GPT Reviews

OpenAI's Data Partnerships, where organizations collaborate to create diverse training datasets for AI models. They also explore the massive funding of $500 million into German AI startup Aleph Alpha. Additionally, they review three cutting-edge research papers on topics such as open-world multi-task agents, dense prediction, and multimodal autoregressive models.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:29 Openai Announces the OpenAI Data Partnerships

04:11 Lidl owner and Bosch Ventures co-lead $500M Series B into German AI startup Aleph Alpha

06:10 Tweet by Jerry Liu

08:05 Fake sponsor

10:11 JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models

11:54 PolyMaX: General Dense Prediction with Mask Transformer

13:33 Mirasol3B: A Multimodal Autoregressive model for time-aligned and contextual modalities

15:18 Outro

The launch of Humane's AI Pin, a wearable brain powered by OpenAI, and Amazon's ambitious AI model codenamed 'Olympus.' The team also explores interesting internet finds, including a blog post about being acquired by Google. They review research papers on the appropriate use of evaluation benchmarks for large language models, the rule-following ability of LLMs, and learning unsupervised world models for autonomous driving.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:34 Humane officially launches the AI Pin, its OpenAI-powered wearable

04:07 Amazon dedicates team to train ambitious AI model codenamed 'Olympus' -sources

06:26 What I learned being acquired by Google

07:52 Fake sponsor

10:41 Don't Make Your LLM an Evaluation Benchmark Cheater

12:15 Can LLMs Follow Simple Rules?

14:07 Learning Unsupervised World Models for Autonomous Driving via Discrete Diffusion

15:53 Outro

ChatGPT outage and the increasing demand for OpenAI's services. They also explore Meta's new policy requiring political ad campaigns to disclose their use of AI. In the "Random Reads" segment, they highlight the ArXiv AI paper summarizer, Bulletpapers, which won the Anthropic Hackathon. The hosts then delve into three fascinating research papers: Neural MMO 2.0, OtterHD-8B, and leveraging large language models for automated proof synthesis.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:23 ChatGPT is back online after a 90-minute ‘major’ OpenAI outage

03:28 Meta will require campaigns to disclose use of AI in political ads

04:51 ArXiv AI paper summarizer, won Anthropic Hackathon

05:08 Fake sponsor

07:25 Neural MMO 2.0: A Massively Multi-task Addition to Massively Multi-agent Learning

08:05 OtterHD: A High-Resolution Multi-modality Model

09:37 Leveraging Large Language Models for Automated Proof Synthesis in Rust

11:31 Outro

First-ever AI-negotiated contract, Baidu's shift to Huawei chips, and Meta's decision to bar political advertisers from using generative AI ads tools. The team also reviews three cutting-edge AI research papers, including CogVLM's visual language model, Google DeepMind's framework for classifying AGI models, and GateLoop's data-controlled linear recurrence model.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:13 An AI just negotiated a contract for the first time ever

03:40 Baidu placed AI chip order from Huawei in shift away from Nvidia

04:59 Meta bars political advertisers from using generative AI ads tools

06:20 Fake sponsor

08:13 CogVLM: Visual Expert for Pretrained Language Models

09:54 Levels of AGI: Operationalizing Progress on the Path to AGI

11:10 GateLoop: Fully Data-Controlled Linear Recurrence for Sequence Modeling

12:57 Outro

OpenAI's new models and developer products, Microsoft's partnership with Inworld AI for AI characters in Xbox games, and explore thought-provoking research papers on Kolmogorov neural networks, learning from mistakes in large language models, and prompt injection attacks.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:38 New models and developer products announced at DevDay

04:08 Microsoft is bringing AI characters to Xbox

06:08 AI App Graveyard (dang.ai)

08:17 Fake sponsor

10:38 On the Kolmogorov neural networks

12:02 Learning From Mistakes Makes LLM Better Reasoner

13:19 Tensor Trust: Interpretable Prompt Injection Attacks from an Online Game

15:21 Outro

Elon Musk's new AI venture, xAI, and its chatbot Grok, as well as Andrej Karpathy's experiments with ChatGPT's "Advanced Data Analysis" tool. They also explore AI policy developments and the geopolitics of AI. The team reviews three cutting-edge AI research papers on pretraining data mixtures, idempotent generative networks, and fast language-audio pre-training.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:07 Elon Musk debuts ‘Grok’ AI bot to rival ChatGPT

