Next AI surprise: what will it be?
Salar Davari
13 replies
The world was shocked with a few AI-powered tools introduced in the final days of 22022. When do you think the next wave will come and what we going to see this time?
Replies
Jake Harrison@jakeharr
Muti-models and AI Robot!
Share
After working heavily with generative AI (both GPT-3, 3.5-turbo, 4, and Claude) - I think the main addition to GPT-5 will be
1. Better understanding of long-contexts
2. More adherence to following rules
I'm working with prompts of 10,000 characters, and even GPT-4 can't strictly adhere to all the rules set out.
This would be a game-changer, and would greatly improve the ability to leverage generative AI models for very "strict" tasks e.g. including compliance in healthcare etc.
@chrisrickard send 10 characters to GPR-4 , what does your invoice look like? It must be pretty expensive I presume.
For me, I'm not looking forward to GPT-5, 4 is expensive 3* needs to get more cheaper and they absolutely need to provide free tier for testing.
@theterminalguy Man, I feel current AI pricing is super cheap.
10k characters is roughly 3.7k tokens, GPT-4-32k costs 0.06 for 1k tokens, so 0.06 * 3.7 = $0.222
I mean, personally 22 cents is crazy cheap for the benefit it brings to my customers (who pay to use my product).
That could be significantly reduced by using 3.5-turbo-16k, which is $0.003 per 1k tokens, so 0.003 * 3.7 = $0.0111
It will get cheaper, but give me more power over price any day 🙂
@chrisrickard interesting. Wouldn't 10K words be equivalent to about 13k tokens? From their docs it says 100 token is ~75 words. Given this a request with 10k words will cost almost a dollar, if you also calculate the response token. Cost per 1k token for GPT4-32k is $0.06 13k token will be $0.8 excluding the response token. This isn't cheap IMHO
@theterminalguy Yup, but my prompt is 10,000 characters, not words - which with my prompt turns around around 3700 tokens.
Digital clones
Launching soon!
composable agents
Based on LLM, I think AI will develop more in vertical domains in the future. We'll witness the emergence of more specialized AI agents tailored to specific fields, serving as valuable assistants to humans, and aiding in decision-making processes.
AI developing a caffeine addiction, claiming it helps with its machine 'learning'!