wangza @wangza

A new RCT on banning smartphones in the classroom - marginalrevolution.com/marginalrevolu…

Helped my teenage son get setup with an account at @River recently so he could save some of his income from his summer job in BTC. He was stoked to receive this in the mail the other day - very cool! @WalkerAmerica @Thebitcoinway_ @TheSurvivalPodc @theothermjordan @_DannyKnowles

Google loves keeping valuable websites hidden from you. Here are 11 cool websites you probably didn’t know existed (until now):

Conclusion after my trial use of the much vaunted (and overhyped) DeepSeek's AI platform: this chatbot is simply just an AI version of the PRC Ministry of Foreign Affairs. 🙃

One of the great memos of the LLM era, very much bears rereading. So many investors read it and still invested a sky-high valuations anyway. Can’t say people like me and the unnamed memo writer didn’t warn them.

Haider. @haider1

a year ago

FROM THE PAST 🤯 remember 2 years ago, there was a leaked internal Google document claiming open-source AI will surpass google and openAI how true is this today?

keyboard_arrow_left Previous keyboard_arrow_right Next

17 19 159 22K 38

Perplexity is one of the most critical metrics when designing large language models, yet most people can't explain how it works. Perplexity quantifies how uncertain a model is when predicting the next word in a sequence. It tells us how many options the model considers on average when making predictions. For example, if a language model has a perplexity of 10, it means that the model is choosing between 10 plausible options for the next word at any given step. The lower the perplexity, the better the model's predictions align with the actual text. A good way to think about perplexity is to look at how "perplexed" a model is at any given time. When the model is confident about what comes next, it will have fewer choices and a lower perplexity score. When the input is ambiguous or unfamiliar, the model's uncertainty increases, raising the perplexity score. You'd never want to trust a model with high perplexity, so it's a great metric to decide whether you want to ask humans to step in before using the model's output. Fortunately, it's very simple to compute the perplexity of a model, so we can use it during training to monitor how well a model is learning. The metric is also easy to understand, explain, and interpret. Keep in mind that perplexity doesn't measure accuracy. A model can have low perplexity but still make incorrect predictions. For instance, it might confidently predict the wrong word. We can't use perplexity to compare different models unless they use the same tokenization, dataset, and training methods. A model's perplexity score depends heavily on these factors, making it less useful as a universal benchmark. Finally, perplexity doesn't capture a model's ability to understand long-term dependencies in text. It evaluates predictions token by token, missing the bigger picture in complex language tasks.

Here is a link to the book: bit.ly/49lBHVq A few notes about the book: 1. It's a good mix of theoretical concepts, code, and practical case studies. 2. You'll get a ton of time building real-world RAG pipelines. There's plenty of practice here. 3. The main focus of the book is on balancing performance and costs. I love books that acknowledge that "the best solution" is full of tradeoffs.

at a panel session organised by B2B sales with boring corporate exec attendees, only saving grace was the (other) startup founder on the panel

(Sorry I botched the name a bit) Cursor editor: cursor.com Get pro for $20, then in Cursor settings select Sonnet 3.5. Then watch all the videos on how to use and practice. (I think both the setup above and the usage is somewhat beginner unfriendly, maybe someone can link to good videos / guides)

One of these currencies is not like the others. Sesame Street on #bitcoin

The push for overt financial slavery takes another step here in Thailand.

I'd like to have a real conversation about whether AI is a risk for human extinction. Honestly, I don't get how AI poses this risk. What are your thoughts? And, who do you think has a thoughtful perspective on how AI poses this risk that I should talk to?

Ep. 7 of BOB Radio is here ⬇️ youtu.be/mJF_r4HsNs0 Our special guest: Mr. Albert Buu, (@subnetmask) Founder and CEO of @neutronpay recorded straight from Da Nang where they hosted @LightningconVN. Are you ready for Hyperbitcoinization in SEA?

A Masterclass in Marketing 4 years of Marketing Degree in 7 minutes One of a kind. Bookmark and come back to this often.

BOB Radio ep. 8 is here!! Our special guest: Mr. @louishliu, Founder & CEO of @MimesisCapital, possibly the world's first #Bitcoin family office. How Louis innately understood scarcity? Why did he start Mimesis? What did he invest in? What's next? 🔥🔥 youtu.be/Dj0yy7PBX4U

No need to endlessly scroll on your streaming platform—you can now watch all #FutureForward short films right now! Experience the inspiring journeys of people and companies taking bold climate action, right from your living room. Check it out: theclimatepledge.com/future

Let's start with this prompt from Midjourney V4 This is already quite a verbose prompt for MJV4. Now let's run it through my little magic ChatGPT prompt. Hold on to my beer.

keyboard_arrow_left Previous keyboard_arrow_right Next

Make custom videos with just text and your input video. He cleverly used @LumaLabsAI and @runwayml's Gen-1 for this (great work) The results are amazing.

#BOBtagged Book Thief Cafe Chiangmai: Legit coffee and even better ambiance.

keyboard_arrow_left Previous keyboard_arrow_right Next

Good gem 💎

Piccolo ⛏️⚡️ @PlebPiccoloJr

4 years ago

Best resources to learn how to #Bitcoin - for all stages of development - bitcoiner.guide - lopp.net - learnmeabitcoin.com

3 22 71 0 24