Hey everyone, welcome to another installment of What’s New in AI, a weekly newsletter summarizing the trends happening in the world of artificial intelligence. Lots of news this week, from investments to GPT authors to new AI model training. Let’s dive in:
The OpenAI - Microsoft Deal
The hottest news of the week comes from an actively developing OpenAI - Microsoft investment deal. Over the last few days, multiple sources have been reporting on terms being negotiated between the startup and the behemoth.
Microsoft will close a $10 billion investment deal into OpenAI before the end of this month.
Microsoft will contribute a total of $13 billion in capital to OpenAI, underscoring how important the technology behind ChatGPT and DALL-E 2 is to its future.
Documents show that should OpenAI's technology become successful and profitable, Microsoft would be able to make as much as $92 billion from its collective investment.
Venture capitalists that participate in the tender offer of employee shares could receive up to $150 billion.
OpenAI projected a loss of $508 million in 2022.
(Source)
Microsoft is making a huge bet on OpenAI if this is true. That said, it makes sense as they continue to plan the rollout of GPT into their core suite of consumer tools such as Word and Outlook.
(Source)
While at a loss in 2022, OpenAI seems to be making moves on the revenue generation front. TechCrunch reported that OpenAI has begun piloting a professional premium version of ChatGPT.
(Source)
Let’s see how the deal pans out in the coming weeks, but seemingly OpenAI is about to get a lot more cash as Microsoft doubles down on their bet.
Another AI Unicorn
Venture dollars are seemingly continuing to pour into the AI startup ecosystem. DeepL, an AI-based language translator startup, just announced that they raised over $100M at a $1B+ valuation, which means another AI unicorn.
The startup is also not confirming or disclosing other financials, but the investor source said that the $1 billion valuation was based on a 20x multiple of DeepL’s annual run rate, which was at $50 million at the end of last year. In the current fundraising climate, this is a pretty bullish multiple, but it speaks to the company’s growth, which the investor noted is currently at 100%, and the fact that DeepL’s breaking even and close to being profitable.
(Source)
Other AI companies also announced their rounds this week, including Inbenta raising $40M and Seek AI raising a $7.5M preseed + seed round.
Clearly AI is the new hype market, the big question is how long it lasts.
Wolfram Alpha with GPT
Stephen Wolfram, CEO of Wolfram Research, wrote a post on his thoughts of bringing computational knowledge superpowers to ChatGPT with Wolfram|Alpha.
There are all sorts of exciting possibilities, suddenly opened up by the unexpected success of ChatGPT. But for now there’s the immediate opportunity of giving ChatGPT computational knowledge superpowers through Wolfram|Alpha. So it can not just produce “plausible human-like output”, but output that leverages the whole tower of computation and knowledge that’s encapsulated in Wolfram|Alpha and the Wolfram Language.
(Source)
Combining the Wolfram|Alpha computational language / powers along with GPT models would seemingly create a very powerful LLM. In fact, this has already happened this week in the LangChain project. The open source project aiming to combine LLMs with other sources of computation or knowledge just added Wolfram|Alpha as part of their toolset to integrate with GPT. The speed of the AI community is amazing to watch.
Research Paper credits GPT
ChatGPT was recently listed as an author on a research paper exploring the use AI for for medical education. The study utilized the capabilities of ChatGPT to take the USMLE medical examination test and achieve a passing grade. Apart from the fact that this test can pass an exam that medical students study days for, the other interesting outcome is that GPT wrote at least some of the research paper.
The resulting implication of listing a language model as an author on a research paper is potentially huge. There may be a shift coming in our understanding of authorship, authorship criteria and the role of AI in the scientific process.
Using AI for training AI
Recent tweets from Elicit and @mathemagic1an highlighted an approach for training AI models. They used a language model to generate synthetic training data for another AI model. This approach, known as RLAIF, allows resource-strapped startups and research groups to leverage AI to create more data for training, potentially increasing their efficiency and effectiveness.
Fun Bites
Actor and genius marketer Ryan Reynolds uses ChatGPT to write an ad script for his company, Mint Mobile.
Broadn.io just released a search tool for podcasts (with a focus on Lenny’s Podcast first). It’s an interesting implementation of search through spoken content using AI.
For those working with ChatGPT prompts, @LinusEkenstam posted a great Twitter thread on how to optimize prompts to get to a better answer faster.
Flair AI just announced a waitlist for their AI powered product photo shoot technology. If you’re interested in trying it out, comment on the tweet below.
That’s all the highlights for this week. Check out my Twitter list of AI builders who I’m following.
We also have a discord community! If you want to participate in conversations relating the latest developments in AI, join here: https://discord.gg/kApBrN8uqv