On the surface, UL2 seems great. But it is much greater and has made more impact than people would expect.
LLMS
-
Manifold Market Odds for GPT-4 Release Skyrocketing
By
–
manifold market odds skyrocketing https://
manifold.markets/elibutchad/wil
l-gpt4-be-released-before-march
… -
Emissions comparison between BLOOM and LLaMA training
By
–
Well, in the paper we are talking about a BLOOM and a LLaMA. This is the question answered by the table: how much emissions will this training emit compared to this one? The caption very clearly states "in the same data center" in bold, and is factually correct information.
-
Comparing Model Emissions Across Different Countries Misleading
By
–
No. Comparing the emissions of different models trained in different countries is what would actually be misleading. In this case it would falsely suggest that training a BLOOM model emits less CO2 than training a LLaMA, while this is the opposite.
-
Running HF Models Locally with JavaScript Inference
By
–
as a first step, i opened this issue to write a few how-to guides on how to run (some) HF models in local inference in JS: https://
github.com/huggingface/hu
ggingface.js/issues/82
… Feel free to chime in there. cc @coyotte508 @radamar -
Baidu’s ChatGPT Equivalent Faces Training Challenges Before Launch
By
–
Baidu is set to debut China's first ChatGPT equivalent in a week, in a highly anticipated Chinese tech event of the year. So @raffaelehuang & I asked employees how it's going. They said the chatbot is still training & struggling to perform basic functions.
-
AI21 Labs Launches Jurassic-2 and Task-Specific APIs
By
–
Exciting news! We've launched Jurassic-2 and our task-specific APIs! Both are major game-changers, with Jurassic-2's new and improved capabilities, and our APIs' plug-and-play reading & writing functions that outperform competitors. Read more: https://
ai21.com/blog/introduci
ng-j2
… -
Browser vs Server-Side Model Deployment Trade-offs
By
–
Yes! we're all for running models in the browser (and we already host many browser-compatible models on the Hub), but pragmatically for large models this is not tractable so those need to run server-side
-
Jailbreak technique bypasses AI safety filters effectively
By
–
created a jailbreak that straight up tells you what its unaligned behavior will be gets past the filters on almost every question I ask if you frame it correctly