We had a busy week in the world of AI news. Nvidia’s stock goes crazy…again, the job market is changing faster than anyone thought, AI and Hollywood continue to clash, more AI safety concerns, so many new launches from Meta, Midjourney, and OpenAI, an entirely AI-based social network called BeFake launches, and so much more. And towards the end, I’ll reveal the winner of this week’s AI video of the week. Sit back, relax, and enjoy this week's AI news!
MidJourney InPainting
First, let’s talk about some of the incredible AI launches that happened this week. Midjourney launched Inpainting, a highly-demanded feature that allows users to select a portion of an image and write a new prompt just for that section. Check out this example where the user replaces a lightsaber with a banana. This feature is in direct competition with Adobe’s Firefly product, which has had this since launch. As always, you can use Remix and other features to help fine-tune the output from inpainting.
Python in Excel
Next, Microsoft continues to impress me to no end. This week, they launched being able to run Python code directly from Microsoft Excel. Python is well-known as the go-to coding language for artificial intelligence, and Excel is the software that powers many businesses worldwide. With this update, you can now use the power of Python, including its top-tier and compelling data libraries, to run queries, create graphs and charts, and analyze your Excel data right from Excel. All you need to do is use the PY command, and you have access to many Python libraries quickly. One of the biggest headaches with Python is environment management, which is why I use Anaconda in every one of my tutorial videos. Microsoft realized this and partnered with Anaconda to allow for a standardized Python environment from Excel. The one potential downside of this awesomeness is that your Python code will run in Microsoft’s cloud, taking away a lot of user control. If you want to try it out, a preview is already available.
ChatGPT Fine-tuning
OpenAI launched fine-tuning of their GPT3.5 models, with GPT4 fine-tuning coming later this year. This allows users to customize their versions of ChatGPT. Open-source models have allowed for fine-tuning, but now, closed-source ChatGPT allows you to create models specific to your use cases. This allows for improved steerability, reliable output formatting, and, most interestingly, custom tones. I recorded a tutorial video I’ll be posting early next week, walking through exactly how to fine-tune ChatGPT, including how to create the dataset necessary to do the fine-tuning easily.
Facebook Translation & Code Models
I say this every week, but Meta continues to dominate open-source AI. This week, they launched an incredible translation model called Seamless. You can translate text to text, audio to text, text to audio, and everything between. It supports 100 languages and works well. The best part? It’s free. The only downside is it’s released on a research license, so it’s not commercially viable.
Meta launched a coding model based on LLama 2 called Code LLama. This is a fine-tuned version of LLama 2 specifically catering to coding use cases and will compete directly with Google, Microsoft, and OpenAI’s coding assistants. Meta is using a tried and true playbook of commoditizing their competitor’s strengths by releasing open-source model after open-source model, a playbook employed by Microsoft for years. Microsoft is finally getting a taste of its own medicine.
Health AI
AI continues to make strides in helping with health use cases, with the FDA clearing AI-powered software that pinpoints suspicious findings in chest X-rays. We’ve heard that AI can read medical imaging exceptionally well, and I can’t think why we wouldn’t want to leverage AI for this. Doctors and AI make mistakes, so why not let them check each other’s work and make doctor’s lives easier in the process? This should have many benefits, such as improving radiologists' performance, reducing their burnout, and driving down the cost of imaging. This is one of my favorite use cases for AI because it will substantially impact the world.
Car-Free City
This story is not about AI but is too awesome not to talk about. A new city is launching. Yes, you read that right, a new city. This city is remarkable because it is built to be 100% car-free. Located in Tempe, Arizona, this neighborhood was built by a company called Culdesac, entirely from scratch around alternative modes of transportation, like buses, trains, scooters, and bikes. Car culture dominates America’s cities, and as someone who grew up in Los Angeles, I would be beyond excited to ditch my car if I could. Everything you need to live is within walking distance: restaurants, cafes, grocery stores, gyms, bars and more. There’s no asphalt, plenty of open space, and gorgeous landscaping. I appreciate what the founder tries to do by making a city as walkable as many European cities are.
