If you want to generate a heatmap of existing text, you will have to take a different approach here.
The naive solution I could come up with would be really expensive with openai, but if you have an open source model, you can write up custom inference that goes one-token-at-a-time through the text, and on each token you look up the difference in logprobs between the token that the LLM predicted vs what was actually there, and use that to color the token.
The downside I imagine to this approach is it would probably tend to highlight the beginning of bad code, and not the entire block - because once you commit to a mistake, the model will generally roll with it - ie, a 'hallucination' - so logprobs of tokens after the bug happened might only be slightly higher than normal.
Another option might be to use a diffusion based model, adding some noise to the input and having it iterate a few times through, then measuring the parts of the text that changed the most. I have only a light theory understanding of these models though, so I'm not sure how well that would work
I've spent the last few months using Claude Code and Cursor - experimenting with both. For simple tasks, both are pretty good (like identifying a bug given console output) - but when it comes to making a big change, like adding a brand new feature to existing code that requires changes to lots of files, writing tests, etc - it often will make at least a few mistakes I catch on review, and then prompting the model to fix those mistakes often causes it to fix things in strange ways.
A few days ago, I had a bug I just couldn't figure out. I prompted Claude to diagnose and fix the issue - but after 5 minutes or so of it trying out different ideas, rerunning the test, and getting stuck just like I did - it just turned off the test and called it complete. If I wasn't watching what it was doing, I could have missed that it did that and deployed bad code.
The last week or so, I've totally switched from relying on prompting to just writing the code myself and using tab complete to autocomplete like 80% of it. It is slower, but I have more control and honestly, it's much more enjoyable of an experience.
Drop in a lint rule to fail on skipped tests. Ive added these at a previous job after finding that tests skipped during dev sometimes slipped through review and got merged.
I do like the control you have over organizing. The only thing I was shocked was I couldn't revert a series of changes (checkpointing) like you can can in Cursor. Sometimes the LLM does it a different way I dont like, so I want to go back and edit the prompt and try again.
I grew up in a town called Seymour, which is in the same county as Gatlinburg/Pigeon Forge. It's Dolly country. She's a hero - she's done so much for her community. Much of the money you spend at Dollywood goes to help so many public causes. One of them is the Imagination Library, which gives out free books every month from birth to 5 years of age. I was one of the first families that got to take a part in it. I certainly don't remember much from when I was 5, but I do remember getting those books. I imagine it had a positive impact on my growth.
We'd go to Dollywood a few times a year - she would give out free tickets to people who worked in Gatlinburg to go. It's really well run, and their water park is great too. Growing up, we'd ride the train when we visited. While I didn't appreciate it much as a kid, when I grew up I realized how awesome of an opportunity that was.. I moved away from Tennessee about 12 years ago, one of the biggest things I have missed is Dollywood and their big steam train.
Words cannot describe the love and admiration myself and the vast majority of Tennesseans have for Dolly Parton. She is the closest thing to a living Saint that many of us will ever witness.
I discovered her music 'recently'. The 'imagination library' is a thing she has been doing for years, and I read about this on CNN or BBC or something like that, and I was deeply moved. When I read more about it (if I remember well) she started this initiative because her father couldn't read, and she wanted to make sure every kid has books to read. When I read that last part I thought to myself "shit.. this is a KIND person" (like wow!! she is a great human being!).
She bought the property that is now Dollywood specifically to provide both amusement and jobs to people in East Tennessee.
Imagination Library is mostly an umbrella organization that provides logistical support to the many, many locally-funded groups that actually send the books.
She's very kind, and apparently very charming and down-to-earth despite having a deity-quality voice, but she's also really smart and knows how to leverage her celebrity to help people.
If you've never listened to it, watch https://www.youtube.com/watch?v=CMrfM711vXI - it's "Jolene", one of her most famous songs, but slowed down like it was a 45 RPM single played at 33 1/3 RPM like an LP. Not a single note is off, even that slowly.
I can't really show an interactive demo, but my team at my day job has been fine tuning OpenAI models since GPT-3.5 and fine tuning can drastically improves output quality & prompt adherence. Heck, we found you can reduce your prompt to very simple instructions, and encode the style guidelines via your fine tuning examples.
This really only works though if:
1) The task is limited to a relatively small domain (relatively small could probably be misnomer, as most LLMs are trying to solve every-problem-all-at-once. As long as you are having it specialize in a specific field even, FT can help you achieve superior results.)
