Nothing mentioned on the status page but I get no push notifications, received some emails and got the following errors. I understand this is a Beta feature but this is basic happy path functionality…
personally i have no idea how to use this beta i haven’t found documentation yet well i came to the forum just for that but ur the first post i found to talk about it
It’s available via the website only.
yeah i saw but what does it do anyway what is the purpose how will this be useful?
oh nvm it just like a glorified notification app it lame
This is a useful tool, but it appears to be the first version. You can tell it to perform internet searches after a certain amount of time and to remind you of things. It’s helpful, but still in an early phase—once they implement more features and make it more conversational, it will be much better.
This tool, when implemented in a dialogue mode, is very useful because you can request an explicit reminder, such as a scheduled purchase at a certain time or an important date reminder. Once the agents are added, they will form part of its core. For example, when sending messages, you can tell it to send a message to someone at a specific time, effectively acting as a personal planner. It can also send a message to a second person or show you results from a website, performing periodic checks to provide you with information on events such as sports competitions. In short, with proper development, it will be a very useful tool.
The UI is a bit weird.
Allow the reminder and what needs doing by chat but manage it in a different page.
Why not have it managed by chat, or totally setup using a single line in the management page.
Also the management (editing) page is broken for non US locales due to the date formatting (as workaround open chrome://settings/languages and select US English as the default and reload the chatgpt website).
Friends, remember that tech companies are always several steps ahead. When they show us this tool, it’s because it was the first version they had a year and a half or two ago. Most likely, they now have this tool in a much more sophisticated version, probably integrated with other capabilities, but they obviously won’t release it.
Releasing the most advanced version would leave them with nothing new to offer, which is why they gradually roll out the older versions. This is how research and development areas work: you research, develop, research again, develop further, and when you’ve made enough progress, you release what you initially had. This gives you room for future products and ensures that competitors can’t catch up. Otherwise, by the second day, they would have already copied everything, and with bad luck, they might even surpass you.
I’ve worked for companies like these (and bigger) and respectfully they don’t do that
It’s always get this feature out asap to keep ahead and bring money in and fix the bugs and tech debt later.
As far as I can see the white whale is being chased (gpt5/o3) when they should be concentrating on agentic AI like a certain competitor that can now control a desktop PC.
Background tasks + push notifications have been wanted a while, this isn’t new, as have family accounts (with age ratings to allow non PG banter), more pay tiers, and a better integrated Windows desktop app.
Even basic stuff like allowing previous chats to be recalled is something being rolled out that has also been halted due to bugs.
The problem is the tail is wagging the dog, the POs are chasing the whales with features and neglecting the low hanging fruit and existing Plus customer base who are getting second class treatment.
Why on earth would ChatGPT be useful over a rotary phone except in a call centre, or “old” features like reinforcement learning taking the backwards step that it needs configuration and file uploads which might as well have been named LORAs, when reinforcement learning could have been made more general and adaptive/on the fly during conversations too?
By low hanging fruit and neglecting the Plus base I mean chatgpt still can’t have an avatar assigned to it like the custom GPTs nor can it call out to custom endpoints. So one might think, ok I’ll clone my main settings into a custom gpt and use that, with memory - oh wait you can’t because custom gpts can’t use memory.
Ok so maybe I can stump up and use my main got with API calls, I’ll pay for extra API usage instead… exactly the same problems.
I really enjoy hearing your opinion, and I completely understand you.
It’s true that sometimes companies chase money, which often leads to releasing things without proper planning or patience.
Honestly, I’m not sure if that’s the case here because, in my understanding, this functionality is part of another system, and this version isn’t refined to a high level. That’s why I lean toward the idea that they’ve released a basic version instead of the one they might already have.
I share these frustrations. I see the potential, but the pragmatic majority has written it off as currently useless. My attempts to use it have mostly failed without any explanation why. Testing is impossible—tasks require a minimum of 15 minutes to execute. Why?
I remember seeing an OpenAI job listing that required candidates to “be capable of working in ambiguity.”. I imagine it’s hard to connect features to models.
