Midjourney Open Now, GPT-4o Fine-Tuning, and Top 100 GenAI Apps

PLUS HOT AI Tools & Tutorials

Aug 23, 2024

Hello and welcome to our weekly roundup!

So, today, we have a lot of good news and one pretty bad conclusion for founders of AI startups. I suggest we start with a nice release from OpenAI, pay attention to Midjourney on the web, and take a look at the best startups according to a16z. Then, turn to the podcast, in which Cohere CEO Aidan Gomez gave his opinion on the state of the AI industry.

Shall we begin?

What do you think about creating a startup for sale? Or maybe you are looking for a business to acquire? Either way, this post is for you:
Buy & Sell AI Startups: from $1K TO $100M
Stepan Ikaev
·
Aug 22
Read full story

This Creators’ AI Edition:

Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁

Featured Material 🎟️

Fine-tuning Now Available for GPT-4o

OpenAI is finally launching fine-tuning for GPT-4o. Developers can now customize the model with custom datasets and get better performance for specific use cases at a lower cost. A nice bonus with this release: OpenAI offers 1 million training tokens per day at no cost to each organization until September 23.

From coding to creative writing, fine-tuning can have a large impact on model performance across a variety of domains. This is just the start—we’ll continue to invest in expanding our model customization options for developers.

To demonstrate what GPT-4o fine-tuning can do, OpenAI has been working with other developers over the past few months. In this way, the partners have identified several use cases and are now sharing their progress.

The first partner on the list is Genie by Cosine. As OpenAI explains, it’s an AI software engineering assistant that can autonomously identify and resolve bugs, build features, and refactor code in collaboration with users. It can reason across complex technical problems and make changes to code with higher accuracy and fewer tokens needed.

With a fine-tuned GPT-4o model, Genie achieves a SOTA score of 43.8% on the new SWE-bench, a verified benchmark announced last Tuesday. Genie also holds a SOTA score of 30.08% on SWE-bench Full, beating its previous SOTA score of 19.27%, the largest improvement in this benchmark.

Another partner was Distyl, which ranked 1st in the BIRD-SQL benchmark. The tuned GPT-4o allowed it to achieve 71.83% execution accuracy in the leaderboard and excel in tasks such as query reformulation, intent classification, chain-of-thought, and self-correction, with particularly high performance in SQL generation.

Sounds great.

Keep your mailbox updated with key knowledge & news from the AI industry

How To Use It

As OpenAI explains, GPT-4o fine-tuning is now available to all developers on all paid usage tiers.

If you want to try the new features, visit the fine-tuning dashboard(opens in a new window), click “create,” and select gpt-4o-2024-08-06 from the base model drop-down. GPT-4o fine-tuning training costs $25 per million tokens, and inference is $3.75 per million input tokens and $15 per million output tokens.

Of course, you can also fine-tune the GPT-4o mini in addition to the basic model. To do so, visit the fine-tuning dashboard and select gpt-4o-mini-2024-07-18 from the base model drop-down. For GPT-4o mini, OpenAI offers 2M training tokens per day for free through September 23.

Regarding data privacy and security, OpenAI noted that fine-tuned models remain entirely under the user's control, with full ownership of business data, including all inputs and outputs. This way, the company ensures that your data is not shared or used to train other models.

Now GPT-4o can be called a truly complete product. By the way, if you're thinking about testing the fine-tuning, you'd better do it soon, as long as it's free. After September 23, the new service will cost $25 per million tokens, while using the customized model will cost $3.75 per million input tokens and $15 per million output tokens.

News Of The Week 🌍

Midjourney Releases Web Image Generation

For now, the inconvenient Discord. The company has launched an updated image generation platform. It is available in any browser and allows you to generate 25 images. To use Midjourney, you only need to sign up with a Google or Discord account. The web version of Midjourney will generate four pictures at once. You can choose the one you like, edit it a bit, and save it. Here's what you need to do once you log into the platform:

Click “Create” in the upper left corner.
Type the prompt in the line at the top and press Enter.
Wait for the generation (it takes about 7 seconds).

I have nothing against Discord (a nice app), but using Midjourney through it is annoying. So the fact that it can now be used via the web is excellent news.

a16z Updates Top 100 Best GenAI Apps

There are lots of graphs, beautiful pictures, and charts. Andreessen Horowitz (a16z) showed a new list of top consumer genAI apps for the second half 2024. An interesting feature of this ranking was the number of newcomers.

This time, almost 30% of the companies were new, compared to the previous March 2024 report. Here, by the way, are the 50 most popular apps among users:

As for newcomers, the current list includes Luma AI, Viggle, SeaArt AI, Udio, ChatBot App, and a few more cool apps.
You can find the full report with all the data here.

Cohere Founder Speaks Out on Business Models in the AI Industry

Cohere founder and CEO Aidan Gomez visited the 20VC podcast, where he talked a lot about AI and how this market is evolving. In particular, he says that selling access to models is quickly becoming a “zero-margin business.” For now, these AI models cost more than they make, which is a big problem for independent developers.

To learn more, I highly recommend watching the entire podcast:

(Fortunately, it's not eight hours long)

If you briefly summarize what Gomez is talking about, the problem is that modern models are very cost-intensive. And while big companies like OpenAI, Google, and Microsoft can offset the costs, for startups, this situation looks pretty dire. Probably some of the best ways out for them would be to sell their business to bigger players.

Sharing is caring! Refer someone who recently started a learning journey in AI. Make them more productive and earn rewards!

Refer a friend

Luma Labs Unveils Dream Machine 1.5 Update

Startup Luma Labs confirms that it will not stop on its achievements and plans to impose competition to Runway Gen-3 and Kling AI. The other day, the company updated its Dream Machine 1.5 platform. At first glance, you won't notice any differences, but in reality, the developers have significantly improved their model's capabilities.

Now, Dream Machine perceives user prompts more accurately, creates logos and graphics, and generates more realistic videos with people and other objects. The system is also noticeably faster: a five-second video takes less than two minutes to generate.

It's nice to see new players jumping on the hype train, but not getting lost to the abundance of competitors. Let's hope we hear about Luma many more times. Even after the full release of OpenAI's Sora.

OpenAI Partners with Conde Nast to Improve ChatGPT and SearchGPT

Condé Nast—the publisher of storied outlets such as The New Yorker, Vogue, and Wired—is OpenAI's next big partner. The companies did not disclose the partnership terms, but the agreement's basis is clear. OpenAI will use publications from popular outlets for licensed training for its ChatGPT and SearchGPT platforms.

In addition, Condé Nast CEO Roger Lynch hinted in a memo that the “multi-year” deal will include payment from OpenAI in some form.

I'm starting to feel like I'm in Groundhog Day, but come on. Let's run through the current list of media outlets OpenAI works with. It now consists of the Associated Press, Axel Springer, The Atlantic, Financial Times, News Corp, Time, and Conde Nast. Who's next?

Useful Tools ⚒️

Paperguide – Discover, read, write, and manage research with ease using AI

Selix AI – Give your outbound idea - watch your AI execute it

Hexus AI – Your AI creator for engaging product demos & docs

Evidently AI – Open-source evaluations and observability for LLM apps

Myko Assistant – Research assistant for web scraping and finding contacts

Myko Assistant is a tool designed to streamline research and data analysis. It provides accurate, verified information on various topics, from potential clients to company research. With this platform, you can access the data you need, save time, and boost your productivity. It can be a valuable asset for professionals looking to stay ahead in today's fast-paced business environment.