Is prompt engineering dead? No, it’s SoTA. GPT-4 with good prompts (dynamic k-shot + self-generated CoT + choice-shuffled ensembles) beats Med-PaLM 2 on all nine of the MultiMedQA benchmarks it was fine-tuned for, without fine-tuning:
GENERATIVE AI
-
RLHF vs DPO: Training Methods for AI Models Compared
By
–
For those without context, I’m referring to this sort of thing. Competition going right now as to whether we need “RLHF” (which I think is just reinforcement learning w/ a trained reward model) or if we can do “DPO” (essentially supervised learning) no one really knows https://
x.com/yoavartzi/stat
/yoavartzi/status/1730252149370548598
… -
Leveraging Large Language Models for Enhanced Cybersecurity
By
–
Learn how leveraging large language models and #generativeAI throughout the #cybersecurity stack can help bolster digital security.
-
Foundation Model for Celebrity AI Generation and Fine-tuning
By
–
Basically, I'm a "foundation model" for celebrities.
I can be fine-tuned into Harry Potter, Johnny Depp, Snoop Dogg, Taylor Swift, or a slightly overweight version of Mark Zuckerberg. -
LLMs Automate Podcast Transcript Editing First Pass
By
–
I happily paid for transcripts at every podcast I’ve ever done and now with LLMs to do the first pass editing this will be even better.
-
Smaller Open-Source LLMs Can Outperform Larger Commercial Models
By
–
Hot take: bigger isn't always better. Especially when we're talking about #LLMs. Smaller, faster #opensource LLMs can outperform much larger and more expensive commercial LLMs when #finetuned for a task. Try fine-tuning for free: https://
pbase.ai/40YVJRH. -
NVIDIA’s Groundbreaking Evolution in AI Technology Revolution
By
–
Crucible Moments by @Sequoia takes a deep dive into our company's groundbreaking evolution. Tune in to learn about our pivotal role in the tech revolution and the future of #AI. #CrucibleMoments https://
cruciblemoments.com/episodes/nvidia -
Combining LLM self-improvement and inferred consensus
By
–
Fantastic work — imagine combining this with “LLMs Can Self-Improve”! Btw, I suggested LLM-inferred consensus in a tweet last year — AFAIK I thought of it from reading Xuezhi Wang et al. but there may be earlier examples. Case study in “Ideas are cheap; verification is hard.”
-
Fireside chat at Slush on building with OpenAI and AI future
By
–
Just wrapped up a great fireside chat at @SlushHQ with @vhankes about building with @OpenAI and the future of AI. The energy in Helsinki is electric! Meeting awesome founders has been inspiring, and the perfect way to celebrate ChatGPT’s 1st birthday!