The AI Daily Brief: Artificial Intelligence News and Analysis - The Most Important AI News This Week: ChatGPT Browse, Apple Quartz, GPT Robot Dogs

Episode Date: April 28, 2023

Discussed on the show: ChatGPT Privacy - https://twitter.com/DataChaz/status/1650961357913616402 ChatGPT Browse - https://twitter.com/zacknotes/status/1651360024877203457 HuggingChat - https://twitter.com/ClementDelangue/status/1650908484936908808 Microsoft Designer - https://twitter.com/yusuf_i_mehdi/status/1651625352131907584 Google Bard - https://www.theverge.com/2023/4/21/23692517/google-ai-bard-chatbot-code-support-functions-google-sheets Apple AI Quartz - https://twitter.com/aakashg0/status/1651600006405660673 Sully Omarr on Apple's M2 opportunity - https://twitter.com/SullyOmarr/status/1651275484573966336 Harvey funding - https://www.sequoiacap.com/article/partnering-with-harvey-putting-llms-to-work/ A16Z Pinecone - https://a16z.com/2023/04/27/investing-in-pinecone Replit - https://techcrunch.com/2023/04/27/replit-funding-100m-generative-ai/ Palantir - https://www.vice.com/en/article/qjvb4x/palantir-demos-ai-to-fight-wars-but-says-it-will-be-totally-ethical-dont-worry-about-it ChatGPT in Robot Dogs - https://twitter.com/svpino/status/1650832349008125952 Rate of Change - https://theaibreakdown.beehiiv.com/p/rate-change Midjourney updates - https://twitter.com/nickfloats/status/1651314922326994944 Watch the original video: https://www.youtube.com/@TheAIBreakdown

This episode of the AI breakdown was first released as a YouTube video on Friday, April 28th. It is a weekly overview of all the most important news and headlines from this fast-moving industry. From new chat GPT features to funding rounds from startups to Britain spending hundreds of millions of dollars on their own sovereign AI. This is the AI news you need. Welcome back to the AI breakdown. For this episode, we're doing a week roundup of news. We're going to talk about everything from chat GPT's updates to a number of big company initiatives in the space to funding rounds to a country's involvement to chat GPT powered robot dogs. Anyways, let's dive in and let's start with chat GPT.

So there are three announcements or three updates that came to chat GPT this week. The first was this sort of incognito mode, right? A privacy enabled mode that allowed users or allows users to disable their chat. history. And when they do, that means that Open AI won't train their models or improve their existing models with user data. Seems like a really important and obvious step for the company also potentially has to do with rules and regulations and consternation, particularly from Europe, around privacy laws in that country. Now, relatedly, ChatGPT or Rather, OpenAI, also announced that they'd be having a forthcoming business mode, which is private by default.

So it has this sort of chat history disabled idea. And that means that that private businesses don't have to worry about open AI models accidentally revealing their private information that it's been trained on. Lastly, I started to notice more and more people seeing a browse mode in chat GPT. So this is GPT 3.5, but it can access the internet. It's very nascent, it seems. Not everyone has access to it. I don't, for example. I've seen some people like Robert Scobal asking when they can have access to it, but this is obviously a potentially game-changing feature. Right now, chat GPT's data set only goes up to 2021. It can't go out and search information on the internet. It's part of why people have been so excited about auto GPTs is that they have that

capacity. Now, there are a bunch of safety issues that relate to this when it comes to these AIs being able to go out and browse the internet. But the industry is moving full steam ahead when it comes to it. And so I think it seems pretty likely that we're going to have this chat GPD browsing feature roll out more broadly in the weeks to come. There have also been some updates when it comes to chat GPT open source competitors. Last week, we saw StableLM from Stability.AI, and this week the team behind HuggingFace announced Hugging Chat, which is an open source prototype interface powered by Open Assistant that is a chat GPT alternative.

If you're interested in how the early versions of Hugging Chat perform as compared to Chat GPT, I did a video about that, which you can go check out. Now, while there is some debate around open source LLMs, Clem from Hugging Face makes his position clear. He says, I believe we need open source alternatives to chat CPT for more transparency, inclusivity, accountability, and distribution of power. Speaking of distribution of power, let's turn to the big companies and what they've been doing this week. So first up, we have Microsoft. They announced last year their designer platform. And as you can see, basically what it is, is it's a Canva alternative that uses AI to try to move from concept or what you need to an actual

design very quickly. They've started to make that much more widely available this week, and people have been pretty excited about it. It's a very practical tool and sort of cements that Microsoft is really here to play when it comes to the whole AI revolution. Now, a company that has been kind of behind or feeling behind, at least, when it comes to AI, is, of course, Google. Just a few months ago, it sounds like, based on New York Times reporting, Samsung told Google that they might be replacing them as the default browser on their phones, and that freaked Google out. They are reportedly working on a new search experience, Project Magi. And last week, at the very end of the week, they announced that Bard would now have the ability to help around coding and coding support.

