At higher temperatures (>1.25) GPT-4's writing will often dissolve into nonsense, and that risk increases the longer it writes. You can also use the top-p parameter in tandem (try 0.9 to 0.99) to get more creativity without it going completely off the rails.
LLMS
-
Tuning AI Temperature for Response Diversity and Weirdness
By
–
Is this useful? Sometimes! Temperatures up to ~1.25 can help when you *really* need diversity in short responses to the same prompt, or if you want writing that's deliberately weird. But for most tasks, even temp=1.0 is too high unless you also reduce top-p.
-
Cerebras Enables 50k Context Windows for AI Models
By
–
Leveraging context windows up to 32k is an accomplishment that only a handful of companies can enable. At Cerebras, we empower everyone from beginners to experts with the ability to leverage context windows up to 50k with just a few clicks! Learn more :
-
GPT-4 Closed-Source Value: Boosting Developer Productivity Tenfold
By
–
GPT-4 has awesome value even as a closed-source solution. It provides proof of existence and will motivate other groups. For $20/mo (and some caution), developers can 10x their speed of producing high-quality code.
-
OpenAI Open Sources Evals Framework for GPT-4 Model Evaluation
By
–
Less publicized but highly awesome aspect of GPT-4 launch was that OpenAI open sourced an evals framework, allowing us to crowdsource model evaluations at scale . The repo is getting some very high quality PRs (rewarded with GPT-4 access). I <3 evals; `pip install evals` https://
x.com/kondrich2/stat
/kondrich2/status/1635687409668227072
… -
GPT-4 API Access Expansion Announced
By
–
Hearing about people getting access to GPT4 API… pic.twitter.com/1wJpJVXYSJ
— Yohei (@yoheinakajima) 16 mars 2023Hearing about people getting access to GPT4 API…
-
Talking Language AI #6 Conference Announcement Coming Soon
By
–
We are 1-week away from Talking Language AI #6 with @davlanade and @JayAlammar
! Sign up now, https://
info.cohere.ai/talking-langua
ge-ai-06
… -
GPT-4 as a Multiplier: Amplifying Engineering Capabilities
By
–
GPT-4 should be thought of as a multiplier to your capabilities. If you are a 1x engineer, you can become 10x. If you are a 0x engineer, you will stay at 0 or may even go negative. I just asked GPT-4 to write OpenCV code for Photometric Stereo, and it produced beautiful code.
-
GPT-3.5-Turbo Praised for Academic Research with 10x Price Reduction
By
–
I also think gpt-3.5-turbo is great for academic research, with a 10x reduction in price
-
Llama Model Enables Emergence Research with Strong Performance
By
–
And by the way, I think Llama is great for enabling emergence research to be done, strong performance and multiple model sizes.