Notes: Now this I’m really into. Hit the stars button in the Notes app, and AI can auto-format and summarize the messy notes you took during your hourlong meeting. You know, the one where you jotted everything down but didn’t really listen to what was said? A similar summarize tool is built into the web browser, too.
https://www.wsj.com/tech/samsung-galaxy-s24-artificial-intelligence-features-dc37e134?page=1.
I don’t know. Macworld recently speculated that “Apple’s biggest opportunity may be in the least-sexy area of AI: building features for convenience and productivity. Just as Microsoft is building AI tools into Office, Apple will eventually have to build AI features into its iWork suite (Pages, Numbers, Keynote). It’s not the kind of stuff that grabs headlines or makes memes, but it can be very useful.”
Samsung has partnered with Google, and they been working on AI for at least seven years. I’ve read that Apple has been using AI to improve their hardware and photo processing, etc. But Siri sure hasn’t received an AI love.
I’m kind of with you on this. My intro to AI is more of me dipping my toes into the water. I’m sure I’m thinking way too small, but I can’t wait for AI to help ease the mundane burden of a lot of my admin tasks. Use recent text messages received and suggest times to meet up with people. Reminding me to circle back with some friends if I said I would. Help me at work summarize documents for my boss.
I just used ChatGPT to help me solve a Python “problem”. It was insanely simple and something I could have coded myself (and I wasn’t using Copilot or anything), but I asked it to generate code to sort an excel file (again, this is NOT normally something you would need Python for and definitely not AI). What I found intriguing was when it didn’t work (I capitalized a column heading in the excel file but not in the code), I copied back the output to the AI. It then re-wrote the code with debugging built in. Then I copied that output back to the AI. And it solved my problem, rewrote the code, and it worked.
It’s the mundane stuff I’m interested in so AI can help me be more effective, whether that’s in my personal life or at work. I see it as an assistant in my life.
But Apple doesn’t do mundane. Apple gives us new features that make work and play even more powerful. Widgets to personalize the Mac and get more done “while stunning new screen savers, big updates to video conferencing and Safari, along with optimized gaming make the Mac experience better than ever.”
All I want is a digital assistant that works.
While I don’t necessarily disagree, I would point out I think the new autocorrect based on their LLM is pretty mundane…and FANTASTIC. It’s one of their most positively-reviewed new features from what I can tell.
That’s good news. I can’t say that I’ve noticed any difference in AutoCorrect, but that could be because I’m still using an iPhone 11
We should probably see if it’s actually any good before we worry about what Apple is doing to compete with it. Samsung generally is not great about software, they make nice hardware, but they tack a bunch of bloat on top of it on their phones.
Fair point, but I hope Apple is waiting to see how Samsung does. They need the pedal down on this and should be embracing AI for the consumer-facing parts of their software.
While LLM functionality is useful, so far majority of these solutions require submitting your notes/content to a company’s servers to get them processed. That is problematic even if they promise not to use it for training. I try to keep as much of my content in end-to-end encrypted apps (including Advanced Data Protection) and as far as I know, no one has implemented tires in an encryption preserving way.
I am hopeful Apple will do some sort of cool thing with online/offline hybrid approach to solve this.
What they did with iCloud Private Relay was and is an amazing way to hide your IP address from websites without the hassles of dealing with an always on VPN.
They did great work with differential privacy too.
I’ve found ChatGPT pretty useful when I’m having an off day and not seeing the simple fixes.
For me, this would be nightmare. The essential activity for me is to go through those messy notes to understand and organise them, identify the missing bits, remind myself of what needs more attention than I gave it at the time, follow up on the (all too frequent) incomprehensible abbreviation. If my oh-so-efficeint AI PA does it for me, I’ll end up with well formatted and nicely presented but content, but al the value of the post-meeting activity will be lost
I’ve been there because at work I take copious meeting notes --some people think I use an AI transcription tool, it’s just me typing at the speed of light. Then I need to take the time to review everything and, like you, I feel that this process is exactly where my value as a consultant resides: synthesis, conceptualization, decision, and the narrative to explain it to another business stakeholder. It takes a lot of time for me.
