What can art historians and computer scientists learn from one another? Actually, a lot! Amanda Wasielewski joins us to talk about how she discovered that computer scientists working on computer vision were actually acting like rogue art historians and how art historians have found machine learning to be a valuable tool for research, fraud detection, and cataloguing. We also discuss the rise of generative AI and how we this technology might cause us to ask new questions like: “What makes a photograph a photograph?”

This week we’re talking about LLMs with Simon Willison. We can not avoid this topic. Last time it was Stable Diffusion breaking the internet. This time it’s LLMs breaking the internet. Large Language Models, ChatGPT, Bard, Claude, Bing, GitHub Copilot X, Cody…we cover it all.

Mat & Johnny interview everyone’s favorite LLM (Natalie with a special hat on) to see if it’d make a good hire as a Go dev. Also, Mat tries to turn it into his very own creepy robot by asking personal questions about his co-hosts. Things get weird. In a good way?

• Use any model from OpenAI, Anthropic, Cohere, Forefront, HuggingFace, Aleph Alpha, and llama.cpp.
• Full playground UI, including history, parameter tuning, keyboard shortcuts, and logprops.
• Compare models side-by-side with the same prompt, individually tune model parameters, and retry with different paramaters.

Try the hosted version if you prefer…

Daniel and Chris explore the intersection of Kaggle and real-world data science in this illuminating conversation with Christof Henkel, Senior Deep Learning Data Scientist at NVIDIA and Kaggle Grandmaster. Christof offers a very lucid explanation into how participation in Kaggle can positively impact a data scientist’s skill and career aspirations. He also shared some of his insights and approach to maximizing AI productivity uses GPU-accelerated tools like RAPIDS and DALI.

By modifying a classical neural network, we introduced a new model inspired by biology which shows improved memory performance. It helps modeling brain processes and may eventually improve AI-based language tools such as ChatGPT.

This app uses Apple’s Core ML Stable Diffusion implementation to achieve maximum performance and speed on Apple Silicon based Macs while reducing memory requirements. (Also runs on Intel based Macs.)

Jerod & the gang catch you up on what’s new and poppin’ in the web development world. We go deep on GitHub Copilot X and the latest AI advancements, take a bathroom break while Nick talks about TypeScript 5 & continue the debate about the future of React.

Taxy uses GPT-4 to control your browser and perform repetitive actions on your behalf. Currently it allows you to define ad-hoc instructions. In the future it will also support saved and scheduled workflows.

Taxy’s current status is research preview. Many workflows fail or confuse the agent. If you’d like to hack on Taxy to make it better or test it on your own workflows, follow the instructions below to run it locally. If you’d like to know once it’s available for wider usage, you can sign up for our waitlist.

Ok that’s cool… 🤯

Here it is using Google Calendar with the prompt “Schedule standup tomorrow at 10am. Invite david@taxy.ai”

This open letter by the Future of Life institute has been signed by 1380 people (so far) including notable technologists such as Steve Wozniak, Stuart Russell, Emad Mostaque (Stability AI) & Elon Musk.

Powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable. This confidence must be well justified and increase with the magnitude of a system’s potential effects. OpenAI’s recent statement regarding artificial general intelligence, states that “At some point, it may be important to get independent review before starting to train future systems, and for the most advanced efforts to agree to limit the rate of growth of compute used for creating new models.” We agree. That point is now.

Therefore, we call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4. This pause should be public and verifiable, and include all key actors. If such a pause cannot be enacted quickly, governments should step in and institute a moratorium.

The team behind this took LLaMA and trained it on ~800k GPT-3.5-Turbo responses. The result is a sophisticated assistant you can easily run on your laptop by downloading a 4GB file and cloning the repo.

We are seeing an explosion of AI apps that are (at their core) a thin UI on top of calls to OpenAI generative models. What risks are associated with this sort of approach to AI integration, and is explainability and accountability something that can be achieved in chat-based assistants?

Beth Rudden of Bast.ai has been thinking about this topic for some time and has developed an ontological approach to creating conversational AI. We hear more about that approach and related work in this episode.

Steve Yegge is very excited about LLMs and thinks the rest of us should be as well:

There is something legendary and historic happening in software engineering, right now as we speak, and yet most of you don’t realize at all how big it is.