2) You have high quality examples (you don't need a lot, maybe 200 at most) Quality is often better than quantity here.
Often, distillation is all you need. Eg, do some prompt engineering on a high quality model (GPT-4.1, Gemini-Pro, Claude, etc.) - generate a few hundred examples, optionally (ideally) check for correctness via evaluations, and then fine tune a smaller, cheaper model. The new fine tuned model will not perform as well at generalist tasks as before, but it will be much more accurate at your specific domain, which is what most businesses care about.
200 examples at most, really?? I have been led to believe that (tens of) thousands is more typical. If you can get excellent results with that few examples, that changes the equation a lot.
Could you elaborate on what better way to communicate "I feel like you aren't listening to me"?
Also, this dynamic is one of a boss to employee, not of a relationship peer. The boss in this instance needs to communicate something very clearly: "This is a fire-able offense if this keeps happening." Is it better in this instance to be more aggressive?
> Could you elaborate on what better way to communicate "I feel like you aren't listening to me"?
The problem with that phrase is it is trivially untrue - the person is listening to him. It would be like me saying "you're not reading what I'm writing"; you are reading it. You might not understand it, you might not agree with it, maybe the writing is even improper in some way. But you can't write a response to it if you didn't read it. The concern he's raising isn't a real one and Jerry can't address it because he's already listening and he has no control over how his manager feels about said manager's misconceptions.
You'll notice that Jerry had to spend a few rounds of begging and pleading to figure out he needs to say "Yes, I will be accurate and careful from now on, so you can trust what I say. If I have a concern, I will raise it with you directly and honestly" to get through this and end the encounter. There are two things to note here:
1) Manager should have led with this if that is what he wanted. Not "I’m worried you’re not listening to me" but "To make sure you've heard what I've said, could you please repeat back in your own words".
2) The manager shouldn't be asking for Jerry to repeat stuff back to him in that way in the normal course of events, it is somewhat unprofessional/a stupid power play. Nothing Jerry is committing to in that sentence is a real change in behaviour. He was probably already trying to be accurate and trustworthy. He didn't realise there were any concerns here until his manager exploded. Changes in behaviour would be something like "There was a conflict between X and Y, in this instance I prioritised X. Next time I will prioritise Y." One of the issues is the manager did such a bad job of steelmanning and drawing out Jerry's reasoning behind the behaviour we can't tell what he did wrong. Many of the accusations ("attitude problem", "didn't like Jerry's tone", "not working in good faith", etc) are absurd and ungrounded. In my opinion we can't really figure out what Jerry's mistake was from the article; Manager wanted him to write certain code, he wouldn't because [reasons] and we aren't being told in any depth what [reasons] were. They may have been bad [reasons] but the manager should be interrogating and dealing with them instead of complaining about being "listened to".
"I feel like you aren't listening to me" happens to be a pretty classic phrase used by people who are inarticulate by the way. They can't get a message across and they don't think to look at their own communication to find the problem; but they know that they can't accuse the other person of being too stupid to understand (the other classic :[ ). So the problem becomes that somehow the other person just isn't listening.
I'm in my mid-thirties and most of my friends have ditched Facebook. I didn't really realize this until when I used it to create an event for a house party... I was somewhat surprised that only 2 people out of 15 even saw it. I ended up resorting to good old text message and that worked, but it was tedious. Not sure how popular this will become, but having a social-media-less event invite/broadcasting system would be nice, and having one that most people with an iPhone have access to covers much of my friend base
I thought email was a common denominator but I learned most people don’t check email or check it rarely. So different from the days when everyone had email.
I still use FB and so do many of my friends my age (mid to late 40s). But a bunch have also migrated to Instagram.
Among the younger generation, you’re a millennial if you’re on instagram because they’ve moved to TikTok. FB folks are over the hill. There’s a generational divide and pride in being trendy.
WhatsApp is only a thing among my international friends — many Americans don’t have it.
The only universal now is text messages but it feels so clunky (even with iMessage).
I wonder if it is rooted in similar things though. Right, like with email. People don't really read or check emails because spam became a serious problem. Then with social media, looking at facebook, there is definitely a big different in ad space in facebook between the time I used to use it to now. Where ads have effectively become the "spam" equivalent for social media. Ultimately, did success of these technologies also lead to its demise. Email was so good, so it made sense for a market of spammers. Facebook became a prime place for ads, and as ads become more and more of the platform, people started to consciously or subconsciously step away to other platforms.
