for those who haven't heard about this:
openAI only gives you the top 5 logprobs per response (max). we (mostly @justintchiu
) came up with an algorithm to get logits from openAI. essentially for a different token (not top-5), u can binary search logit bias to see the minimal
LLMS
-
Extracting Full Logits from OpenAI API via Binary Search
By
–
-
Language Model Inversion: New Research and Open Source Tools
By
–
paper (language model inversion) http://
arxiv.org/abs/2311.13647 github (open_logprobs) -
Language Model Inversion Attack Research Paper
By
–
language model inversion http://
arxiv.org/abs/2311.13647 -
OpenAI API Changes Due to Language Model Inversion Research
By
–
achievement unlocked: OpenAI changes its API because of your research you can still get logits from GPT-4 by using the argmax bisection method detailed in Language Model Inversion. it's pretty expensive, though we should have stockpiled logits while we had the chance…
-
ChatGPT introduces new DALL-E image style and aspect ratio controls
By
–
ICYMI: ChatGPT is A/B testing additional DALL-E controls which help enhance your prompts. It comes with a selection of different image styles which you can preview, as well as an aspect ratio control. https://t.co/oDwkUzuFsS pic.twitter.com/ODqtZpXUiZ
— 🚨 AI News | TestingCatalog (@testingcatalog) 7 mars 2024ICYMI: ChatGPT is A/B testing additional DALL-E controls which help enhance your prompts. It comes with a selection of different image styles which you can preview, as well as an aspect ratio control.
-
LoRA Land’s Mistral-7B Models Cut Inference Costs Significantly
By
–
LoRA Land's 27 fine-tuned #opensource #mistral-7b #LLMs no only rival #GPT-4, but they're also very cost-effective! Assuming 2M tokens (90% input; 10% output) per adapter per day, you can cut inference costs by 45%-65% vs. GPT-3.5 Turbo and 97%-98% vs. GPT-4 Turbo
-
Groq Reports Strong Demand Across Enterprise and Developer Segments
By
–
"We’re seeing a ton of demand across the board, from Fortune 50s all the way to individual developers… a wide variety of use-cases including AI copilots, financial trading platforms & content generation." — @lifebypixels in @cerebral_valley
: -
Observations on Sora video platform development and upcoming release
By
–
In case you are wondering: the Sora video platform is WIP Currently, its subdomain exposes 2 URLs:
– /new – /favorites Feels like we will get a totally standalone tool which is not a part of ChatGPT which we already know What will come first? Sora or GPT-5? -
Prompt Length Strategy: Short Daily Use vs Detailed Consistency
By
–
Yep. I find for day-to-day stuff my prompts are shorter. I go 'mega' when I need a more consistent and repeatable output but often get away with a medium-sized prompt (1-2 paragraphs) otherwise.
-
Large Multimodal Datasets: Unsung Heroes Behind AI Breakthroughs
By
–
The unsung hero behind many breakthroughs (including Stable Diffusion, and GPT-4) is large multimodal datasets. Last October, Assistant Professor Ludwig Schmidt from the University of Washington introduced DataComp, an approach to enhance data quality.