A bunch of years ago I started making my own beer with my friends Eric and Rhett (of whom the latter is an old school award-winning brewer that got started in the 80’s and now owns a brewery in VA Beach), and the former being a nutjob that would throw whatever was in the kitchen … Read more
LL3 debuts on April 18th. It feels like a year since then as far as progress. There was a 4bit bnb quant out in 4 hours after the release. Meta and the model wall… What’s the point? The second one person gets access the model is essentially public… Very large context sizes and fine tunes … Read more
Long-term digital archiving is actually much more involved than it appears. On one hand, you might think you can simply make a copy of something and store it indefinitely. On the other, it would be naïve to consider this a secure long-term storage method. Numerous natural factors, including magnetism, gravity, and environmental conditions (temperature, moisture, … Read more
Purpose built models that are part of an ecosystem of models that have certain knowledge. You don’t buy a dairy farm just to make mac and cheese. Why use a 1 trillion parameter model to ask it how long to boil an egg for. Nothing new here but… I think we need to break apart … Read more
One major problem with machine learning and the resulting output is the dataset. So many models have so much junk in them because the data wasn’t clean or processed correctly during training, or data was included that wasn’t necessary, contradictory, redundant, or otherwise unhelpful or misaligned for training. LLM’s generally get trained on so much … Read more
Real-time ingestion via NWS API’s (so many). Correlate NDBC, NWS, WPC, NHC, and whatever else you want to alert on any type of severe weather. Use cases:
Ok maybe but- for-profit companies like Google and Microsoft developing regulation ..not so much. Regulating the internet has failed, why bother regulating AI, especially when it’s growth has been fueled by the internet and it’s dependencies. Regulating AI (and models especially) almost guarantee creating closed source. closed source means $$$ for someone so as usual, … Read more
For triage having multiple subject matter expert bots may be excellent instead of asking questions to one (huge/slow/expensive) model. I’m not saying the best idea is to use a bot to diagnose head trauma, this is just an example. Additionally, you can update specific information per subject as things change (new solutions, guidance, etc). For … Read more
Ingested a few hundred hours of Car Talk and got the bot answering as a team. I think this has a lot of potential, especially if you’re training on a lot of knowledge. This turns into a conversational output and it’s pretty funny too. Also captures their personalities while talking about other topics. These are … Read more
<rant> Generally, the current thought is to train a model on a ton of data- broad data, usually. I think the idea is to turn it into a human. But is this really what we need. I see modular models A model that knows all Maths A model that knows all Art and etc and … Read more