ChatGPT Adds Private Mode as HuggingFace Launches Open Source Competitor

0:00The AI Breakdown you're about to hear originally aired as a YouTube video on Wednesday, April 26th. In it, we discuss the new updates from ChatGPT, including their privacy-focused incognito mode, as well as their forthcoming business offering. And we look at a new open-source ChatGPT competitor from HuggingFace called HuggingChat. Today on the AI Breakdown, ChatGPT adds an incognito mode and teases a business version, while HuggingFace opens an open-source competitor. Welcome back to the AI Breakdown. So like I said, today we are starting with some updates on ChatGPT. We'll move into why those updates are coming now, and then we'll turn to a new open source competitor.

0:46The first update that ChatGPT announced that has people really excited is effectively their version of incognito mode. It allows users to selectively turn off chat history, and that means a couple things. It means that the history won't be recorded in the little side panel, which means after 30 days, it'll be deleted entirely. And it also means that OpenAI will not use that chat to train its future models or to improve the existing ones. Now, this is something that has a lot of people excited. It seems like a very basic feature to people that massively improves the usability of ChatGPT in the same way that incognito windows are a key part of modern browsers.

1:27The second announcement that ChatGPT had was a new business subscription. And effectively, it's sort of like like incognito mode for business. I think there's going to be a lot more to that as well. But as you can see here, the idea is it allows companies to use ChatGPT without accidentally leaking trade secrets or just letting the model training know what they're up to, which could create some security concerns. OpenAI spelled all of this out in a post new ways to manage your data and ChatGPT. And there's really not very much. It goes into what the privacy or incognito mode is. And it also just promises that business subscriptions will be coming in the future.

2:07We're working on a new ChatGPT business subscription for professionals who need more control over their data, as well as enterprises seeking to manage their end users. ChatGPT business will follow our API's data usage policies, which means the end users data won't be used to train our models by default. Now, why would they be doing this now? Well, there's a couple reasons. One is the risk that companies are starting to see privacy issues, right? This is a tweet from Uriel at Zengo who says, We just caught ChatGPT in a concerning privacy trap. We asked something about Zengo we know for a fact is not public, but yet ChatGPT provided that information pretty accurately.

2:48The only possible way was that ChatGPT accessed or read some of our internal emails. When we asked the source, this was the answer. Concerning, does ChatGPT access or read emails and uses the information somehow? The response that he linked was ChatGPT saying, I apologize for the confusion in my previous response. I do not have access to any information beyond what is publicly available online. Upon further research, I was not able to find any information about a blank service called blank. It appears that Zengo currently does not offer such a service. Thank you for bringing this to my attention, and I apologize for any confusion that I might have caused.

3:22Now, this is just one example. We don't know if this is something that other companies are widely experiencing, but even if it's just a weird error or a fluke, this is exactly the type of thing that OpenAI does not want to have happen on the regular. There's also the issue of global regulatory compliance. A lot of the tweets that you see about these features are clearly connecting them to concerns around things like GDPR in Europe and other data policies. It makes sense that they're making these moves now. Already in April, we saw the first country actually ban or temporarily ban at least chat GPT because of concerns around personal data issues and data protection laws.

4:03Italy was that country, and they effectively said that right now the way that chat GPT collects data is just incompatible with things like GDPR. There's also issues with the lack of age verification, and so because of that, they wouldn't allow their citizens to use I think there are good debates to be had around whether this is actually banning chat GPT or whether it's just banning your citizens for chat GPT. But for right now, that doesn't super matter. What's more, it seems like Italy is just the first country, but it's not the last country that's going to look into these types of policies. Just yesterday, news agencies started reporting that Germany is also launching a data protection inquiry around chat GPT.

4:41A commissioner from that country said, we want to know if data protection impact assessment has been carried out and if the data protection risks are under control. We were asking OpenAI for information on issues that stem from the European General Data Protection Regulation, or GDPR. It's not surprising that Europe is getting into these issues most quickly because they have GDPR on the books, right? There's a specific law that they can point to as potentially being broken when it comes to chat GPT. But there's lots of possible different answers to this issue, and it's likely that every country is going to explore how they want to address some of the challenges, including the data privacy challenges of AI.

