The 5 Most Important Stories in AI This Week

0:00Today on the AI Daily Brief, we are counting down the biggest stories in AI this week. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:17All right, friends, quick note. Today, we are doing a little bit different type of episode, a countdown of the top stories of the week episode. That means no headlines versus main, just diving right into the meat of the show. And the way we're going to do this is not actually counting down five stories from five to one. we're actually going to divide it into different categories, and our first category is the most exciting product releases of the week. For me, this has to kick off with MidJourney, which is finally open to everyone on the web. For years, MidJourney has been a Discord-only experience, and over the last six months or so, they've been gradually letting more and more people onto their web-based interface, which is a custom-designed web application for generating and editing MidJourney images.

0:55The web-based experience massively decreases the barriers to entry, and taking that even farther, the company is also offering free trials, which is something that is also different for them. Midjourney continues to be the AI product that I use the most, and so I'm excited personally for more people to have a chance to check it out. However, Midjourney is not the only image generator that's generating conversation this week. In fact, if anything, the most discussed image generation release this week is Ideogram's version 2.0. The specific reason that Ideogram is catching so much attention is how good it appears to be at handling text.

1:28Text generation has been the Achilles heel of image generators for some time. It's still something, for example, that Midjourney really struggles with. But Ideogram seems to nail it as this thread of advertising-related prompts visually shows off. In this generation, a stylish backpack hanging from the edge of a mountain cliff with a breathtaking view of the valley below. The text Adventure Awaits is written in bold adventurous letters on the top. The tagline Carrier World is placed below, emphasizing durability and style for those who seek adventure. and it looks like a professional advertising image.

1:57There are a bunch more examples of this, and it really does represent a sea change moment for image generation if this capability for text generation in images is starting to become more commonplace. One more product release, which is somewhat related, Luma Labs has released their Dream Machine 1.5. This is a quick update since Dream Machine is only a couple months old at this point, and already people are stacking these technologies together. Take for example this Cthulhu-themed website animation, where an ideogram image with this incredibly cool text was then animated with Luma Labs. Our second category is coolest model developments.

2:31And in this category, once again, we have two different topics. First of all, fine tuning is now available for GPT-4.0. OpenAI says that this was one of the most requested features from developers and basically allows people to now fine tune the state of the art. As an additional incentive, OpenAI announced that they're offering a million training tokens per day for free to every organization through September 23rd. Now, to be fair, some people are skeptical of this. Santiago at SVPino writes, Fine-tuning is now available for GPT-4.0 and GPT-4.0 Mini, but I need somebody to help me understand why anyone would use this.

3:03Fine-tuning is hard, and it takes a significant investment from a company to get it right. Why would a company spend all of that time to fine-tune a model that OpenAI controls? You can't download the model from OpenAI, you can't host it yourself, etc., etc., etc. From what I've heard from people, and this has surprised many, the simple convenience of operating inside OpenAI's ecosystem simply overwhelms a lot of these other concerns. Whether that continues to be the case remains to be seen, but for now, this is a big update that many will be excited to take advantage of. The other model development that I thought was interesting, which is something we talked about on yesterday's show, was Microsoft's release of three new Fi 3.5 models.

3:38These are all smaller models, comparatively speaking, with really impressive performance. And to me, the interesting thing about this is how much the competition in AI is not just at the far ends of the biggest model capabilities, but also in terms of finding ways to do more with less. Driving the pursuit of the small model competition is, I think, commercial concerns. Companies want to be able to run models on devices, and they want to be able to offer capable models more cost-effectively. These are two major considerations when it comes to model adoption, and so in many ways it's not surprising to see them become a bigger area of focus for many of the big labs.

4:12Third, we move into the biggest stories in enterprise AI, and I think the leader here has to be Salesforce announcing two new agents. The company announced the Einstein SDR agent and the Einstein sales coach, and to me this represents an example of how fast AI is changing even in the enterprise sector. Companies haven't even really wrapped their head around generative AI that's used and maintained by their employees to say nothing of these new agent capabilities, but that doesn't mean the technology is slowing down. Salesforce's arguments with these agents is basically that no sales team has enough pipeline, which is a fairly compelling pitch.

4:45And the purpose of these agents is then to not only increase pipeline, but in the case of the sales coach, also help people better convert. One of the interesting things to watch for with announcements like this is how providers like Salesforce position their technology vis-a-vis existing humans. It's very clear for Salesforce that this is not about replacing people, but about rapidly scaling individual capacity in a context where, as they put it, there basically is no such thing as too much capacity. There are always more sales leads.

5:36Venice is an alternative. Venice is a powerful AI app for text, image, and code generation that respects you as a sovereign individual and believes privacy and free speech are not only human rights, but necessary for civilizational advancement. Private, permissionless, and uncensored, you can try it for free without an account. AI Daily Brief listeners receive a 20 % discount on Venice Pro. Visit Venice.ai slash NLW and enter the discount code NLWDailyBrief. That's NLWDailyBrief. All one word. Today's episode is brought to you by Super Intelligent, which is, of course, our platform that helps you learn how to use AI tools and perhaps even more importantly, gives you ideas on the best use cases that are actually going to help you achieve whatever it is you want to achieve.

