how fun it is to be a creature without a fixed essence who has to spend millennia discovering what they're capable of.

Next week we’re back to AI and startup life, but today, we’re still exploring St. Augustine. One of my favorite places in the USA.

Friends, this week marks a monumental step forward in my path towards pushing for a greater ecology of accountability in the age of AI. Thrilled to officially announce the AI Accountability Lab (AIAL) @theaial.bsky.social at Trinity College Dublin is launching today, Thursday Nov 28, 2024 aial.ie

Good morning and Happy Thanksgiving from St. Augustine. Used Apple AI to remove the orange barriers on the grass, but it did a so so job. However, it will only get better. Photo: Castillo de San Marcos, St. Augustine.

Any recommendations for the best resources to learn DuckDB (videos, blogs, etc.)? Hoping to do some digging this weekend so I can implement next week. Thanks in advance.

Talked to a friend tonight and now I’m not sure if I should be thankful or sad for not having family on thanksgiving to talk politics with.

Anthropic released an interesting thing today: an attempt at a standard protocol for LLM tools to talk to services that provide tools and extra context to be used other the models modelcontextprotocol.io

We need emergency reviewers for NAACL in the Linguistic Theories, Cognitive Modeling, and Psycholinguistics" track. Please DM.

This is one of my favorite books from this year’s reading list. Yes. I use old programming cards for bookmarks. Got thousands from my Grandfather-in-law years ago.

One of the features both early Twitter and Digg.com had was an open platform API that any developer could build awesome applications/visualizations against. To me this is one of the most important (technical) reasons @bsky.app would thrive and grow.

Concerning, not because of the obvious conclusion (that’s inaccurate), but that somehow the best next prediction logic in Gemini decided to respond with something that is obviously not statistically accurate, indicating a major failure to predict best next response gemini.google.com/share/6d141b...

Walking the dogs tonight, nature was really doing its best work.