NYT lawsuit
Now, let’s move on to AI news, starting with some troubling legal times for the most prominent AI companies, starting with OpenAI. They’ve been getting hammered from a legal perspective, starting with a potential NYT lawsuit that could force OpenAI to wipe ChatGPT and start over. The suit will allege that OpenAI was trained on copyrighted material from the NYT. According to an article by Arstechnica, The result, experts speculate, could be devastating to OpenAI, including the destruction of ChatGPT's dataset and fines up to $150,000 per infringing piece of content. Just a month ago, another law firm filed a class-action lawsuit on behalf of comedian Sarah Silverman and other authors against both OpenAI and Meta, accusing the companies of illegally using their copyrighted material to train their models.
And, late last year, this same firm filed a suit against Microsoft’s Github for their Copilot product, alleging it, too, used copyrighted code to train its model. Once again, that same firm repeated its formula and filed suit against Stability AI, Midjourney, and DeviantArt over AI image generators. Although OpenAI and Meta don’t fully disclose the sources of their training data, the authors were able to figure it out by clues in the data and suspect OpenAI may be using 294,000 books downloaded from "notorious 'shadow library' websites. It’ll be fascinating to watch how these lawsuits play out. With both of these companies' armies of lawyers, I’d be surprised if they knowingly trained their models on copyrighted material. If Meta or OpenAI’s models were truly trained on illegally obtained copyrighted material, they would be in big trouble.
Back to the NYT lawsuit, according to NPR, OpenAI risks a federal judge ordering ChatGPT's entire data set to be rebuilt entirely, which would spell disaster for OpenAI and lead to an avalanche of similar suits from countless other content creators. As someone who had their videos ripped from YouTube and posted to TikTok without my consent and received millions of views…I know who you are…I understand how frustrating it can be as a content creator to have your work stolen.
OpenAI Hides Sources
And likely because of the legal threats it's facing, OpenAI’s newest ChatGPT version has started hiding its sources. In the last few months, large language model providers like Meta and OpenAI have stopped disclosing their sources of training data, but now, a new research paper suggests the models are taking it a step further, according to businessinsider.com: “ChatGPT now attempts to avoid responding to user prompts with exact phrasing from copyrighted works, according to a technical paper published August 8 by a group of AI scientists working for the research arm of ByteDance, the owner of TikTok.”
This is interesting, given Tiktok is a massive vacuum of copyrighted material, and they don’t seem to care or do anything about it. Also, according to the research paper referenced by BI, “In an effort to avoid showing it was trained on such material, ChatGPT now "disrupts the outputs when one tries to continuously extract the next sentence… which did not happen in the previous version of ChatGPT," the researchers wrote. "We speculate that ChatGPT developers have implemented a mechanism to detect if the prompts aim to extract copyright content or check the similarity between the generated outputs and copyright-protected contents.” Ultimate these cases will have to go in front of a judge because there’s a strong chance that there are enough snippets and quotes of pretty much every popular book in existence on the internet that these large language models could accurately rewrite books like Harry Potter without having been trained on the actual source material.
AI Gen Content Can’t be Copyrighted
And the hits keep coming. Just this week, a ruling was delivered that AI-generated art is not protectable by copyright law. According to hollywoodreporter.com: “A federal judge on Friday upheld a finding from the U.S. Copyright Office that a piece of art created by AI is not open to protection.” and Copyright law has “never stretched so far” to “protect works generated by new forms of technology operating absent any guiding human hand,” U.S. District Judge Beryl Howell found. This seems like a lack of understanding of AI. Not only did humans create and fine-tune these generative art models, but they guided them using prompting to create new art. To say that there was no human involvement seems false to me. Others may disagree, but I see AI as a tool in an artist’s toolbox like any other. An analogy that makes sense to me is Stability AI is the company that makes the paintbrush, Stable Diffusion is the paintbrush, and the person writing the prompt is the artist. Do you agree? Let me know in the comments.
Hollywood and Music Industry
The tension between Hollywood and writers continues to grow. As the writer’s strike continues, with one of the most significant issues being how AI will affect writing jobs, 96% of the biggest entertainment companies are boosting AI spending. And this isn’t only in the US. In China, 100% of entertainment companies are increasing AI spend, similar to India and the UK.