LLMs aren’t just the biggest change since social, mobile, or cloud–they’re the biggest thing since the World Wide Web. And on the coding front, they’re the biggest thing since IDEs and Stack Overflow, and may well eclipse them both.

Steve’s been in the industry a long time. He worked at Amazon back when AWS was just a demo on some engineer’s laptop and he worked at Google when Kubernetes was just a demo on some engineer’s laptop.

The point: when Steve Yegge gets excited about something it probably means more than when most people get excited about something.

This week we’re talking with Georgi Gerganov about his work on Whisper.cpp and llama.cpp. Georgi first crossed our radar with whisper.cpp, his port of OpenAI’s Whisper model in C and C++. Whisper is a speech recognition model enabling audio transcription and translation. Something we’re paying close attention to here at Changelog, for obvious reasons. Between the invite and the show’s recording, he had a new hit project on his hands: llama.cpp. This is a port of Facebook’s LLaMA model in C and C++. Whisper.cpp made a splash, but llama.cpp is growing in GitHub stars faster than Stable Diffusion did, which was a rocket ship itself.

This project brings stable diffusion models onto web browsers. Everything runs inside the browser with no server support. To our knowledge, this is the the world’s first stable diffusion completely running on the browser. Please checkout our demo webpage to try it out.

Josh Comeau:

Over the past few months, I’ve spoken with lots of early-career devs who are getting more and more anxious about AI. They’ve seen the increasingly-impressive demos from tools like GPT-4, and they worry that by the time they’re fluent in HTML/CSS/JS, there won’t be any jobs left for them.

I couldn’t disagree more. I don’t think web developer jobs are going anywhere. And I’m getting pretty sick of the FUD? being spread online.

So, in this blog post, I’m going to share my hypothesis for what will happen. Things are going to change, but not in the scary way people are saying.

Neural search and chat-based search are all the rage right now. However, You.com has been innovating in these topics long before ChatGPT. In this episode, Bryan McCann from You.com shares insights related to our mental model of Large Language Model (LLM) interactions and practical tips related to integrating LLMs into production systems.

Sualeh Asif from Control (an AI code editor):

We’ve been using GPT-4 for a few months internally, and we thought we’d highlight a few examples](https://github.com/anysphere/gpt-4-for-code) that have been both particularly impressive and really useful to us.

Here it’s converting a Python dict of member functions to esoteric but correct-on-first-try C++ code 👇

Justin Searls dives deep into whether AI tools like ChatGPT actually threaten knowledge worker jobs and provides helpful ideas around what to do about it.

Having spent months programming with GitHub Copilot, weeks talking to ChatGPT, and days searching via Bing Chat as an alternative to Google, the best description I’ve heard of AI’s capabilities is “fluent bullshit.” And after months of seeing friends “cheat” at their day jobs by having ChatGPT do their homework for them, I’ve come to a pretty grim, if obvious, realization:

The more excited someone is by the prospect of AI making their job easier, the more they should be worried.

The plugin is open source on GitHub. Photoshop is not 😉

We’ve all experienced pain moving from local development, to testing, and then on to production. This cycle can be long and tedious, especially as AI models and datasets are integrated. Modal is trying to make this loop of development as seamless as possible for AI practitioners, and their platform is pretty incredible!

Erik from Modal joins us in this episode to help us understand how we can run or deploy machine learning models, massively parallel compute jobs, task queues, web apps, and much more, without our own infrastructure.

With the recent proliferation of generative AI models (from OpenAI, co:here, Anthropic, etc.), practitioners are racing to come up with best practices around prompting, grounding, and control of outputs.

Chris and Daniel take a deep dive into the kinds of behavior we are seeing with this latest wave of models (both good and bad) and what leads to that behavior. They also dig into some prompting and integration tips.

We’re super excited to welcome Jay Alammar to the show. Jay is a well-known AI educator, applied NLP practitioner at co:here, and author of the popular blog, “The Illustrated Transformer.” In this episode, he shares his ideas on creating applied NLP solutions, working with large language models, and creating educational resources for state-of-the-art AI.

We’ve been hearing about “serverless” CPUs for some time, but it’s taken a while to get to serverless GPUs. In this episode, Erik from Banana explains why its taken so long, and he helps us understand how these new workflows are unlocking state-of-the-art AI for application developers. Forget about servers, but don’t forget to listen to this one!