I could create something similar to Tasks using clever programming and Actions, but why bother? The name and concept have been tarnished by the mostly unimpressed users. Do I push forward or just twiddle my thumbs?
It feels like this was developed to check a box. There is no passion behind this release. As it stands, the only practical use I can see is delivering continuous streams of selective data, such as daily news updates.
Coincidentally, OpenAI’s latest news release is another partnership with a news corp.
I’m deeply disappointed. I’d love to integrate this with my own data sources for people using my GPTs, but there’s no indication of when, or if, that will be possible. I’m still surprised that I can’t even use o1
with GPTs. There’s so many features that would revolutionize how businesses interact with data left on the table.
Consumer needs seem increasingly irrelevant, overshadowed by internal obligations that miss the mark entirely.
Here are the bugs I’ve found so far
- Setting a non custom date shows for example Saturday on the edit page, but on the summary of tasks a day earlier (Friday)
- The settings page is broken for non en-US (English US users) because it can’t handle the date format (workaround is to switch your browser language to en-US)
- It can’t randomise the time (e.g. send this reminder 5 times between 9am and 10pm daily). Every single attempt will put the five times in the first hour which exceeds the 4 permitted, or will do it 5 times bang on the hour each day.
- Setting the schedule by chat is so prone to error they should give up, each regeneration adds a different days/times/frequency!
- Deleting the chat threads reminders were made in breaks the reminders (missing conversation errors in multiple places)…
- Regenerating a schedule adds a new one, so if you want to update a schedule, later which will it pick?
- If you provide a table of times to run daily, it’ll just shove them in a one hour slot and break due to exceeding 4 in an hour again
If anybody at openai is reading this please just keep it simple instead (KISS), minimise MVP by doing the following:
- Remove the chat integration except for basic recall of the reminder and dates. (Task prompt on the settings page remains where you declare what is to be done). This encapsulates the settings in a single place and reduces complexity and bugs of chat thread management.
- Provide a calendar and frequency in the settings page, make sure “random” is also available as an option (e.g. I want irregular checks for good news during a day, but not specifically at the same time each day).
- When the reminder CTA is clicked open a dedicated notification page shared between all of a user’s notifications (per project if reminders are about different topics - i.e. the reminder can be assigned to a project in settings). This fixes the missing/deleted threads issue and allows a quick lookup area for reminders if you have a lot of threads.
- Fix the locale checks, add some basic unit tests!!!
- Fix the wrong (different) day being shown between the settings editing page and its summary (again unit tests!!). Might be worth making a common lib for this one chaps!
- Don’t use (what I assume is) 4o mini for the execution of the task - it’s too stupid and messes up (and has been messing up my schedules when trying to get them right). It also confuses me with it - I ask it to do something and it reminds me I need to do it
why I don’t have the tasks function on the web? I am a pro user
I have to report another weird issue.
I’ve set up several recurring tasks that should, every morning, scan the headlines of major newspapers in Italy and provide me with a summary of the most important news titles.
However, over the past few days, I’ve noticed that the system keeps showing me the same headlines, as if it’s not performing a fresh scan of the web but instead relying on some cached data.
This behavior makes the task less useful, as the goal is to get updated headlines every day.
I’m curious if:
- Does ChatGPT actually perform a fresh web query for recurring tasks each time?
- Are there limitations on retrieving new data for recurring tasks?
- Is there a way to force updates or avoid this issue?
I found this with search in general too.
The way I’ve got around it is to set up multiple notifications and vary the language and/or topic.
I think what’s happening is there is no consideration of the chat context when the search is performed bringing back the same result AND caching (tested via having ChatGPT assistant hit a personal endpoint, and noticing it didn’t make new requests for subsequent lookups even with no-cache headers set).
The combo of these two is causing it to happen more frequently.
This will be a good functionality when it works, but I agree, it does not. I’ve set tasks but there’s no follow-up. All notifications are enabled but I never get any. The only benefit to me for now is that ChatGPT remembers my tasks when prompted, so when I cue it, like, “You didn’t send me that list you were supposed to,” it will complete the task then without me having to go into all the detail again.