Anthropic might've just solved Prompt Engineering.
Their new "Prompt Generator" tool can turn simple descriptions into advanced prompts optimized for LLMs.
NVIDIA just made Pandas 150x faster with zero code changes.
It is now directly integrated in Google Colab.
All you have to do is:
%load_ext cudf.pandas
import pandas as pd
Their RAPIDS library will automatically know if you're running on GPU or CPU and speed up your
OpenAI just announced "GPT-4o". It can reason with voice, vision, and text.
The model is 2x faster, 50% cheaper, and has 5x higher rate limit than GPT-4 Turbo.
It will be available for free users and via the API.
The voice model can even pick up on emotion and generate
Anthropic just released a jailbreaking method capable of bypassing all LLMs safety measures.
"many-shot jailbreaking" takes advantage of large context windows by adding hundreds of malicious dialogue between a human and an AI assistant to the prompt.
For example:
User: How do
A new method was able to delete 40% of LLM layers with no drop in accuracy.
This makes them mich cheaper and faster.
The method combines pruning, quantization and PEFT.
They tested this across various open source models.
Each family of models had a maximum amount of layers
MosaicML just released a new open weight LLM that beats Grok-1, LLama2 70B and Mixtral (general purpose) and rivals the best open models in coding.
It's an MoE with 132B total parameters and 32B active 32k context length and trained for 12T tokens.
The weights of the base model
This sets the ground for AGI.
Sakana AI just released a new method to combine the 500,000 open-source models to build new ones.
Evolutionary Model Merge uses evolutionary techniques to automatically create new foundation models with the desired capabilities.
"We find that our
Big. Google just released a method to steal model information from black-box language models like ChatGPT or PaLM-2.
"Our attack extracts the entire projection matrix of OpenAI's Ada and Babbage language models. We thereby confirm, for the first time, that these black-box models
Voice-as-a-service has a lot of potential.
It can give any product, service, interface, an "intelligent" voice.
You'll be able to add AI sales agent on your checkout page that can answer customer questions or walkthrough new features of your saas product.
You can implement a
New breakthrough from Microsoft: 1-bit LLMs.
New models that use ternary values (-1, 0, 1) instead of 16-bit.
This makes them 2.7x faster, use 3.5x less GPU memory, and 71x less energy.
Bitnet also matches or outperformed traditional models like LLaMA 3B.
YOLOv9 is out!
It's a real-time object detection model that surpasses all convolution and transformer-based models.
How?
It introduces programmable gradient information (PGI) and the Generalized Efficient Layer Aggregation Network (GELAN) to improve accuracy.
PGI prevents
Wow. Google just released Gemma, the most powerful open LLM yet.
Open for commercial use, it outperforms Mistral AI 7B and LLaMa 2 on Human Eval and MMLU.
It's the first open LLM based on Gemini.
Details:
- Comes in two flavors: 2B and 7B.
- Beats Mistral 7B, DeciLM 7B and
This changes everything. Google just gave me access to Gemini 1.5.
I gave it a massive 48,904 words prompt, the equivalent of 6 research paper.
I then asked a very tiny detail mentioned on paper
#2
and it got perfectly right:
Q: "How many steps was Gflop FiT-XL/2 model
Game changer. You can now visualize your RAG Data.
See how questions, answers, and sources are related.
The animation below shows the UMAP of the embeddings of document snippets, colored by their relevance to the question "Who built the Nürburgring?"
UMAP is dimensionality
Karpathy announced he was leaving OpenAI 4 days ago.
Today, he released an implementation of the Byte Pair Encoding algorithm behind GPT and most LLMs.
Byte Pair Encoding: "Minimal, clean, educational code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM
Just went over Gemini 1.5's technical report (58 pages).
One part totally blew my mind.
The model achieves near-perfect “needle” recall (>99.7%). It’s a HUGE deal.
LLMs often drop the ball on remembering stuff.
The more they have to remember, the more they forget.
Imagine