03:57 Andrej Karpathy Tweets

05:39 Your guide to AI: November 2023

07:41 Fake sponsor

10:27 Pretraining Data Mixtures Enable Narrow Model Selection Capabilities in Transformer Models

12:02 Idempotent Generative Network

13:31 FLAP: Fast Language-Audio Pre-training

14:59 Outro

Cohere's Embed v3, a language model that improves search applications and retrieval-augmentation generation systems. They also explore Apple's plans to incorporate AI into their Apple Watch for advanced health monitoring, including hypertension and sleep apnea detection. Additionally, they discuss a joint statement on AI safety and openness, emphasizing the importance of transparency and broad access in AI governance. The team also reviews three fascinating research papers, covering topics such as chip design, multimodal human-AI interaction, and real-time reconstruction of visual perception using brain decoding.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:23 Cohere Releases Embed v3

03:58 Apple Plans Hypertension, Sleep Apnea Detection for Next Watch

05:41 Joint Statement on AI Safety and Openness

07:36 Fake sponsor

10:40 ChipNeMo: Domain-Adapted LLMs for Chip Design

12:33 LLaVA-Interactive: An All-in-One Demo for Image Chat, Segmentation, Generation and Editing

14:05 Brain decoding: toward real-time reconstruction of visual perception

15:56 Outro

DeepMind's upgraded AlphaFold AI system, which can now predict the 3D structures of proteins, DNA, RNA, and more. They also explore Quora's Poe, an AI chatbot creator economy that allows bot creators to earn money. Additionally, they touch on Yann LeCun's warning about the concentration of AI power in the hands of a few companies. The episode also features reviews of research papers on pretrained models for computer vision tasks, offline reinforcement learning, and text rendering strategies for pixel language models.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:37 DeepMind updates AlphaFold: A glimpse of the next generation of AlphaFold

04:01 Quora’s Poe introduces an AI chatbot creator economy

05:35 Opinion: AI one-percenters seizing power forever is the real doomsday scenario, warns AI godfather

07:28 Fake sponsor

09:30 Battle of the Backbones: A Large-Scale Comparison of Pretrained Models across Computer Vision Tasks

10:52 Unleashing the Power of Pre-trained Language Models for Offline Reinforcement Learning

12:32 Text Rendering Strategies for Pixel Language Models

14:01 Outro

BlueSea Frontier Compute Cluster, a floating AI paradise with unparalleled compute power. They also explore the Phind Model, which beats GPT-4 in coding tasks, and the creation of a game called "Angry Pumpkins" using AI technology.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:16 BlueSea Frontier Compute Cluster: The Future of AI Innovation

04:08 Phind Model beats GPT-4 at coding, with GPT-3.5 speed and 16k context (phind.com)

06:20 Building Angry Birds with AI

08:22 Fake sponsor

10:44 OpenCog Hyperon: A Framework for AGI at the Human Level and Beyond

11:56 Multimodal ChatGPT for Medical Applications: an Experimental Study of GPT-4V

13:53 A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learning

15:39 Outro

President Biden's executive order on safe and trustworthy AI, the release of a "final" Beatles song created with AI technology, and exciting developments in the world of language models. They also explore a groundbreaking open dataset called RedPajama-Data-v2 and highlight three research papers on training large language models, modeling truthfulness in language models, and code generation.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:48 President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence

04:32 The Beatles: ‘final’ song Now and Then to be released thanks to AI technology

06:15 RedPajama-Data-v2: an Open Dataset with 30 Trillion Tokens for Training Large Language Models

07:43 Fake sponsor

09:38 FP8-LM: Training FP8 Large Language Models

11:06 Personas as a Way to Model Truthfulness in Language Models

12:23 CodeFusion: A Pre-trained Diffusion Model for Code Generation

14:11 Outro