I’ve messed around with GPT-3 and other language models, and while they’re incredibly impressive at generating text or answering questions, you can tell they’re not “thinking.” They’re just really good at predicting what comes next based on patterns in their training data. For example, I’ve tried asking them stuff that requires reasoning or connecting ideas in a way that isn’t obvious, and they either mess it up or give something that sounds smart but doesn’t really make sense when you think about it. It’s not AGI—it’s more like a really advanced autocomplete. AGI would need to handle totally new problems and learn like humans do, not just regurgitate patterns.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
GPT 3 is bad. Use GPT4, Sonnet 3.5, or DeepSeek V3. Or use the reasoning models
Might also need to properly structure your prompt
They are not perfect but in my experiments for simple use cases they are good enough. Also be mindful with how these models work with tokens. Anything that involves maths is fundamentally tricky for LLMs due to their architecture
I just see the prompt as some sort of juiced up google it will give you a vague road to follow or a refresher if you already worked with the material or knowledge. So instead of prompting google and having to skip ads hoping to click the right link you probably get more stuff to fine tune your search by prompting a SOTA LLMs.
Having typed, im not wondering if the younger gen of people replace "I will google it" with "I will prompt it"
Having typed, im not wondering if the younger gen of people replace "I will google it" with "I will prompt it"
Aider bench is the benchmark not aidanbench so the above post is confusing.
All we can be certain of in the global competition is that Google has become an incredibly strong contender after initial failures, and China is “close enough.” It’ll be interesting to see what Open AI does given their disappearing moat, and Anthropic is similarly screwed.
On a closer look this is a fake benchmark. Aider bench is the benchmark.
Average OpenAI engineer (genius IQ, international math and physics olympiad gold medalist, )
vs
Average DeepSeek engineer (random new grad PhD from China T10 uni, likes AI)
To the absolute surprise of nobody, OpenAI officially announces for profit company:
It got to reap the benefits of nonprofit status for nearly a decade and then convert to a for profit.
vs
Average DeepSeek engineer (random new grad PhD from China T10 uni, likes AI)
To the absolute surprise of nobody, OpenAI officially announces for profit company:
It got to reap the benefits of nonprofit status for nearly a decade and then convert to a for profit.
Last edited:
OpenAI is done for, its Googles game to lose