As the scale of models increases, the computational effort needed for training grows rapidly. Check out a pair of new approaches that can significantly improve model performance through thoughtful pre-training and fine-tuning with only minimal compute. https://
goo.gle/3UicYbD
LLMS
-
Efficient Pre-training and Fine-tuning Approaches for Large Models
By
–
-
GPT-3 Improved Rhyming Generates Green Day Song Lyrics
By
–
The new version of GPT-3 is significantly improved at rhyming! Here, I use it to generate the lyrics for a new Green Day song.
-
CICERO by Meta AI: Fluent Conversations with Multiple Diplomacy Players
By
–
What impresses @ml_perception the most about #CICERObyMetaAI is its ability to hold fluent, human-like conversations with up to Diplomacy players simultaneously. Learn how we created an interface between language and strategic reasoning models:
-
Chains for User Input and Memory in LangChain
By
–
Yes to both! I think they are probably slightly different things, ones getting it know if it needs more info (adding in a “user input” chain) and another is getting it to keep the overall goal in mind (memory probably good for this)
-
LangChain 0.0.24: Logging Improvements and Advanced AI Models
By
–
LangChain version 0.0.24 Add chain name for better logging (@Hansenq)
Bump default @OpenAI model to `text-davinci-003` (!!!) (@shyamalanadkat)
PaL (Program-aided Language Models) Chain (!!!) with help from @aman_madaan et co. https://github.com/hwchase17/langchain
… -
Prompt Injection Vulnerability Persists in GPT-3 Models
By
–
So prompt injection is still a thing with the new text-davinci-003 GPT-3 model… but not sure how much this matters. cc: @goodside
-
OpenAI Announces New Version of GPT-3 Language Model
By
–
@OpenAI announces a new version for its GPT-3 language model https://actuia.com/actualite/open ai-officialise-une-nouvelle-version-pour-son-modele-de-langage-gtp-3/ … #AI #ArtificialIntelligence
-
Zero-shot style requests less controlled; retry stylized ‘step-by-step’ examples.
By
–
Zero-shotting style requests works somewhat now, but still not as controlled as using demonstrations. Need to retry some of my stylized "Let's think step by step" examples on the new model.
-
GPT-3 explains the Black-Scholes formula in an uwu voice
By
–
New GPT-3 version (text-davinci-003) explains the Black-Scholes formula in an uwu voice:
-
GPT-4 Leap: Early 2023 Startup Opportunities
By
–
GPT-4 will be as big of a leap from GPT-3 as GPT-3 was to GPT-2. Early 2023 is going to be wild. I can't wait to build with it and invest in the best startups who leverage it (culture 2.0, diverse, user-first, intuitive design, strong business sense, brilliant team).