So this, it seems, might be their play to try to get back into the game against competitors like OpenAI and their relationship with Microsoft. Now, speaking of companies who haven't really announced their play into the AI space yet, obviously the big one that we're all waiting on is Apple. This week we got a little bit of information maybe that there's an AI-powered health app codenamed courts that's coming out in an iOS to be released in September. Now, a lot of people have noted that this is sort of not the full extent of what you'd expect from Apple. Kosh here says quite a tame entry point for the world's most valuable company. And I don't believe for a second that this is all that Apple has in mind when it comes to AI. I'm going to do another video about this at some point. but if you want to kind of preview of something that I think is a lot closer to the right way of thinking,

go check out Sully Omar's thread on Apple's approach, what it means to have M2 chips in all these computers with the excess capacity that they give, and how that might connect into Apple's AI strategy. I will include a link to this thread in the notes. Of course, it's not just big companies that are responding to the AI revolution. It's also countries. We've seen some responses be to want to ban these tools. Italy banning chat GPT, Germany potentially having an inquiry, all of these European countries

being concerned around data. Well, the UK is going with a slightly different approach. This week, the Prime Minister and the Technology Secretary pledged an initial 100 million pounds to establish a foundation model task force. This is designed to make Britain competitive relative to the chat GPTs of the world. And while there is a pretty significant budget here, according to later interviews with the Prime Minister Rishi Sunak, it's not necessarily the case that they're going to for sure create a chat GPT competitor. They want to have the capacity to do so, but the idea or the first job of the foundation model task force is to make a recommendation to the British government about exactly how they should play. But again, you're starting to see

different approaches to how countries are going to handle AI from banning to building and probably everything in between. Next, we move on to startup funding. And on any given week, there's no shortage of companies in the AI space that are getting venture capital dollars. It is definitely the biggest magnet for venture capital in the world right now. The first that I'll mention is Harvey, which was funded by Sequoia, and basically the idea is to translate tools like ChatGPT and these large language models into a business-ready form. Andresen Horowitz led a $100 billion series B round into Pine Cone.

Pine Cone they describe as the storage layer for LLMs. They say, Pinecone is an external database where developers can store relevant contextual data for LLM apps. Rather than sending large document collections back and forth with every API call, developers can store them in a pine cone database, then pick only the few most relevant for any given query, an approach called in-context learning. It's a must-have for enterprise use cases to truly bloom.

Basically, what they're trying to solve here is the idea that, again, if you have a business that's trying to take advantage of this entirely new category of computing, which is really what LLMs are, they don't necessarily have all the data that you need. There might be proprietary data that needs to be fed into the system. Pine Cone is an external database that plugs into these LLMs that makes that data available so that these tools can work at scale for enterprises. So you're kind of seeing a theme, right? And when it comes to AI transformation for the enterprise, these are the key pieces of infrastructure that are being built now. The other big funding round announced this week was for Replit. And this is a company whose most in-demand feature seems to be

its ghostwriter tool, which is effectively an AI-powered tool that helps coders code. More effectively, more efficiently helps them answer their questions. With this $100 million raise, these guys are now the latest AI unicorn. Speaking of unicorns, another tech unicorn came out with an AI demo, and it didn't necessarily have the exact response they wanted. Palantir came out with its new artificial intelligence platform that integrates AI into military decision-making, and it should. showed basically AI controlling drones, and there was a lot of response to this that suggested

that perhaps this wasn't the best demo. I think the Vice headline here sums it up. Palantir demos AI to fight wars, but says it will be totally ethical. Don't worry about it. Now, AI's integration with the military is on the one hand almost totally inevitable, but at the same time it does raise a number of these security questions and AI safety questions that feel like they don't always get enough coverage as this conversation has gotten so much more breathless with the advent of all these new technologies. And speaking of things that scare the hell out of some people, this is a video on your screen of one of those Boston Dynamics robots that I'm sure you've seen around. New York City just hired one as a police officer. Well, now they're integrated with chat GPT. They say,

we had a ton of fun building this. But of course, I think a lot of the people watching it are thinking about these cute little interactions about asking how much battery it has left, but about entirely freaky scenarios in a dystopian future ruled by robot dogs. Now, as for me, one of the things that I've been thinking a lot about this week is the rate of change. There was a very viral thread on Reddit that showed basically a series of comparison images that were all created by Mid Journey, and these were produced a year apart with identical prompts.

So one was produced in April 2022. and one was produced in April 2023. And you can see if you're watching this, that there is an incredible difference between these images, just a stunning rate of progress in a single year. And what it makes me think about is the fact that so few people actually anticipated this speed of change even a year ago, even two years ago,

that it makes one wonder if we should just basically not trust our sense of how fast things will happen. Many of the conversations about things like AI ethics and safety are predicated on assumptions or predictions around how fast different types of things will or won't happen. And in a world where it seems like every assumption we have about how fast things are happening are being challenged to the upside because they're happening faster, maybe we need to recalibrate in a pretty fundamental way. Now, when it comes to mid-jurney, it sounds like they are in the middle of training V6.

They had wanted to come out every 30 days with a new version, but there are a few things in the works that took a little bit longer, but they're also already working on V7. And of course, in Text to Video Land, Runways Gen 2 is started to make it out in beta, and people are putting together entire short films and movie trailers. And it's pretty remarkable to see how quickly this has come. Text to image is one thing, and it's a mind-blowing technology, but text to video, where you can actually imagine a world and then you see. it spit back at you is something else entirely. Now the naysayers will, of course, as they did on my recent Twitter thread, point to the hands. It's always the hands. The hands are too wonky, too weird. But that obviously obscures the fact that this is changing so quickly and all of us are just racing to catch up.

Hopefully these videos, the ones that come out every day and the one that I did here with this recap help you feel a little bit more like you understand what's going on. I will keep doing them as much as I can to help you on your journey. So until next time, guys, peace.