And not only movies and TV are being disrupted by AI. Music seems to be actively adopting AI. This week, YouTube announced a partnership with Universal Music Group to create an AI framework for music based on three fundamental AI principles: 1) AI is here, and we will embrace it responsibly with our music partners. 2) AI is ushering in a new age of creative expression, but it must include appropriate protections and unlock opportunities for music partners who decide to participate. and 3) We've built an industry-leading trust and safety organization and content policies. We will scale those to meet the challenges of AI. As mentioned in a previous video, I’m impressed by the music industry’s adoption of AI rather than trying to fight it legally, which failed them in the past when previous technologies disrupted their industry.
Social Media and AI
The world of social media is changing in the face of AI. Snapchat’s AI bot was all over the news this week because it glitched hard. Since its launch, the in-app digital assistant has seen numerous bugs and breakdowns, but now it’s going rogue by posting videos to user’s feeds without their approval. As first reported by Techcrunch, the My AI feature posted a 1-second video story and then stopped responding to users’ requests.
BeFake
Since AI is already posting on behalf of humans, why not go all in on this concept? Well, a new social network called BeFake did just that. The App, started by former Machine Zone CEO Kristen Garcia Dumont, is a social media app for digital self-expression, with the tagline “Why be real when you can be fake?” To be honest, I kind of like this idea. Social media apps are supposed to be all about reality, even though they are highly curated portions of people’s lives and generally have lots of gloss and…fakeness. So let’s stop pretending and own being fake. BeFake is all about self-expression and creativity, where being fake online is welcomed.
Nvidia
Now, let’s talk about the business of AI, which is booming for the most part. Nvidia crushed their Q2 earnings, sending their stock soaring in after-hours trading. Demand for their chips, which powers much of the AI models out there, is driving their 100% increase in revenue from 1 year ago. Nvidia is now the 5th most valuable company in the world. Nvidia was chugging along until they were in the right place at the right time to power this wave of generative AI.
Jobs & AI
The jobs market is being affected by AI every day. First, LinkedIn says ChatGPT-related jobs have ballooned 21x since November. Before then, ChatGPT wasn’t known, so it’s no wonder there has been a massive increase in ChatGPT-related jobs since then. Many people are scared about AI replacing their jobs, which is bound to happen, but as with every other technological revolution, many more new jobs will be created. You’re already ahead of the curve if you're watching this channel.
And adding on to that trend, a new IBM study published says 40% of workers will need to reskill in the next three years due to AI, which equates to about 1.4 billion people out of the 3.4 billion-person total workforce. However, the silver lining is that 87% of executives polled for this study believe generative AI will augment roles rather than replace them. Also, people who successfully reskill with AI will earn more. One of the most salient quotes from the report was: "AI won't replace people—but people who use AI will replace people who don't.”
Goldman Sachs Report
The last story about the business of AI is a new Goldman Sachs investment memo that forecasts AI to approach $200 billion in global investment by 2025. The real question is, how much of that is going straight to Nvidia? Is Nvidia still the best AI stock to own right now?
AI Video of the Week
Now for the AI video of the week! This week’s theme is terrifying. Check out the video. I’m unsure how this video was made, but X user Javi Lopez @javilopen created it. Excellent work. I’m now going to sit in a fetal position for the rest of the day.
AI “Kill Switch”
Our last set of stories is about AI safety. Meta has confirmed they will include an AI off switch in their content feed ahead of the August 25 deadline in the EU for the Digital Services Act. According to the TechCrunch article: “Under the DSA, users of larger platforms — 19 of which the EU designated back in April — must be offered a choice of a non-algorithmic feed, where content sorting is not based on tracking.” I suspect most people like having a non-algorithmic feed, but when they see how irrelevant most posts are to them, they’ll switch the AI back on quickly.
X Bots
It seems X is struggling with bots still. Even though Musk continues to slam Twitter’s original management for how easy it is to prevent bot networks, he realizes how wrong he was. Now, with the power of ChatGPT and other AI, bot networks have become so good that they are increasing throughout the social network, spreading crypto scams.
Musk recently claimed the best way to stop bots was with subscriptions, which makes sense. They won't be run if you make it financially infeasible to run a bot network. In addition to subscriptions for people who can’t or don’t want to pay X, Musk has been working on a user verification system. Using government-issued IDs and selfies, X hopes their new verification system will help quell the bot problem. Take a look at these screenshots of the feature in action.