>People don't really read or check emails because spam became a serious problem.
With the tabs in Gmail, very little leaks through to my primary inbox that isn't relatively immediately relevant (and not a lot of mail total). Often don't look at Promotions at all and maybe glance at Updates once a day or so.
Email is useful for me though, yes, a lot of my interaction with my circle of friends is over texts.
The problem for me is not so much real spam, this gets filtered. The problem is the massive amount of work required to unsubscribe or clean up automated emails from apps and websites, both transactional and non-transactional.
I know way too many techy and non-techy people who have thousands of unread email messages from those apps.
A lot of people I know don't really answer to real email anymore, unless they know something is coming. It became just something you use to make accounts with.
Even corporate email is dying. 99% of my inbox is transactional emails from SaaS apps and spam from apps I forgot to delete. And 90% of the rest is spam from recruiters or people trying to sell me some product. Only 0.1% is legitimate.
Statistically, email is not for people anymore, period.
Experiences differ. I did go on unsubscribe jags from time to time at my last employer because I ended up on email lists from a lot of events.
But really, I get 5-10 emails a day now in my primary inbox and I don't really have many filters. I DO get a lot in Promotions and Updates, but most of the stuff in Promos I can safely ignore and I mostly keep my eye on Updates if I'm expecting something I might want to deal with there.
Email is still my primary channel for the most part.
There is still a lot of "spam" if you don't spend the effort creating filters or unsubscribing to the new notification list that companies like to make every few months. Hell, my inbox is covered in invoices, receipts, disclosures, required actions, ToS changes, etc., even though I've spent some time setting up filters for some of the common receipts.
Sounds like lots of transactional mail for services you signed up to with that mail. Sign up to less crap or use a different mail from the one you use to communicate with real people.
I used to use a separate email when I ordered things etc. Once Gmail tabs came in, I pretty much stopped doing so because it was too much trouble to monitor a second email address because I actually care about receipts, order tracking, etc. a lot of the time.
I think you've hit the nail on the head of the problem.
A lot of comments online claim that people don't care about spam, or think that advertisements are a good thing for a free service, or at the very least won't change their habits if given an alternative. If that's the case then what's a better explanation for your observations?
I argue that people do care, even if it's perhaps not expressed in words.
We have a family email domain for my extended family, administered by a few retired but very tech-savvy relatives (both had long IT careers) and it’s roughly 50:50 whether a message sent to everyone@ lastname.com will actually show up in people’s inboxes or not. It’s probably 75:25 that a reply all to that list will show up, but modern email is a dumpster fire.
Is this using some cloud-based email host where you don't have any control over the spam filter? Otherwise, whitelisting (verified) senders from your own dowmain should be very much possible.
E-Mail isn't some magic that randomly drops mails. Mail servers are even resilient against network problems and will retry dilevery MANY times. What you are describing is NOT normal and would make using it for business basically impossible, which is not the case since email is still the primary b2b communication method for many companies.
I uses the business version of Office 365 for e-mail. It works well. I never have a problem with e-mails not being delivered or going into a SPAM folder. I am not saying your family did anything wrong. What I am saying is e-mail works well for some people.
Yeah, unfortunately that seems to be the best way to handle this kind of thing but unfortunately that costs $6/person/month so our ~50 person casual email list for organizing fantasy football and family reunions would cost almost $4k/year.
I don't remember the exact timeline but I think SMS became free (bundled with mobile phone plan) in the US before WhatsApp became popular. And most of us don't interact via chat very much internationally. So (probably) most people just default to SMS/iMessage unless there's a reason to do something differently. And even the one person I regularly communicate with chat in Europe, we default to Facebook Messenger.
I'm in my mid 40s, my friends mostly use email for organising events more than a week or two in the future, google chat or WhatsApp for more spontaneous things.
Very occasional FB invites for things when casting the net wide, like, I'm back in town and having a picnic, everyone come.
My wife is late 40s and just deleted her facebook account, and she's the most FOMO person I know - and she did this because of zuck capitulating to trump. A lot of people have had it with companies supporting fascists.
Lol so you/your wife were OK with all the spying and manipulation via ads but not being negative enough towards the democratically elected president is where you draw the line? Hysterical.
