kevinh456

Would love an English translation that makes sense.

That’s because it is, friend.

I anxiously await the 34th report of bribery.

Github bro. Github.

Fable, the same model as mythos with extra safety controls, was much faster, more accurate, and more token efficient than previous models. What I got done with it in 48 hours accelerated my personal project from concept to deployed prototype.

Kinda cool but it would be great if you made a couple changes…

The ui is confusing. It’s hard to get back to the main ui as you navigate around.

I couldn’t find any stuff. Your app needs to “show me the money shot” when it loads so I see what it’s like and useful for. Even if something isn’t nearby, I need to see the whole interaction loop of the app.

Have you considered partnering with local freecycle groups? There’s whole communities doing this already

Lol. Sorry. This will not work.

You can't control the way apps cache their data. Each one is going to load and cache its data in its own way. There is no "data to download" like you're thinking.

DL is a definition list not a description list.

Burn baby burn. Token inferno.

Captchas are not effective. You can pay 2captcha less than a penny per captcha and humans solve them for you.

Tell me how you feel about DST.

You anchor on what "AI" actually is. "Artificial Intelligence" is, IMO, useless phrase. Make them clearly define what they're talking about.

AI is a bunch of different technologies that have many uses—neural networks, natural language processing, OCR, speech recognition, machine learning, computer vision, image classification, upscaling models, and our favorite new friends "generative pre-trained transformers" (GPT) and "large language models" (LLM) that make up key parts of "generative AI."

Once you make them specify what they're talking about, you talk about the nature and inherent limitations in the technology.

I like to call GPT and LLM "statistical binary string predictors." IE: given a string of binary, predict the expected binary string based on the inputs. It's an amazing technology, don't get me wrong. We're starting to see the limits already though.

Limited context windows. Larger context/training == lower quality results. More input tokens = lower quality. In some respects, newer models are now regressing from earlier models because they're chasing benchmarks and not the real world use cases.

Start to dive into the details. Ask them to admit the problems with LLM and GPT. Ask them how they see these problems getting resolved. Most AI fanboys don't understand the technologies involved. Expose it.

It took way too fucking long to figure out what they’re doing. Maybe they should have paid for pr or marketing or something.

How did this ai written crap get so far up on hacker news.

Shopify does somehow to this day.