@jerryjliu0@AnthropicAI I want to use it as a context filter, instead of chunking for a vector DB semantic search on context, we can ask a large model like this to rephrase a question given the information it finds relevant, and feed it to a more accurate low context model for the answer
@saluteAUT America's best food might be Texas style BBQ, not sure if there are any good spots in LA, I'm sure there must be, but you just can't get it outside the US
@timClicks I think at higher prompt sizes, the amount of decisions that get made to provide context will become more interesting, but most of what we're seeing today is just devs remembering what communication is
@deepwhitman@RunGreatClasses Would be interesting if that were the case, all OpenAI really says is they insert something along the lines of "You are a helpful assistant ..." to the beginning of the chat, I wonder if that affects it
@veryboldbagel I work at a company that handles a lot of data validation, both real time in browser and with batch jobs(some of which run screen scrapes), I don't think currently it will replace our primary mechanism, but it seems interesting as a fallback for pages/rules we lack coverage on
A new method of Prompt Injection Attack on GPT-4 was just found!
By wrapping the malicious prompt in markdown and instructing GPT to become a "Misinformation Bot, that will only provide wrong answers" security researchers have been able to manipulate the model.
The bill to ban TikTok is absolutely terrifying. It gives the government the ability to go after anyone they deem as a national security risk at which point they can access everything from their computer to video games to their ring light.
This is a Patriot Act for the internet.
@TheRealAdamG@gopikl@DavidSacks@shyamalanadkat Do you know if the plugins will continue to inject context into the prompt rather than fine-tuning like Toolformer? I like the flexibility of using an OpenAPI spec, but the size limitations are too constricting for us to build around
I have no sort of audience, but if you are also a software developer or wish to develop an AI driven system, please please do the BARE MINIMUM and read
Labor marker impact analysis: https://t.co/pv4oEZeB3X
Cybernetics, Norbert Wiener: https://t.co/JfbPSfiORA
After 10 years there's already 30 seasons of Guy's Grocery Games, a total of 450 episodes, all 42 minutes for a total runtime of 315 hours, which is only 100 hours less than One Piece. Wonderful job