Yep, groups was essentially all I used FB for until we moved to Discord (which much better for us), I was so glad when I could stop checking FB completely.
Problem with Discord is you have to enforce real names otherwise you have to limit it to people you know.
Young people I know (except for gamers) find Discord a bit sus because you don’t have any baseline with regard to name or profile pic. Also who already knows who. Discord doesn’t expose any social network outside of the specific server.
You would think Discord would be the community of choice for Gen Z but in reality it’s limited to gamer and gamer adjacent folks.
Turns out identity and known social network are still things people look for to achieve a base level of trust for real time chat.
Reddit and HN are more topic driven, but chat somehow feels more personal.
It's also the only bit of Facebook that hasn't turned into an endless stream of trash. I expect that not to last either, if you're looking for an idea then a localised marketplace alternative with social proof should be on your radar.
For a long time they were heavily promoting "Ships to You" non-local goods. Annoying. Lots of dropshipper type stuff rather than a local unique items. Marketplace seems to have backed off that in the last year(s) though, my feed seems very local, one-off, and "real.
It still has a lot of trash, but 90% of it is trash you experience as a seller. Scammers are still really common, and I doubt the moderation has gotten much better since I failed to sell an empty aquarium because they couldn't be convinced it didn't have fish in it (although based on everything else on Facebook, there probably is just no moderation now).
For people in their early 20s to mid 30s in the NYC area, I'm starting to see mass adoption of an app called Partiful for managing social invites and events, it has a lot of nice features for sending invites, RSVP management, sending text blasts out to attendees (you can schedule reminders the day before or whatever).
My first thought. I’m surprised it’s not everyone’s first thought. Everyone in the bay that I know uses that for parties. Clearly every tech company is aware off the ubiquity of that app at least
Yeah, this is straight up f.lux 2.0 where Apple saw an idea take off, and unlike 'Nightshift' where they connected it to their new 'Health' product to stimulate Apple Watch purchases, they connected Apple Invites to social behaviors to stimulate iMessage and iCloud adoption and revenues.
My social group also uses Partiful. It works great, but it's a little worrying that it's so useful while being free: I can't see how this possibly could make money, so I assume the enshittification is coming any second now.
I can't imagine something like this is expensive to host, minus perhaps the text messages. But presumably they could charge for those (and make a little off the top).
There’s no right or wrong answer but as someone who used to work in publishing, the typeface seems contrived to be amateurish. I wonder if it’s supposed to evoke a more “authentic” unpolished feel? (YouTubers actually find too much polish reduces engagement among younger people)
The pictures are also a bit amateurish but this is more a function of the inviter. On other platforms much of the design choices are made for you so there’s a lower bar but for me, partiful seems to want to hit the kind of “having street cred” aesthetic.
It’s remarkable how this has changed. Back in what I call the “Facebook golden age” (2012-2016), before it turned to complete crap, it was unthinkable to host an event that was NOT organized by Facebook. I recall throwing birthday and holiday parties and all I had to do was scroll through my friends list and invite everyone and that was that. Everyone would see it and everyone would RSVP.
Here friends just send a message on WhatsApp. I do not know anyone who has hosted a house party of 79800 people so that they are struggling with this. But then again I guess some geographies have it more complicated, isn't it?
A (for most of the world, in any case) possibly surprising fact about the US is that WhatsApp is not very popular there.
This indeed causes problems when wanting to create a quick ad-hoc group for a party invitation etc., if at least one of the invitees is not an iPhone user.
The only reason I have WhatsApp is that a couple non-US friends use it from time to time. No one I know in the US does anything other than standard text messaging whether or not it ends up being iMessage.
It causes problems if one of the iPhone owners isn't an active iCloud+ subscriber:
> Creation of invitations requires an iCloud+ subscription.
This isn't about making life easier on people, this is about getting you to subscribe to Apple's services for access to a REST API. Apple gets some benefit of the doubt, but this is literally Slop-as-a-Service.
I can't tell what you're arguing here – are you misunderstanding what you quoted from the FAQ? Only the person who creates the event needs to have an iCloud+ subscription. Everyone else can RSVP to it regardless of whether they have an iCloud+ subscription or even an Apple device at all.
> Do invitees need to have an Apple device with the app to attend an event?
> Apple Invites is for everyone. Guests don’t need the app, an Apple device, or an account to RSVP to an event.