But your comment made me think… why is it valuable for us to do the post-meeting activity? What’s that value for? For whom? You seem to imply that it’s for you, but you need that value in order to make a decision yourself or help someone else to make that decision. Why go through our lengthy notes to extract the main points that were discussed, the agreements and key action points that people commited to when that can be done by the PA itself?
I cannot be more excited about that because I think that in the long(ish) term we will still be at the helm, just with faster information. And if some day the PA becomes smart enought to make these decisions for me, well that day I will know I need to go look for some other activity to make a living.
There is much more “information” than the notes. Post-processing notes from something that I’ve attended taps into my memory of things that maybe I didn’t even realise at the time - a facial expression or tone of voice from a participant; how fluent and fluid an argument was; the sense of who might be bulldozering or too passive; my lifetime of experience and connections to ideas and people. Processing notes is a pretty good way of tapping into all that and reinforcing my own understand as well as expressing things in a form ready for an audience or collaborator.
An AI might be able to summarise my notes, but they can’t possibly summarise my intuitions and deep knowledge, and that could be disastrous.
Another thing I’ve been experimenting with AI…
I’m sliding into a new position at work pretty soon, and the type of job I’ve done before but this time it’s in a new topic area if that makes sense. So I’m trying to read up on products/reports that are coming out…some are from Congress and others are from within my organization. Some apply to how things should be done/how they can be improved. Others are budget type documents that are newer to me. I’ve been feeding the information (all public, nothing proprietary) into ChatGPT and having conversations about it. I’m distilling 350+ page reports into a conversation with the software.
Very cool.
I just did the same, using the Leo AI in Brave browser (which uses Mixtral and Claude Instant and Llama 2 13b for chats) for some basic help cleaning up old versions of Python. I’m not a programmer and Python is only installed when I’m trying to use some tool that requires it. So I had a real mishmash of versions, NONE of which are currently being used for anything (as far as I can tell). The AI chat guided me through finding out where things were and if they were being used and even how to redo the ‘Current’ symlink and point it to the new current version (3.12).
As I mentioned, I’m no programmer, just a copy-paste hacker with just enough savvy to understand basically what is going on and why. But I rarely actually code anything. So AI Chat saved me untold hours scrolling through Google search results. To me it’s like having a chat with a savvy Python expert who flips through the manual and spits out relevant steps. IT SAVES ME TIME. It rather makes doing or fixing things a nice quick task vs something I just put off because of the headache…
For me it’s getting weird and wonderful SQL statements that I can’t quite work out for myself. Usually works. Playing with sample perl scripts usually works too and when it doesn’t I know enough to put it right.
However, getting ChatGPT to generate R code has been fraught with frustration. Code that does not work. Uses functions from unknown libraries. Given up on this and reverted to hard slog of reading introductory texts.
When I graduated in 2003, Google was hiring anybody graduating in AI. I had three colleagues who finished their Ph.D. in machine learning in that same year, and they are all still working at Google today. Since then, I’ve seen almost all my friends who have done Ph. D.s in AI end up working there (especially before the latest GenAI hype). So, Google has been developing its AI capabilities for over 20 years!!
I recommend trying JetBrains AI—it is far better than OpenAI for coding. It uses multiple LLMs for different tasks, and the code quality is of a magnitude better. I only use it, in general, to generate small code snippets, such as individual (React) components and functions in data science scripts. I don’t use it to generate complete applications, as I prefer to handle the main program logic myself and want to ensure it is secure and efficient.
My biggest criticism is that it often generates imperative code, while I always code deterministically. Additionally, it doesn’t always create pure functions. So, it often needs some tidying up.
It’s worth it as it makes me 25% more efficient at least.
Took a look at JetBrains AI web site but am going to stick with OpenGPT for now as it has two advantages over JBA. Firstly it is free-to-use (at least for 3.5) and secondly does not require a programming-languae specific IDE to integrate in to for use. Old school software engineer here using either MacVim or occasionally emacs. And as my current “work” is not lengthy product development projects — more tinkering with things to support annual/potentially quinquennial political activities — the need for a sophisticated programming oriented AI LLM is not critical. Plus ChatGPT generated code already requires “tidying up” so maybe there would be no real advantage for me to switch.