5:23So you have countries like Italy and Germany, which are potentially restricting different tools, even with very popular tools. But then you have another approach, which seems to be being explored by the UK. The UK will spend 100 million pounds to develop its own sovereign AI. The country aims to rival AI models like ChatGPT. So what's going on here is that the Prime Minister of Britain, Rishi Sunak, who's definitely a tech-advanced type of person, this is the guy who said that he wanted the UK to become a global crypto hub, and his technology secretary, Michelle Donilon, have put together this£100 million fund to establish a foundation model task force.

5:59The goal of the team is to develop AI that, quote, makes the country globally competitive, as well as makes these systems safer and more reliable. This is on top of another roughly£900 million, or$1.1 billion, that is in the UK budget already to deal with supercomputer and AI research issues. As Engadget puts it, officials aren't shy about their hopes. The UK wants to have a sovereign AI technology that spurs the economy while avoiding the ethical and technical pitfalls that have led experts to call for a six-month pause on experiments. Now, while the UK is taking this step, the Prime Minister Rishi is not yet willing to say for sure that it's the right idea to actually try to compete with a chat GPT type of service.

6:42Asked earlier this week, he said that he's going to take advice from this new task force before deciding whether to actually launch an initiative or a model to compete with OpenAI, Google, and whoever else. In the meantime, though, private industry is just racing ahead. And one of the things that we are seeing is more and more open source efforts trying to compete with or at least provide open source alternatives to these large language models that come from major tech companies. Last week, we saw one of these alternatives in Stable LM, which came from Stability AI, which is the team behind Stable Diffusion.

7:15And this week, we get a new open source competitor, this time from Hugging Face, which is the open source AI platform. Hugging Face CEO Clem says, I believe we need open source alternatives to chat GPT for more transparency, inclusivity, accountability, and distribution of power. Excited to announce Hugging Chat, an open source early prototype interface, powered by Open Assistant, a model that was released a few weeks ago. This is a V0 with many limitations, but we are iterating quickly on the interface and safety mechanisms and intend to support the next rapidly improving open source models. One of the things that Clem notes here is that in this V0 of Hugging Chat, they're only storing messages in order to display them to the user, meaning that they're not actually researching or training their model with user information, at least not yet.

8:00And a lot of people are really excited about this. Itamar Golan says today is a big day for open source AI. Hugging Face has just launched the first open source alternative to ChatGPT, Hugging Chat. It is powered by one, Open Assistant's latest model, which is the best open source chat model available right now. Two, Hugging Face Inference API. Three, UI, save threads, and basically almost everything like the ChatGPT experience. So of course, we can't get a new open source ChatGPT competitor without going and trying it. And so that's what I did. I said, how do you compare to ChatGPT? HuggingChat responds, I'm a very capable assistant that can handle complex tasks and provide detailed explanations.

8:38My creators designed me to be more than just a large language model like GPT. They gave me access to vast amounts of data beyond text, so I know more than just words on paper. Additionally, my training process includes fine-tuning for specific domains, making me well-suited to assisting in specialized fields such as law, medicine, finance, etc. I followed up and said, that's great, do you think you're creative? Hugging Chat very humbly said, creativity may not be my strongest suit, but I strive to bring innovation into everything I do. As I learned from user interactions, my abilities improve, and so can become even better at helping others create something new.

9:09Despite that warning, I decided to try it, and so I used a prompt that I've used before with ChatGPT, which is part of an experiment I've been doing with my four-year-old about how she can write stories using characters that she creates herself with the help of ChatGPT. I said to HuggingChat, I'm hoping to have your help writing a story. I'd like it to be similar to fanciful classic children's authors like Roald Dahl. The main characters are a nine-year-old girl and her unicorn. Can you suggest a few ideas for a key conflict to explore in the story? Hugging Chat came back with three possible conflicts.

