GPT Reviews

Apple's new MLX framework for on-device AI could shake up the AI race with its optimized design for Apple silicon and ecosystem of devices.

Duolingo's shift towards using AI to create more content and cutting contractors raises concerns about how AI technology will affect jobs in the long run.

The papers discussed in this episode showcase exciting advancements in open-source language models, including DeepSeek LLM's superior performance compared to GPT-3.5 and Alibaba Group's proposed system for supporting exceptionally long context lengths.

"Self-Contrast" is a new method proposed to improve the reflection capacity of Large Language Models (LLMs) by adaptively exploring diverse solving perspectives and generating a checklist to help LLMs re-examine and eliminate errors or inconsistencies.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:35 Apple ML Research releases MLX for on-device AI

02:53 Duolingo Cuts 10% of Contractors as It Uses More AI to Create App Content

04:26 Attacks on machine learning models

05:37 Fake sponsor

07:22 DeepSeek LLM: Scaling Open-Source Language Models with Longtermism

09:09 Infinite-LLM: Efficient LLM Service for Long Context with DistAttention and Distributed KVCache

10:51 Self-Contrast: Better Reflection Through Inconsistent Solving Perspectives

13:06 Outro

OpenAI's GPT Store launch, Perplexity AI's natural language search engine, and two papers proposing new approaches to improve LLMs' reflection capacity and expand their capabilities.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:42 OpenAI’s GPT Store launching next week

03:01 AI-powered search engine Perplexity AI, now valued at $520M, raises $73.6M

04:39 Our 2023 Year in Review

05:56 Fake sponsor

07:49 Self-Contrast: Better Reflection Through Inconsistent Solving Perspectives

09:34 Instruct-Imagen: Image Generation with Multi-modal Instruction

10:59 LLM Augmented LLMs: Expanding Capabilities through Composition

12:48 Outro

Microsoft's Copilot key for PC keyboards, Samsung's upcoming AI advancements in their smartphone series, a framework for generating photorealistic avatars that gesture according to conversational dynamics, and MIT CSAIL's exploration of how language models learn about the visual world and their potential for training visual representation learning systems.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:37 Microsoft wants to add a Copilot key to your PC keyboard

02:59 Galaxy Unpacked 2024: Opening a New Era of Mobile AI

04:53 Efficient LLM inference

06:15 Fake sponsor

08:20 From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations

09:50 Incremental FastPitch: Chunk-based High Quality Text to Speech

11:04 A Vision Check-up for Language Models

13:00 Outro

This episode covers a range of fascinating topics, from the use of AI in court proceedings to the impressive safety record of Waymo's driverless cars. We also explore cutting-edge research on autoregressive multimodal models and fine-tuning weak language models using self-play. Don't miss out on these exciting developments in the world of AI and technology!

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:36 AI approved for use in court proceedings

02:51 Self-driving cars are safer than human drivers

04:38 Intel GenAI For Yield, TSMC CFET & 3D Stacking, AMD 3D Device Modeling, Applied Materials Material Innovation, SK Hynix HBM 4, Micron 3D DRAM & FeRAM, Hybrid Bonding vs TCB - IEDM 2023

06:00 Fake sponsor

08:17 Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action

09:42 Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models

11:19 Exploring the Frontiers of LLMs in Psychological Applications: A Comprehensive Review

13:06 Outro

OpenAI's move to reduce regulatory risks in the EU around data privacy and their mind-blowing revenue generated by ChatGPT.

The concept of man-computer symbiosis proposed in a paper from 1960 by J.C.R. Licklider and its implications for the future of AI and human-machine interaction.

Gemini, a new Multimodal Large Language Model introduced by Google, and its competitive commonsense reasoning capabilities when evaluated on a range of complex reasoning tasks.

