We've changed- Discovery Unlocked 2.0
A new theme, format, frequency and tone. Also, AI and a lot of it.
Welcome to Discovery Unlocked (version 2.0). Thank you for noticing that we look a bit different. We are trying a new theme, format, frequency, and tone for our newsletters from this week on.
AI is evolving rapidly, and that makes it hard to keep pace with the developments. We cut the noise by curating only the most relevant information. 🚀
If you haven’t already subscribed, now will be a really good time.
Every Friday we will come to you with the following-
(we won’t spam you with another 5000 writing apps, promise)
🛠️ Tools- bringing some of the latest and best AI tools and products being put out in the market
🐣 Thoughts- hot takes, ideas, tweets, blogs, demos, and more that capture the best ideas in the AI space
📑 Developments- news, research, and partnerships, that give a sneak peek into this space
🥷 Random- well, let’s just say this space can be anything
So, here we go..
🛠️ Tools
What if we could use technology to augment our memory? [Product]
Born with a vision to give humans perfect memory Rewind is a macOS app that enables you to find anything you’ve seen, said, or heard. 🤯
a16z's
in his post on Investing in Rewind writes-We’re constantly context switching our brains and the apps we use. Inevitably, we’ve all asked ourselves: “What was I looking at when I was in that meeting last week when someone was talking about the forecast. I know it was important and I wanted to come back to it…darn! I can’t remember”
Prompt Promptly [Hack]
We are going out on a limb and assuming that you would have come across this new thing on the market (TBH old now) called ChatGPT. The effectiveness of using ChatGPTs (or other LLMs) is subject to writing good prompts.And,
has condensed it into a set of steps-Understanding and setting the right context for your prompts
Giving the model a task to complete- Once the context is established, give the model a clear task to complete.
Asking questions- ask specific questions within the scope of the conversation
Considering the output, if it’s not what you were expecting or lacks detail, refine the prompt
Refining the prompt to get better results
I am lazy. Let GPT improve my prompts. [Product]
Metaprompt generates better prompts from low-level prompt input.World’s first self-organizing workspace- Mem [Product]
Take notes and forget about what folder or app it “should” go in. Mem uses AI to connect and organize your stuff along with events, topics, and people.
Sneak peek into MemX- an AI extension of Mem that leverages recent breakthroughs in natural language processing to actually understand what you capture.
Ellie, an AI assistant that send emails for you [Product]
Ellie is an extension that learns from your writing style and crafts replies as if they were written by you. Just add the extension to your browser (Chrome and Firefox supported) and you can quickly craft personalized messages in your own tone, effortlessly. Oh and Ellie is multilingual, it will reply in any language to match the email being replied to.
🐣 Thoughts
@sama on ChatGPT being used for cheating [Interview]
“We adapted to calculators and changed what we tested for in math class, I imagine. This is a more extreme version of that”- Sam Altman, CEO of OpenAI (the company behind ChatGPT)Read snippets from his conversation here, and watch the full interview here.
ChatGPT is 'not particularly innovative,' and 'nothing revolutionary'- Meta's chief AI scientist [Hot take]
Meta's Chief Scientist Yann LeCun in a recent press gathering stated that ChatGPT is just a “Well-Put Together AI System”, despite the buzz around OpenAI's language model, ChatGPT, the technology behind it is not revolutionary and similar systems have been built in the past by other companies and research labs.
Some cool advice for building in AI from the founder of LEX [Blog]
Few quips that we liked-
AI is a new raw ingredient to help you solve some customer problems that weren’t solvable before.
Products using AI aren’t necessarily ‘wrappers’. Calling AI-powered app wrappers would be like calling toasters and Teslas “wrappers for electricity.”
AI technology isn’t anyone’s moat—but that doesn’t mean moats won't be built.
AI-powered applications are mostly not about AI, they are still about - talking to users, understanding your data, and running growth experiments.
📑 Developments
ChatGPT Pro for $42 per month [News]
OpenAI said earlier this month it would be testing a faster, paid version of the AI chatbot, but it’s not clear if the company will stick to the $42 pricing.MSFT 🤝 OpenAI [Partnership]
OpenAI and Microsoft are taking their partnership to new heights with a whopping multi-billion dollar investment from Microsoft! Msft will increase investment in supercomputing systems and Azure will remain the exclusive cloud provider for OpenAI. The companies will also work together to build and deploy safe AI systems and integrate OpenAI technology into apps like GitHub Copilot and Microsoft Designer.
CheatGPT [News]
This week we saw the news about ChatGPT acing an MBA exam at Wharton, and that it could also clear the medical licensing examination USMLE (which is a three-part exam that usually takes students about four years of med school and about two years of clinical rotations to pass). With these developments, there has been a growing concern in the academic community about students using ChatGPT to cheat on tests and homework.
Cheating on homework should be the least of our concerns- POSSIBLE SOLUTION [Research]
LLMs like GPT can write like humans, and this poses a risk of being used for social engineering and election manipulation campaigns that exploit automated bots on social media platforms, the creation of fake news and web content, and the use of AI systems for cheating on academic assignments. Furthermore, the rise of this synthetic data on the web complicates future dataset creation efforts, as synthetic data is often inferior to human content and must be detected and excluded before model training. Hence detecting machine-generated text becomes super important. A team of University of Maryland researchers has published a paper that posits “Watermarking” the machine-generated content for ease of traceability. [Read the paper here.]
Atomic AI raises $35 million for AI-driven RNA drug discovery. [Funding]
AI-powered tools are helping in the complex process drug discovery. The company has raised a $35 million A round, led by Playground Global, with participation from 8VC, Factory HQ, Greylock, NotBoring, AME Cloud Ventures, as well as angels Nat Friedman, Doug Mohr, Neal Khosla, and Patrick Hsu. (The company previously raised a $7 million seed round.)
🥷 Random
ChatGPT does not understand Bollywood 💔😟
We asked ChatGPT about Bollywood classic Sholay, and let’s just say “Mogambo khush nahi hua 😶” (translation- Mogambo isn’t happy)
Did you like the new format? Let us know how we can improve.
We are eager to listen and learn. Do let us know things you would want us to cover and things you would want us to change. Email us, ping us on Twitter (Pratik & Tanmay), or simply comment below.
Also, do tell your friends about us.
See you next Friday.
-P&T