A brief summary of language model finetuning
Here's a (brief) summary of language model finetuning, the various approaches that exist, their purposes, and what we know about how they work.
Loading…
llm
Tragedy of the (data) commons
Ben chats with Shayne Longpre and Robert Mahari of the Data Provenance Initiative about what GenAI means for the data commons. They discuss the decline of public datasets, the complexities of fair use in AI training, the challenges researchers face in accessing data, potential applications for synthetic data, and the evolving legal landscape surrounding AI and copyright.
Masked self-attention: How LLMs learn relationships between tokens
Masked self-attention is the key building block that allows LLMs to learn rich relationships and patterns between the words of a sentence. Let’s build it together from scratch.
Detecting errors in AI-generated code
Ben chats with Gias Uddin, an assistant professor at York University in Toronto, where he teaches software engineering, data science, and machine learning. His research focuses on designing intelligent tools for testing, debugging, and summarizing software and AI systems. He recently published a paper about detecting errors in code generated by LLMs. Gias and Ben discuss the concept of hallucinations in AI-generated code, the need for tools to detect and correct those hallucinations, and the potential for AI-powered tools to generate QA tests.
The world’s largest open-source business has plans for enhancing LLMs
Ben and Ryan talk to Scott McCarty, Global Senior Principal Product Manager for Red Hat Enterprise Linux, about the intersection between LLMs (large language models) and open source. They discuss the challenges and benefits of open-source LLMs, the importance of attribution and transparency, and the revolutionary potential for LLM-driven applications. They also explore the role of LLMs in code generation, testing, and documentation.
LLMs evolve quickly. Their underlying architecture, not so much.
The decoder-only transformer architecture is one of the most fundamental ideas in AI research.
Practical tips for retrieval-augmented generation (RAG)
Retrieval-augmented generation (RAG) is one of the best (and easiest) ways to specialize an LLM over your own data, but successfully applying RAG in practice involves more than just stitching together pretrained models.
The framework helping devs build LLM apps
Ben and Eira talk with LlamaIndex CEO and cofounder Jerry Liu, along with venture capitalist Jerry Chen, about how the company is making it easier for developers to build LLM apps. They touch on the importance of high-quality training data to improve accuracy and relevance, the role of prompt engineering, the impact of larger context windows, and the challenges of setting up retrieval-augmented generation (RAG).
We chat search from both sides now
In this episode, Ben chats with Elastic software engineering director Paul Oremland along with Stack Overflow staff software engineer Steffi Grewenig and senior software developer Gregor Časar about vector databases and semantic search from both the vendor and customer perspectives.
Explaining generative language models to (almost) anyone
Here’s a simple, three-part framework that explains generative language models.
Breaking up is hard to do: Chunking in RAG applications
A look at some of the current thinking around chunking data for retrieval-augmented generation (RAG) systems.
How to train your dream machine
Ben and Ryan talk with Vikram Chatterji, founder and CEO of Galileo, a company focused on building and evaluating generative AI apps. They discuss the challenges of benchmarking and evaluating GenAI models, the importance of data quality in AI systems, and the trade-offs between using pre-trained models and fine-tuning models with custom data.
Is GenAI the next dot-com bubble?
The home team talks about the current state of the software job market, the changing sentiments around AI job opportunities, the impact of big players like Facebook and OpenAI on the space, and the challenges for startups. Plus: The philosophical implications of LLMs and the friendship potential of corvids.
How do you evaluate an LLM? Try an LLM.
On this episode: Stack Overflow senior data scientist Michael Geden tells Ryan and Ben about how data scientists evaluate large language models (LLMs) and their output. They cover the challenges involved in evaluating LLMs, how LLMs are being used to evaluate other LLMs, the importance of data validating, the need for human raters, and more needs and tradeoffs involved in selecting and fine-tuning LLMs.
Diverting more backdoor disasters
In the wake of the XZ backdoor, Ben and Ryan unpack the security implications of relying on open-source software projects maintained by small teams. They also discuss the open-source nature of Linux, the high cost of education in the US, the value of open-source contributions for job seekers, and what Apple is up to AI-wise.
How do mixture-of-experts layers affect transformer models?
This new LLM technique has started improving the results of models without additional training.
Are long context windows the end of RAG?
The home team is joined by Michael Foree, Stack Overflow’s director of data science and data platform, and occasional cohost Cassidy Williams, CTO at Contenda, for a conversation about long context windows, retrieval-augmented generation, and how Databricks’ new open LLM could change the game for developers. Plus: How will FTX co-founder Sam Bankman-Fried’s sentence of 25 years in prison reverberate in the blockchain and crypto spaces?
Your whole repo fits in the context window
The home team discusses the challenges (hardware and otherwise) of building AI models at scale, why major players like Meta are open-sourcing their AI projects, what Apple’s recent changes mean for developers in the EU, and Perplexity AI’s new approach to search.
A leading ML educator on what you need to know about LLMs
Machine learning scientist, author, and LLM developer Maxime Labonne talks with Ben and Ryan about his role as lead machine learning scientist, his contributions to the open-source community, the value of retrieval-augmented generation (RAG), and the process of fine-tuning and unfreezing layers in LLMs. The team talks through various challenges and considerations in implementing GenAI, from data quality to integration.
It’s RAG time for LLMs that need a source of truth
On this episode: Roie Schwaber-Cohen, Staff Developer Advocate at Pinecone, joins Ben and Ryan to break down what retrieval-augmented generation (RAG) is and why the concept is central to the AI conversation. This is part one of our conversation, so tune in next time for the thrilling conclusion.
Even LLMs need education—quality data makes LLMs overperform
If you’re building experimental GenAI features that haven’t proven their product market fit, you don’t want to commit to a model that runs up costs without a return on that investment.
Best practices for building LLMs
Intuit shares what they've learned building multiple LLMs for their generative AI operating system.
How to beat Doom in just 600 years
Ben and Ryan discuss how complex images (and maybe even interactive games) are being encoded in living cells, the latest trends in prompt engineering, and the educational benefits of gaming.
The hardest part of building software is not coding, it's requirements
Why replacing programmers with AI won’t be so easy.
