SearchGPT, Gemini 1.5 Flash Public Release, and $12,3B for AI Startups
PLUS HOT AI Tools & Tutorials
Hello and welcome to the weekly recap!
Last week, I wasn't wrong when I stated that we were in for a new wave of releases in the artificial intelligence industry. Because it has started. Today, we have 5 (!) announcements at once (all from major players). There is also new proof that startups have room to grow, and a major leak about Runway and its models.
Let's figure out what's going on.
Thinking about becoming a solopreneur but don't know where to start? Check out these AI tools that will speed up your work and buy you a whole bunch of time:
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 🎟️
OpenAI Announces SearchGPT
Multiple leaks about the search engine from OpenAI have finally been confirmed. The company has announced the release of SearchGPT. It's an AI-based search engine that accesses real-time information from all over the internet. As you can guess, it's a direct competitor to Google's Perplexity and AI Overview.
By the way, AI Overview turned out to be a real disaster. Here's our post about it:
So, SearchGPT. The platform offers users a large text box with a single question: “What are you looking for?”. When you answer, the AI searches the web for information and then gives you the result. However, whereas with regular Google, you get a list of links to third-party websites, SearchGPT tries to organize and make sense of them. Yeah, about the same way Perplexity does it.
In one example from OpenAI, the search engine answers the question, when can the user see nudibranchs in Half Moon Bay this weekend? SearchGPT provides information about the location and the best time to see the unusual mollusks. The dialog does not end there: the user specifies the weather forecast and immediately receives a detailed answer.
All this time, the AI communicates in natural language as your personal assistant.
Of course, getting an answer to a specific question is not the only use case. In another example from OpenAI, the search engine summarizes its findings about music festivals and then briefly describes the events, followed by links to sources.
Currently, SearchGPT is just a prototype running in a test mode on the GPT-4 family of models. According to The Verge, about 10,000 test users can access the new AI. In addition, OpenAI is in talks with companies that will support and utilize the new search engine.
Keep your mailbox updated with key knowledge & news from the AI industry
OpenAI also separately noted that SearchGPT was designed to help users connect with publishers by prominently citing and referencing them in searches. Given the company's massive list of existing partners, including News Corp, The Atlantic, and many others, I assume the new model will do just fine with that.
When the full release of SearchGPT will take place is still unknown.
Well, we've waited. I expected OpenAI to be a bit bolder and open access to the public beta at the same time as the announcement. This would have allowed the company to win some more points against the backdrop of the failed Google AI Overview. In the meantime, we're only dealing with a potential (and very promising) competitor to Perplexity.
News Of The Week 🌍
Google Makes Gemini Chatbot Faster
Good news for those who don't want to pay for Gemini Advanced. Google has released an update for the free version of its chatbot. Starting this week, the Gemini 1.5 Flash model powers the chatbot. The company says it improves speed, context understanding, and image recognition. The AI is available in 40 languages and 230 countries.
I often turn to Gemini during the day if I want to perform some simple task (Usually out of curiosity). And, you know, every time I notice that it gives stupid answers, I head to ChatGPT. Maybe at least the new version of Gemini will be a bit more practical.
Meta Launches Its Biggest AI Model
As we wrote last week, on July 23, Meta introduced Llama 3.1, its most capable model to date. It contains 405B parameters, supports eight new languages, and extends the context length to 128 KB. Meta added that partners, namely AWS, NVIDIA, Databricks, Groq, Dell, Azure, Google Cloud, Snowflake, and others, were the first to gain access to Llama 3.
Meanwhile, regular users can also test the new model via WhatsApp and the Meta.ai platform. The developers especially recommend the chatbot's skills in math problems and coding.
Stability AI Releases Stable Video 4D
Stable Video 4D is a platform that converts video of a single object into multiple new videos from eight different angles/views. Stability reports that users can specify camera angles, tailoring the output to meet specific creative needs. It has a single inference, which generates five frames across eight views in about 40 seconds.
The new model from Stability AI is available now at Hugging Face. You can also learn more about how the technology works on the developer's website.
