In the largest-of-its-kind evaluation, we found that GPT-4 provides, at most, a mild uplift in biological threat creation accuracy (see dark blue below.) While not a large enough uplift to be conclusive, this finding is a starting point for continued research and deliberation.
LLMS
-
Early Access Model Leaked: Company Retrains from Llama 2
By
–
An over-enthusiastic employee of one of our early access customers leaked a quantised (and watermarked) version of an old model we trained and distributed quite openly. To quickly start working with a few selected customers, we retrained this model from Llama 2 the minute we got
-
Multilingual Text Embedding Inversion Attacks on Language Models
By
–
just read a cool follow-up to vec2text: "Text Embedding Inversion Attacks on Multilingual Language Models" these folks extend embedding inversion to the *multilingual* setting, where we might not know the language of the encoded text ahead of time they add a
-
Neuralink brain chip, Bard video summaries, Tencent LLM advances
By
–
Top stories in AI today: -Neuralink places first brain chip implant in human
-Yelp launches new AI features
-Summarize a YouTube video or podcast using Bard
-Tencent details multimodal LLM advances
-6 new AI tools & 4 new AI jobs Read more: http://
therundown.ai/p/neuralinks-t
elepathic-human-milestone
… -
ChatGPT ‘About’ pages are now accessible to all users
By
–
A minor ChatGPT update: Custom GPT's "about" pages are now accessible to all users. These About pages contain GPT descriptions and Chat CTA and can be also browsed by non-Plus accounts. h/t @_devalias
-
AI LLM Retrieval Evaluation with LlamaIndex
By
–
Link to blog: https://
srk.ai/blog/004-ai-ll
m-retrieval-eval-llamaindex
… -
RAG Encoder and Reranker Evaluation with Open-Source Models
By
–
Inspired by @ravithejads
, wrote a simple blog post on "RAG – Encoder and Reranker evaluation" using @llama_index on a custom QA dataset. Used open-source embeddings and re-rankers for the evaluation
"JinaAI-base" embedding with "bge-reranker-large" gave the best result among -
RAG and expert AI orchestration
By
–
Even simpler! 🙂 It's based on RAG (Retrieval-Augmented Generation) and an orchestration system we built that calls multiple small expert AIs, each specialized in their own domain. 🙂 A bit like the analogy OpenAI published a few months ago. 🙂
-
Five Deep Topics About AI, Research and Ethics
By
–
What are five topics you can talk about for 30 minutes with zero prep 1. information content of text embeddings
2. architectural mysteries of transformer language models
3. the meta-research process (how & why to pick problems)
4. whether the NBA is rigged
5. shrimp suffering -
Custom GPTs, Bard vs GPT-4, and CodeLlama70b Updates
By
–
I can't tell if AI news has been really slow for the past couple weeks or if I'm just so "in the thick of it" that I'm hard to impress these days. LOL Topics I'm exploring right now…
– You can @ tag custom GPTs
– Bard is outperforming GPT-4 (but not GPT-4 turbo)
– CodeLlama70b