AI is a UX Problem, Code Tools for More Productivity, and Many-Shot Jailbreaking
Product-centric news in the AI world.
Morning y’all!
I’m working from home today and I’m always really grateful to do it, especially as a lot of companies are moving from full-remote to a hybrid approach. I’m grateful for the work but I know that I’m much more productive when I’m in the comfort of my own home and much, much closer to the food that I like. ※\(^o^)/※
But I’m ok with it for now. Culture can be built in a number of ways but statistically the hardest way to build culture is hybrid — most companies will eventually figure this out.
Today is a bunch of product-centric thoughts and industry news that I hope you find interesting and worthwhile! Thanks for subscribing!
\( ﾟヮﾟ)/
— Summer
Princeton NLP has developed SWE-agent. an open source system that turns GPT-4 into a software engineering agent that can help solve GitHub issues. This could unlock quite a bit of productivity for developers.
Replit introduces Code Repair, a low-latency AI agent that auto-magically fixes your code. Again, enhanced productivity at the click of a button.
Jailbreaking is a fun way to test the security of an app and an interesting way to find out what’s possible. Anthropic did a deep-dive into it in large language models. I imagine more of this will become an issue for folks and product builders as the necessary expansion of data grows in demand and need.
Yahoo, in an attempt to stay relevant, acquires an AI news app from the Instagram cofounders called Artifact. Yahoo and “news” has a long history so while it doesn’t surprise me I think it would have been better if they just built their own.
Building an app with prompts is more about the design of the prompt and less about engineering it, at least according to the author of this thinkpiece:
In prompt engineering, we can’t predict how the model will respond to the prompts we craft. Instead of direct programming, we’re steering the model towards the outcomes we aim for.
A good reminder that it’s about humans, not data.
Elecit looks pretty useful, if research papers are your thing.
Ship faster code with AI! It’s a devtool that reviews pull requests and converts GitHub comments into working, tested code.
Apple has been thinking about using the Chinese app Baidu (“Ernie”) for their capabilities but clearly there are privacy concerns, as well as UX issues that have to be considered and solved. It’ll be interesting to see how far this one goes.
Can Intel make a serious comeback and not entirely lose their position in the greater market? At this point it seems like it’s a life-or-death scenario.
Microsoft releases a GitHub repo that has some very popular courses on LLMs, including 18 lessons for beginners. Not a bad place to start, to be honest.
4 optimization techniques for mobile phones and LLMs. I think the next real battle to win will be the device that sits in everyone’s pocket. Winner take all.
Upscayl is a free and open source image upscaler! You can download it and use it for yourself or they have a paid version for macOS if you want to support the developer (or get a .dmg file here for free).
An interesting study on studying 30,000 school policies across 1,000 school websites. The results are fascinating, but, unsurprising.
A fun podcast to put on the background about generative AI and Slack’s use of it in their own community and product. I have to use Slack every single day and although I really am not a fan it does the job.
And that’s about it! Have a great Wednesday folks!
※\(^o^)/※
— Summer