> having one that most people with an iPhone have access to covers much of my friend base
Luckily - you don’t need an iPhone or iCloud account to receive an invite and RSVP to it. Might be harder (or impossible?) to add to photos and music, but you can still get an invite and RSVP to it.
I'm still on facebook and a lot of my friends still are, the main problem we have with facebook events it that almost no one sees them.
This section has been over loaded with suggestions to event you might have no links with of things your remote friends are going to take part of.
Yes, I was also a big Facebook user in my twenties and now I'm in my mid-thirties and it seems Facebook became a lot less useful for this decade of my life.
For the birthdays of children in my social surroundings it seems the best practice has become to create an image with the details of the birthday party. Usually a photo of the birthday child with written Alice is turning 3. Join us for an afternoon of fun at Address on Saturday 16:00.
Usually shared on Whatsapp either in direct messages or in an existing school group if you are inviting the whole class or in ad-hoc group created for the event literally called Alice Birthday Party
I don't think it's that they've entirely ditched FB, but FB is genuinely terrible at surfacing event invites. It would prefer you to have to scroll through a bunch of irrelevant garbage in your feed that it had "recommended" instead so the product team can high five themselves over badly designed engagement metrics rather than worry if the users don't actively despise their product.
Yep I second this, I usually find out that I've been invited to an event when someone makes a post in the event, and often not from the event invite itself.
People currently use Instagram stories for this a lot and it's absolutely wild how Meta hasn't caught on and built in any sort of infrastructure for you to save and keep track of events.
Since Apple was too lazy to make it into a standard, it will probably go the way of App Clips. Niche idea, too few users to adopt it and no stakeholders with enough control to make it popular on other platforms.
iCalendar is RFC 2245[0], written by Microsoft and Lotus. CalDAV is RFC 4791[1] written by Apple, Oracle, and CommerceNet. Those are examples of open standards that Apple happens to use, but aren't Apple standards in the sense that they're something they cooked up internally by themselves.
The story is that these young engineers built a 500$ drone that consumes this kind of data to do mapping. In 24 hours for a hackathon no less.
If the US government didn't already have this kind of tech, they would spend millions just for the same prototype they built. And probably tens or hundreds of millions for a final product.
Is 2.9% and 6.8% acceptable or even 'good' levels, even for the tech industry? I'm not quite sure. A 6.8% performance-related quit/termination seems pretty high from my limited anecdotal experience.
Edit: now that I'm thinking about it, I don't think I worked at a tech company where people were fired/asked to resign more than 2.5x than those that resigned on their own.
2.9% would be excellent if sustained. It would correspond to keeping a significant majority of the employees you want to keep for more than 10 years. That's generally quite rare.
I'm less clear on how to assess the 6.8%. It seems somewhat significant, though if you're hiring many people, that's a period where you might expect churn, as some of them don't work out.
Of course, you can't extrapolate any of this, as 2023 was a year when employees would be very averse to moving, and it was also a year when many companies were coming off of previous hiring sprees. So expect the 2.9% to eventually increase.
> Is 2.9% and 6.8% acceptable or even 'good' levels, even for the tech industry?
I would put 2.9% at the very good to low level. It suggests 100% turnover every 33 years, which is fine, especially for the tech industry.
6.8% for performance strikes me as an indicator of very bad hiring and/or onboarding. A charitable view would be that many years of bad hiring got dumped in one year (so each year only had a small % of bad hires), but I wonder if that was the actual case.
"Non-regrettable" just means the company wasn't too sad to see them go, not that they were necessarily forced out. They could've been a poor performer that found another job on their own, and they wouldn't want to rehire them.
The naive solution I could come up with would be really expensive with openai, but if you have an open source model, you can write up custom inference that goes one-token-at-a-time through the text, and on each token you look up the difference in logprobs between the token that the LLM predicted vs what was actually there, and use that to color the token.
The downside I imagine to this approach is it would probably tend to highlight the beginning of bad code, and not the entire block - because once you commit to a mistake, the model will generally roll with it - ie, a 'hallucination' - so logprobs of tokens after the bug happened might only be slightly higher than normal.
Another option might be to use a diffusion based model, adding some noise to the input and having it iterate a few times through, then measuring the parts of the text that changed the most. I have only a light theory understanding of these models though, so I'm not sure how well that would work