This is probably the first model in the last couple of months that really surprises and has potential for practical application. We’re already used to generating text, images, video, and audio. Now, we're dealing with three-dimensional objects. This sounds very cool for game designers and companies specializing in virtual reality.
Leak: Runway Trains its AI on Videos from YouTube
A new leaked internal document reveals that Runway's third-generation AI video generator collected thousands of YouTube videos and pirated movies for training. Of course, this happened secretly and without the creators' knowledge. In addition to content from independent bloggers, Runway used clips that were created by The New Yorker, VICE News, Pixar, Disney, Netflix, Sony, and others.
A pretty serious misbehavior for a company that raised $141 million from investors including Google and Nvidia and had a $1.5 billion valuation last year. And while average creators are unlikely to have anything to counter Runway, I wouldn't mess with Disney's lawyers.
Sharing is caring! Refer someone who recently started a learning journey in AI. Make them more productive and earn rewards!
Mistral Releases Large 2 with 123B Parameters
Now that's a busy week. The day after Meta's LLama 3.1 release, Mistral showed off its new flagship model. Like all the others, the developer promises the level of the latest advanced models in terms of code generation, math, and reasoning. The main problem that Mistral solves in this version is hallucinations in models.
The solution is quite elegant: if Large 2 can't find the exact answer to your question, it simply admits that it doesn't know something. It sounds a bit strange, but it seems like the right solution.
What's funny is that Large 2 outperforms Llama 3.1 405B in terms of code generation and math performance (which Meta emphasizes), and it does it with three times inferior parameters. Large 2 has only 123B.
VCs Continue to Invest Billions Into AI Startups
According to Crunchbase, from January to July 16, startups developing AI raised $12.3 billion. We are talking about 225 companies of different levels that create platforms for generating text, images, video, audio, and more. That's up from the same period last year.
Here's a breakdown of investments by stage:
198 angel/seed funding deals: $500 million.
39 early-stage deals: $8.7 billion
18 late-stage deals: $3.1 billion
If you have a cool idea for an AI startup but you're worried about not catching the hype train, don't worry. Judging by the momentum, investors will continue to invest in our favorite industry. It's likely that AI will set many new records this year.
Useful Tools ⚒️
QuizRise — AI Quiz Generator that makes learning easy
Tempest AI – Create an infinite game in 2 minutes with no code
Speech to Note – Turn your voice into written notes
AutoApply Jobs – Get hired faster with AI-driven job applications
Studymap AI – Learn almost any skill with the power of ML
I dislike learning new things, so I am constantly looking for tools that will make it easier. And that's what I found this week. StudyMap uses AI to create personalized study plans and helps you stay organized with your learning goals. The platform offers features like progress tracking, smart reminders, and even adaptive learning paths based on your performance.
Shall we give it a try?
Share this post with friends, especially those interested in AI stories!
Weekly Guides 📕
Claude’s AMAZING ‘Custom Instructions’ Feature! (Full Guide)
How To Develop AI Apps 100% For FREE!
How I make ORIGINAL AI videos for faceless YouTube channels [FULL COURSE]
How to Create Professional LinkedIn Profile Picture with AI
This FREE Image to Video Ai Can Control Expressions and Emotion
AI Meme Of The Week 🤡
Well, you know, things happen in life.
AI Tweet Of The Week
I guess even SearchGPT won't solve some of our problems.
(Bonus) Materials 🏆
Inside Mark Zuckerberg's AI Era | The Circuit
New Jersey’s $500 Million Bid to Become an AI Epicenter
Meet the Economist Who Thinks AI Will Help, Not Hurt the Middle Class
At the Olympics, AI Is Watching You
A.I. Can Write Poetry, but It Struggles With Math
Share this edition with your friends!
It’s like ChatGPT and Google had a genius baby! 🚀 Just wrote a fun article about why this could make Google sweat. I think Google might need some new tricks!
Sam and the truly talented team at OpenAI innately understand that for AI-powered search to be effective, it must be founded on the highest-quality, most reliable information furnished by trusted sources. For the heavens to be in equilibrium, the relationship between technology and content must be symbiotic and provenance must be protected.