We had two papers on this topic recently, one using a MCTS approach on formal systems (such as Lean or Metamath): https://
arxiv.org/abs/2205.11491 and one where we combine LLMs with formal provers to verify the generations: https://
arxiv.org/abs/2210.12283
PROMPT ENGINEERING
-
Recent papers on LLMs for formal mathematics and theorem proving
By
–
-
GPT’s Reliability at Repeating Its Prompt Instructions
By
–
If there is one thing GPT is very reliable at it’s for sure repeating its prompt 🙂
-
Stableboost auto-generates and customizes AI image prompts
By
–
Stableboost auto-suggests a few hundred prompts by default but you can generate additional variations for any one prompt that seems to be giving fun/interesting results, or adjust it in any way:
-
Dreambooth Stable Diffusion Finetuning Goes Viral for Personal Pictures
By
–
Dreambooth (stable diffusion finetuning for personal profile pictures) has been going viral last few days as well, for good reasons it's super fun; Unlike other places https://t.co/eIwkwiTY3o lets you play with infinite variations and experiment and play with your own prompts: https://t.co/w1SsH7UQ9R
— Andrej Karpathy (@karpathy) 7 décembre 2022Dreambooth (stable diffusion finetuning for personal profile pictures) has been going viral last few days as well, for good reasons it's super fun; Unlike other places http://
stableboost.ai lets you play with infinite variations and experiment and play with your own prompts: -
Search Accuracy Issues in LLM Entity Resolution Systems
By
–
Results are not always accurate. For example, a search about “what are LLMs” struggles with entity resolution with the law degree and large language models. Feedback for negative results could help fix this. pic.twitter.com/w1JtYehRup
— Perplexity (@perplexity_ai) 7 décembre 2022Results are not always accurate. For example, a search about “what are LLMs” struggles with entity resolution with the law degree and large language models. Feedback for negative results could help fix this.
-
LambdaAPI Giveaway: $330 Cloud Credits for Whisper Fine-tuning
By
–
Last but not least! @LambdaAPI will giveaway 330$ cloud to the 3 participants of the event. Anyone who shares their journey fine-tuning the Whisper model on twitter and tag @huggingface & @LambdaAPI would be eligible. 330$ is enough to get 300 1x A100 compute
-
Agents Now Self-Correct When Selecting Non-Existent Tools
By
–
Previously, if an LLM chose to use a tool that did not exist the Agent would throw an error @johnvmcdonnell changed it so now we tell the agent that tool does not exist and let it self correct… just one step towards making them smarter
-
LaTeX notation request substantially improves math reasoning
By
–
Anecdotally, its math reasoning (excl. arithmetic calculation) improves substantially when you request LaTeX notation.
-
AI behavior with l33tsp34k, hyphens, Zalgo text, and Base64
By
–
Less protocol-complaint but likely also dumber. You see a similar phenomenon if you ask it to speak in l33t5p34k, or with hyphens between all letters, or especially with Zalgo text. Base64 is interesting though because it probably sees a *lot* of translation pairs in pre-train.
-
Prompt injection shatters ChatGPT’s coherent personality illusion
By
–
You can prompt-inject ChatGPT into behaving however you like, but this shatters the illusion of a coherent personality. ChatGPT forces you to type your own disclaimer to prove you understand it’s not real. Otherwise you talk to the protocol droid.