6:22To recognize the end of summer and back to school slash back to work, we are running our best promotion ever. When you sign up for Super Intelligent between now and the end of August, using code SOBAC, your first month will be 100 % free. The platform features over 600 fun, highly practical AI tutorials that get you using AI fast and with an eye to actually transforming how you get things done. We've just launched Super for Teams, so if you have a group of people at your company that want to figure out how to use AI together, I highly suggest you check it out. But for those of you who are using Superintelligent as an individual, once again, if you sign up for super intelligent between now and the end of the month using code SOBAC, you will get your first month 100 % free.

7:04Go to besuper.ai and check it out today. Our fourth topic is our biggest debate of the week, and this will be a surprise to no one. It continues to be California's controversial AI safety bill, SB 1047. Last we discussed it, a group of California congresspeople had sent an open letter to Gavin Newsom asking him to veto the bill, and former House Speaker Nancy Pelosi had added her weight to that position. Since then, we've had two developments from the big labs themselves. OpenAI sent an open letter to California State Senator Scott Wiener, writing, The AI revolution is only just beginning, and California's unique status as the global leader in AI is fueling the state's economic dynamism.

7:43SB 1047 would threaten that growth, slow the pace of innovation, and lead California's world-class engineers and entrepreneurs to leave the state in search of greater opportunity elsewhere. Wiener bit back quickly, arguing that OpenAI's letter quote, doesn't criticize a single provision in the bill, and also saying that the threat that AI talent might leave the state doesn't make any sense because the law would apply to any companies that conduct business in California, no matter where their offices are actually located. Meanwhile, Anthropic, which has been a little back and forth on this bill, but which of course is much more aligned philosophically with the AI safety side of the space that was intimately involved in originally drafting it, has come out in saying while they're not fully endorsing it, they think the benefits outweigh the costs.

8:23Anthropics Jack Clark wrote, You can read the letter for the main details, but I'd say on a personal level, SB 1047 has struck me as representative of many of the problems society encounters when thinking about safety at the frontier of a rapidly evolving industry. How should we balance precaution with an experimental and empirically driven mindset? How does safety get baked into companies at the frontier without stifling them? What is the appropriate role for third parties ranging from government bodies to auditors? These are all questions that SB 1047 tries to deal with, which is partly why the bill has been so divisive.

8:51These are complicated questions for which few obvious answers exist. Nonetheless, we felt it important to give our view on the bill following its amendments. In a section where they discuss high-level principles for regulating AI, they write, The key dilemma of AI regulation is driven by speed of progress. AI technology continues to advance extremely rapidly. On one hand, this means that regulation is urgently needed on some key issues. We believe that these technologies will present serious risks to the public in the near future. On the other hand, precisely because the field is advancing so quickly, strategies for mitigating risk are in a state of rapid evolution, often resembling scientific research problems more than they resemble established best practices.

9:25We believe that this genuinely difficult dilemma is one important driver of the divergence in views among different AI experts on SB 1047 and in general. The resolution that they suggest is quote, very adaptable regulation. They said we've come to the view that the best solution is to have a regulatory framework that is very adaptable to rapid change in the field. They write that there are several ways to accomplish this, including via third-party auditors, frameworks that shape incentives without prescribing behavior, or procedural requirements that require a safety process without describing what is in it.

9:52Down the road, perhaps in as little as two to three years when best practices are better established, a prescriptive framework could make more sense. I understand where they are coming from with this, but boy is that hard to instantiate in real policy. Lastly, and this is really important, they write catastrophic risks are important to address. And once again, as I've said before with this debate, this is the central point of disagreement, and why this discussion is never getting beyond first gear. By and large, the people that are most for this bill believe this, that catastrophic risks are important to address and important to address now.

10:23For many of the people that oppose this bill, they fundamentally disagree with that. They believe that the catastrophic risks simply don't exist in the way that the proponents do. And when you don't believe in the fundamentals of a risk, of course you're not going to support regulation, even light-touch regulation, to deal with that risk. Now, I don't know what's going to happen with SB 1047, But to the extent that we are trying to take inspiration from this in terms of how we might want to approach this conversation on a national level, it seems to me like the clear and obvious thing here would be to separate this discourse into two different dimensions.

10:53Let the first be the risks that everyone agrees on. Build consensus around issues where there's little disagreement. For example, in this bill, whistleblower protections are something that basically everyone agrees on. I haven't seen any real disagreement with that sort of thing. Build a federal framework around the stuff where there is broader agreement, and then intentionally tackle the question of catastrophic risk. Instead of the two sides that have already made up their minds on this issue squawking at each other and screaming past the 99.99 % of people that are in the middle, let each side actually make their case coherently and in a policy context.

11:26Probably too idealistic to hope for, but there it is. Lastly today, the most interesting idea of this week comes from the blog post that we read yesterday from VC Sarah Tavill, where she reminds us that the incredible competition playing out between different model providers is to the huge benefit of the end users and consumers, and also, frankly, startups that get to build new ecosystems. You certainly see that with the speed with which the foundation model companies are putting out new models. But even given what we talked about with MidJourney, Ideogram, and Luma all releasing new products this week, you're seeing it everywhere.

11:59Part of what makes this space so fun to be in is the speed at which it's changing, which is driven by competition. And so even as we have important debates around safety and the future of the world, let's not forget we're also living through an amazing capacity expanding period right now. Anyways, friends, that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.

12:32Thank you.