Self-Play Fine-Tuning (SPIN), a new fine-tuning method proposed by researchers from UCLA, which aims to grow a strong Large Language Model out of a weak one without the need for additional human-annotated data.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:24 OpenAI moves to shrink regulatory risk in EU around data privacy

02:56 OpenAI Has Reportedly Generated $1.6B In Revenue

04:54 Man-Computer Symbiosis

06:31 Fake sponsor

08:14 Gemini in Reasoning: Unveiling Commonsense in Multimodal Large Language Models

10:16 LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning

11:57 Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models

14:07 Outro

Microsoft's Copilot app now available on iOS, Google's potential layoff of 30,000 employees due to new AI innovations, and the emergence of hallucinations as a mainstream research topic. Additionally, three papers explore different aspects of artificial intelligence research, including a large-scale corpus of math-centric text, a novel dynamic scene representation, and a new process-oriented math process reward model called Math-Shepherd.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:42 Microsoft’s Copilot app is now available on iOS

03:01 Google likely to layoff 30,000 employees post new AI innovation

04:50 Jason Wei Tweet on Hallucinations

06:15 Fake sponsor

08:01 Generative AI for Math: Part I -- MathPile: A Billion-Token-Scale Pretraining Corpus for Math

09:28 Spacetime Gaussian Feature Splatting for Real-Time Dynamic View Synthesis

11:19 Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotations

13:09 Outro

T Dictionary.com Word of the Year, ArXiv's move towards more accessible scientific research, and Bill Gates' thoughts on the potential impact of AI. The show also delves into cutting-edge AI research, including benchmarking and analyzing NLP paradigms for biomedical knowledge curation, a novel speech translation model, and the ability of LLMs to generate human-like opinions. Tune in to stay up-to-date on the latest developments in the world of artificial intelligence.

Contact:  [email protected]

Timestamps:

00:34 Introduction

02:00 The Dictionary.com Word of the Year is hallucinate.

03:28 ArXiv now offers papers in HTML format

05:22 Bill Gates on AI in 2024

06:46 Fake sponsor

08:48 Benchmarking and Analyzing In-context Learning, Fine-tuning and Supervised Learning for Biomedical Knowledge Curation: a focused study on chemical entities of biological interest

10:21 Speech Translation with Large Language Models: An Industrial Practice

12:17 ChatGPT as a commenter to the news: can LLMs generate human-like opinions?

14:20 Outro

Stability AI's new paid membership for commercial use of its models, Microsoft Copilot's new music creation feature, Mistral 7B Fine-Tune Optimized's release for free on Hugging Face and as the new default base model within OpenPipe, and the Alignment Research Center's investigation into the ability of language model agents to replicate themselves and adapt to novel challenges in the real world.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:39 Stability AI announces paid membership for commercial use of its models

03:14 Microsoft Copilot gets a music creation feature via Suno integration

05:02 How we built “Mistral 7B Fine-Tune Optimized,” the best 7B model for fine-tuning

06:21 Fake sponsor

08:16 Evaluating Language-Model Agents on Realistic Autonomous Tasks

10:16 LLM in a flash: Efficient Large Language Model Inference with Limited Memory

11:48 A Challenger to GPT-4V? Early Explorations of Gemini in Visual Expertise

13:36 Outro

A failed acquisition attempt by Adobe, new legal protections and improvements to Anthropic's API, and two research papers exploring the potential of large language models for solving geometric problems and measuring language model fit across multiple domains.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:36 Adobe gives up on $20 billion acquisition of Figma

03:03 Expanded legal protections and improvements to Anthropic's API

04:50 How to make LLMs go fast

05:52 Fake sponsor

07:44 G-LLaVA: Solving Geometric Problem with Multi-Modal Large Language Model

09:24 Paloma: A Benchmark for Evaluating Language Model Fit

10:58 Your Student is Better Than Expected: Adaptive Teacher-Student Collaboration for Text-Conditional Diffusion Models

12:59 Outro

From OpenAI suspending ByteDance's account to rising AI job losses, we explore the ethical and legal concerns surrounding AI development. We also highlight promising developments in AI safety and knowledge discovery, such as OpenAI's Prompt Engineering Guide and Preparedness Framework. Finally, we discuss a new technique called weight subcloning, which expedites the training of scaled-down transformers.

Contact:  [email protected]

Timestamps:

00:34 Introduction

01:48 OpenAI suspends ByteDance’s account after it used GPT to train its own AI model

03:06 Recent data shows AI job losses are rising, but the numbers don’t tell the full story

05:23 Prompt Engineering Guide by OpenAI

06:21 Preparedness

07:51 Fake sponsor

09:51 Challenges with unsupervised LLM knowledge discovery

11:11 Weight subcloning: direct initialization of transformers using larger pretrained ones

13:05 Outro