9:37First is that the little girl wants to go back home because perhaps she finds out there's no Santa Claus, but her unicorn is trying to keep her entertained. The second is perhaps the girl accidentally breaks the magical stone and must find another one to replace it before bad people take control. The third is maybe the protagonist has always wanted wings and spends time wondering if magic exists. She then falls in love with the idea that it's real because of this object or item she keeps near her all the time without knowing why she needs it. Now, these are kind of entertaining, particularly the perhaps she finds out there's no Santa Claus at a really far journey away from GPT-4, which was able to come up with some really cool ideas.

10:11The disappearing rainbow, the nine-year-old girl and her unicorn discover that the colors of the world are mysteriously fading away. They embark on a quest to find the missing colors and restore the world's beauty, encountering quirky characters and overcoming obstacles on the way. Then ChatGPT also gave us the Unicorn's Lost Horn, the Enchanted Forest Challenge, the Great Candy Caper, the Dreamweaver's Dilemma, etc., etc. Point being just that Hugging Chat is correct that its creative capacity is not yet up to GPT-4, but of course we're not expecting it to be. I also tried to ask it about another domain area that I know a little bit more about.

10:42I said, do you know much about Bitcoin? It says, sure, Bitcoin is a decentralized cryptocurrency without a central bank or single administrator that can be sent from user to user on the peer-to-peer Bitcoin network without the need for intermediaries. Transactions are verified by network nodes through cryptography and recorded in a public distributed ledger called the blockchain. So far, so good. Then I asked, what is the price of Bitcoin today? It says as of February 20th, 2023, one Bitcoin was worth approximately$1 ,840. I said, are you sure about that? Hugging chat says, I apologize if my previous response is incorrect.

11:15According to my records, as of February 1st, 2023, one Bitcoin was worth approximately$575. US dollars. So obviously there's a little bit of work to be done on Hugging Chat. But that doesn't mean that people aren't still really excited. Andrei Baranovsky, a developer, said tested Hugging Chat open source chat GPT alternative. Asked a question about Fast API. It works well and can provide code solutions. I don't know if the answer is the best, but the future looks promising. Dr. Jim Phan from NVIDIA says Hugging Chat, the open source 30 billion chatbot alternative to ChatGPT. Next step must be Hugging Chat apps.

11:51I think Hugging Face is in a great position to become the Android App Store. In fact, Hugging Face even has an edge over OpenAI. The apps can be other multimodal models already on Hugging Face. Eric Elliott quote tweeted that and reinforced the point. He says, right now, this is not a threat to ChatGPT. These models are not close to OpenAI's language model performance, but this could be exciting in the future when they get better models online and connect all the great Hugging Face models like plugins to the chat. What they're referring to is the fact that the main Hugging Face business is hosting other open source AI models that are available to anyone.

12:26This is really where the open source AI community lives. Tons and tons of projects use Hugging Face already, and that could mean it's a really easy path between Hugging Chat and all of those different tools which could take advantage of it and plug into it. But of course, this brings us back to the question of open source large language models in general and whether that's a good thing. Eliezer Yudkowsky recently discussed this with Lex Friedman and perhaps unsurprisingly was pretty against it. The whole notion of open sourcing, this was always the wrong approach, the wrong ideal. There are places in the world where open source is a noble ideal, but building stuff you don't understand, that is difficult to control, that where if you could align it, it would take time, that is not a place for open source because then you just have powerful things that go straight out of the gate without anybody having had the time to make them not kill everyone.

13:13Agree or disagree, those are the stakes of the conversation. And both sides, both those who are pro-open source LLM models and those who are firmly against them feel intensely and passionately about it. Whatever the debate, the relentless pace of technology marches on and I think that we're going to do nothing but see more of these models come to the fore. What do you guys think? Is the proliferation of these chat GPT open source alternatives a good thing? Let me know in the comments and until next time guys, peace!