The real news for me is GPT 4.5 being deprecated and the creativity is being brought to "future models" and not 4.1. 4.5 was okay in many ways but it was absolutely a genius in production for creative writing. 4o writes like a skilled human, but 4.5 can actually write a 10 minute scene that gives me goosebumps. I think it's the context window that allows for it to actually build up scenes to hammer it down much later.
Cool to hear that you got something out of it, but for most users 4.5 might have just felt less capable on their solution-oriented questions. I guess this why they are deprecating it.
It is just such a big failure of OpenAI not to include smart routing on each question and hide the complexity of choosing a model from users.
One of the major advantages and disadvantages of LLMs is they act a bit more like humans. I feel like most "prompt advice" out there is very similar to how you would teach a person as well. Teachers and parents have some advantages here.
Low security should use passwords. None of that fancy &@73gdb-Whb stuff. Just a regular word. Suitable for Netflix and meditation apps that want a basic login to prove that you paid.
Medium security should use magic links and a simple password that you don't need to write. If you lose your email, the password prevents hackers from taking over your app. If you lose your password, hackers can't take over your device. Suited for something like social media or MMOs, which are targeted very often.
High security might need proper 2FA with auth app, password rotation, stuff like that. Probably shouldn't be necessary unless there's constant active attempts to hack. Everyone gets attacked, especially in the era of AI, but I'm saying at least 10 attacks a day.
You can also layer on extra levels of security, but IMO that's about the level you should expect from users.
Paid APIs are definitely a thing. Payment gateways do this.In the LLM era you can do this too - lots of data processing to be done.
But compare Aider/Claude Code with Cursor. One is much more popular than the other two despite being similar quality. GUIs are there to lower the learning curve.
In this era, it's not too hard to make a simple app in a day with AI as well. Cursor goes much faster than Flutterflow or Bubble. The bar is higher too - you now probably need a prototype before a pitch deck. It's probably not scalable, but that's prototyping for you.
There are almost none because everything goes obsolete in a month. In 6 months, it's deprecated. Nearly every book or guide you find will be outdated by the time it's released and the people who write them are getting exhausted with this loop.
I once wrote slides on how to give AI context so it can code. When I asked a friend to proofread it, GPT-4 came out and it was smart enough to make all my techniques obsolete. I updated my workflow for it, and then Cursor came out, basically automating all the indexing and stuff in seconds while I'd take days.
There's YouTube, but they tend to be brief and look like hype or unedited. Most of the good stuff is experimental, but experimental isn't helpful and shouldn't be. If you want names: techfren and Sabrina Ramonov
The new GPT-4o image update is just amazing. It compounds with the new memory update less 24 hours ago. I could write pages about why but I'd rather just make stuff with it. Anyone who's getting excited about it but not using it is a bit of a hypocrite.
Just treat it like a skateboard. No book. You'll fall many times at first. Have fun with it. You don't have to go anywhere, just go around the block. Join one of the many hackathons out there and do something with it. It'll click.
Make good use of Cunningham's Law - just exclaim that it can't do Y, and someone will explain how it does Y. Like the skater community, people want to see you do tricks.
It was like the search engine era. Everyone threw money at everything and Google won. Looks like Cursor might be the next Google at this rate. But there's still other things to win at in the AI arena, so there's probably some FOMO.
I do believe there's room for segmentation on agents as well. Manus and ChatGPT Deep Research are fairly different and they likely access different sources. Grok seems to play mostly on social media and so on. I think companies need the kind of depth that Cursor has. GitHub tried to do a bit of everything and failed despite their resources.
We use it too, but anything that requires actual funds, we would migrate to AWS or refactor.
Like we'd use to allow PMs to change banners/copy on the fly, but Supabase is better. It does RAG but Supabase does that better too. Firebase functions are really good for prototyping, but once it starts to cost money, we move it to AWS. It's good for feature flagging, and we moved that to Growthbook. Analytics started on Firebase and were moved to open source so we'd own the data. The only thing they do well is Crashlytics and that's free.
If everyone was as bad a customer as us, I would assume they're in trouble.
Unfortunately a lot of people have trust issues with Google. We don't want them controlling data and certainly nothing as core as functions, DB, and feature flagging. Then when people don't use these things, Google kills it real fast.
I suppose, but firebase is used in some applications, though of course a lot of them are just during testing, so assuming everyone was not like you they would make profit, but assuming a mix, they would make only a bit of profit, so it would be more like a small side project/nice feature that google provides that makes them a little bit of money.
They didn't shut down Parse though. You just had to self-host it. But somehow when Big Tech drops support for anything, people just migrate out of it for no reason and it becomes a self-fulfilling prophecy. Like Flutter careers are completely dead in the water now. One of my businesses has been running on Parse Server for half a decade and there's still updates on Parse. And despite being "dead", I still find Parse easier to code with than Firebase, though Supabase is now the favored option.
Fair point. But only a small minority of back end as a service users are comfortable with self hosting something like Parse.
Enterprises really like having someone to sue when things go wrong. And indie devs who pick BaaS offerings do so in large part to avoid the hassle of self hosting.
I'm not sure if it's still valid but it's not too hard, and still much easier to maintain than dealing with AWS. I have basic AWS proficiency, but it's still good to have Heroku to not think about it.
Google laid off the Flutter team. They still seem to be releasing Flutter announcements but it's slow.
One big part of Flutter was that it was a hedge against Java lawsuits back in the day. The other was that it was experimenting with some mobile code, like designing it around lifecycles and all the async UI. But now the same paradigms have been adopted by iOS and Android.
Unlike BE, mobile moves quite quickly too; the last few years have been deprecating a lot of things like push notification security and file access. Mostly to deal with dark patterns. So a platform that isn't in active maintenance tends to fall apart quickly like we've seen with many mobile libraries, Cordova, etc.
reply