The Changelog: Software Development, Open Source - LMMS are the new LLMs (News)

Episode Date: October 16, 2023

Chip Huyen documents the shifting sand of large data models, Herman Õunapuu reviews the Zimaboard, Bryan Braun shares 4 of his most recent VSCode configuration discoveries & Swizec Teller wrote a gre...at summary of the inaugural AI Engineer Summit.

Transcript
Discussion (0)
Starting point is 00:00:00 What up, nerds? I'm Jared, and this is Changelog News for the week of Monday, October 16th, 2023. Except I'm recording it on the afternoon of Friday, October 13th, because hey, we'll be at all things open on Monday. So if some big software news drops over the weekend and you're wondering why I gapped it, that's why. Okay, let's get into the news. Chip Huyan documents in spectacular depth the shifting sand of large data models, moving from LLMs to LMMs. She says, for a long time, each ML model operated in one data mode, text, image, or audio. However, natural intelligence is not limited to just a single modality. Humans can read and write text, we can see images and watch videos, we listen to music to relax and watch out for strange noises
Starting point is 00:01:00 to detect danger. Being able to work with multimodal data is essential for us or any AI to operate in the real world. End quote. Incorporating additional modalities into LLMs produces large multimodal models, LMMs, and everyone's doing it. DeepMind, Salesforce, Microsoft, Tencent, and of course, OpenAI. ChatGPT is already an LMM. Not all multimodal systems are LMMs, though. Chip says MidJourney, StableDiffusion, and DALI don't qualify because they don't have a language model component. Her very long post, which is linked, consists of three parts and is worth studying to keep up with the latest.
Starting point is 00:01:52 Herman Ouanapu calls Xemaboard the closest thing to his dream home server setup. What's a Xemaboard? It's a small, single-board computer that is relatively affordable and comes with an interesting selection of ports, which includes an exposed PCI Express port. Herman wrote up a full review, which is linked. It starts with, quote, what sets the Zima board apart from other single board computers is how polished the product feels. Unlike a board like the Raspberry Pi, this one comes with a case and a cooling setup already attached to it. I suspect that a similarly configured Raspberry Pi 4 or 5, with all the accessories added on top, would result in a price that's quite similar to the cost of a top-of-the-line Zima board. End quote. He goes into details on its storage, power consumption, performance,
Starting point is 00:02:38 and caveats, which is always nice to see in an otherwise glowing review. Herman's conclusion, quote, Overall, I'm very happy with this purchase. The performance is just enough for my services to work reasonably fast. The board uses very little power, and it's silent. And it looks good on my wall. End quote. If the Xima board is new to you, and you have a dream home server setup, give it a look.
Starting point is 00:03:08 It's now time for sponsored news. There's so much going on in the data and machine learning space, it's hard to keep up. Example, did you know that graph technology lets you connect the dots across your data and ground your LLM in actual knowledge? To learn about this new approach, don't miss Neo4j's Nodes online conference on October 26th. At this free online conference, developers and data scientists from all around the world share how they use graph technology for everything from building intelligent apps and APIs to enhancing machine learning and improving data visualizations. There are 90 talks over 24 hours, so no matter where you are, you can attend live sessions. To register for this free conference, visit neo4j.com slash nodes. That's N-E-O, the number four, j.com slash nodes. Or
Starting point is 00:03:55 just follow the link in your chapter data and in the newsletter. Thanks once again to Neo4j for sponsoring Changelog News. Brian Braun has been using VS Code for five years now, but he still learns new things about it all the time. In the linked post, Brian shares four of his most recent discoveries that he wishes he'd known from the get-go. One, use gitignore underscore global to ignore the entire.vscode folder. Two, formatters and linter extensions should always be enabled at the workspace level. 3. The built-in terminal is really great.
Starting point is 00:04:31 4. You might not need GitLens. There's some dissension in the comments, but regardless, it's always interesting to hear how other hackers hack, whether or not you agree with them. The inaugural AI Engineer Summit took place last week. Swizik Teller attended and wrote up a great summary of the event. His takeaway, everyone's dealing with the same three problems. One, good old-fashioned data engineering.
Starting point is 00:05:01 Two, evals and non-determinism. Three, product development. SwizX says AI is, quote, the easy part. Turning your demo into an actual product, now that's the challenge. The summit itself sounds like it was a hit. Up next for the organizers in 2024, an AI World's Fair. That's the news for now, but it's time once again for some ChangeLog++ shoutouts. Shoutout to our newest members, Nathan P, Aaron Y, Christopher W, Johan H, Maurice H, Brian M, Adam G, Anthony B, Jan K, Michael P, and Bob P. We appreciate you for supporting our work with your hard-earned cash. If ChangeLog++ is news to you, that's our membership program.
Starting point is 00:05:51 You can join to ditch the ads, get in on bonus content like extended episodes, and directly support our work. Plus, you get shout-outs like the ones you just heard. Check it out at changelog.com slash plus plus. Changelog plus plus. It's better. Have a great week. Say hi if you're at All Things Open.
Starting point is 00:06:13 Hook us up with a five-star review if you dig the show. And I'll talk to you again real soon.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.