[14th April 2024] Interesting Things I Learnt This Week

 

1. Opera adds built-in support for local LLMs - Opera is adding local AI models to its browser. Users can choose from 150 local LLM variants, keeping their data private on their device. This is part of Opera’s AI Feature Drops Program for early adopters.

My Take : Locally running LLMs will become a commonplace very soon. It requires not just software but hardware support as well. On older hardware it might takes ages to get anything done. But on newer hardware it will be good I guess. The best software to get them integrated is going to be the browser for most folks IMHO . I was hoping Firefox to be the first one to do it, but I wish they work towards it. I have no hope of any of the big personal computing softwares to be doing it. They will push for only their models to be running and working across devices, also trusting them to not harvest data off it, would be a challenge. Never the less, integrating LLM APIs in browser will open up some interesting avenues for web applications. Not sure if it will be taken up or not, but is interesting for sure. 

2. OLMo - A new, open-source large language model (LLM) called OLMo is designed to be a foundation for researchers to study and improve LLMs. AI2, the company that created OLMo, hopes this will lead to more responsible AI. Some of the important points are that OLMo is open source, includes the training data, and is built on a massive dataset of text and code. 

My Take : This is a fantastic step forward! By opening up access to LLMs, researchers everywhere will be able to gain a deeper understanding of how these models work. This will have a ripple effect, leading to improved prompt engineering – essentially better ways to communicate with LLMs and get more accurate and useful results. We can also expect increased predictability in LLM outputs as researchers gain a clearer picture of how these models make decisions. This deeper understanding might even pave the way for faster LLMs in the future.  The potential benefits extend to efficiency as well. Researchers might discover ways to reduce the memory footprint of LLMs, allowing them to run on local machines for specific tasks. This would be a game-changer, making these powerful tools more accessible by letting folks run it locally. 

3. ProtonMail is not sending emails - Author of this post realised that others can email him, but only if they don't use ProtonMail to encrypt the email. If they use ProtonMail and your email address is listed in a directory called WKD (Web Key Directory), then the email will be encrypted and you won't be able to read it. They fixed it by removing their key from WKD.

My Take : This post is a fantastic example of detective work! It really impressed me how the author was able to identify the problem and solve it. It also introduced me to the concept of encrypted emails, something I never knew existed before. This discovery has sparked my curiosity, and I'm definitely planning to learn more about encrypted emails in the future.

4. Monzo's Impersonation Prevention - This article talks about technique Monzo used to combat impersonation scams. Monzo's solution is a call status feature within their app that allows users to verify if they are speaking to a real Monzo representative. If a user is not speaking to a representative, the app will instruct them to hang up and report the call.

My Take : This is a simple yet powerful approach that all banks should adopt. I've received suspicious calls claiming to be from my bank, and it's hard to know if they're legitimate. With impersonation scams so common, a system to verify a caller's identity is crucial for all financial institutions. Ideally, phone companies would be able to identify a caller's company automatically. But until government regulations require this, I don't think the telecom companies will implement it. Till then this verification system seems like the best solution we have for now.