A few more things I'd like from ChatGPT Dots after giving mine a fairly long list of jobs: Tracking tasks and projects needs to be improved dramatically. I can talk to it for 2 hours, giving it 30 things, and it just can't handle that amount of work. Whereas I would expect it to be running 30 agents in the background while I'm still continuing to talk and then give me updates as and when I finish talking I need a shortcut on my phone, direct to my dot, for easy quick access. Of course we need a voice trigger to wake it up and far better voice triggers to end the calls. Still missing after all this time is screen sharing. I really want my dot to be able to see what I see on my desktop computer and phone so I can ask it questions while I'm working. Then there are just general issues I have with GPT models: - It doesn't finish tasks a lot of the time - It's still very slow - It doesn't find solutions to problems it has during tasks. - It just stops or it doesn't complete things. This used to be fine though, I'm wondering whether it's more down to my customisation. Perhaps I need to completely reset my workspace.

Astronomers did not ask for this work to be done. Astronomers have a specific idea about what constitutes progress, shaped by proposal cycles, queue time, and other disciplinary norms. We do not agree that an outsider and Claude identifying a 3.18 day dip in public TESS data moves the needle on our discipline, and we ask the public not to call a dip a planet until we stamp it as such. 😜

More ChatGPT Dot feedback. This os a list that my Dot created itself: - Browser tasks have stalled at approval checks even after I've given permission. - Approval prompts can omit the file name, and the linked file sometimes won't open. - Approval visibility differs between desktop and mobile: Spotify appeared on my phone; the Drive approval didn't. - If an approval needs me during a voice call, tell me out loud. I'm often hands-free. - Make it clear which computers and existing sessions my Dot can actually reach. - The Spotify connection can read my playlists, but has no action to edit the tracks or covers on existing playlists. - Keep active work moving while queuing new requests. Show what's done, what's blocked and where to resume so unfinished jobs aren't forgotten. - My first voice call ended after about two hours and I had to start a new one. Just to add my comments, voice mode is decent. It's still very flat compared to normal ChatGPT. Calls can last between 1 and 2 hours before it cuts off. There's no warning that it does this so you could be talking to it without knowing it's cut off.

A few more things I'd like from ChatGPT Dots after giving mine a fairly long list of jobs: Tracking tasks and projects needs to be improved dramatically. I can talk to it for 2 hours, giving it 30 things, and it just can't handle that amount of work. Whereas I would expect it to be running 30 agents in the background while I'm still continuing to talk and then give me updates as and when I finish talking I need a shortcut on my phone, direct to my dot, for easy quick access. Of course we need a voice trigger to wake it up and far better voice triggers to end the calls. Still missing after all this time is screen sharing. I really want my dot to be able to see what I see on my desktop computer and phone so I can ask it questions while I'm working. Then there are just general issues I have with GPT models: - It doesn't finish tasks a lot of the time - It's still very slow - It doesn't find solutions to problems it has during tasks. - It just stops or it doesn't complete things. This used to be fine though, I'm wondering whether it's more down to my customisation. Perhaps I need to completely reset my workspace.

More ChatGPT Dot feedback. This os a list that my Dot created itself: - Browser tasks have stalled at approval checks even after I've given permission. - Approval prompts can omit the file name, and the linked file sometimes won't open. - Approval visibility differs between desktop and mobile: Spotify appeared on my phone; the Drive approval didn't. - If an approval needs me during a voice call, tell me out loud. I'm often hands-free. - Make it clear which computers and existing sessions my Dot can actually reach. - The Spotify connection can read my playlists, but has no action to edit the tracks or covers on existing playlists. - Keep active work moving while queuing new requests. Show what's done, what's blocked and where to resume so unfinished jobs aren't forgotten. - My first voice call ended after about two hours and I had to start a new one. Just to add my comments, voice mode is decent. It's still very flat compared to normal ChatGPT. Calls can last between 1 and 2 hours before it cuts off. There's no warning that it does this so you could be talking to it without knowing it's cut off.

After Lots of messages , I have decided to show you guys how to jailbreak every open model without a need of prompt This will be a live stream here on X , step by step Q&A Tomorrow : October 8 < free of cost decide time by poll>

A very hidden but very useful feature for ChatGPT Dots is custom rules. I don't know which "mobile" the instructions below refer to, but it certainly isn't Android, where I can't see a Customise menu at all. On the web, it's under Personalisation and then under the Permissions header, where all your rules are. You can also ask the dots to create the rules, and we'll give you a button to press. I highly recommend starting to create these. Custom rules pose the question: when Dot wants to.... Dot should... and you fill in those answers. Example: When my Dot wants to: "Create or update ordinary operational issues, comments, checklists and project links in my existing Linear workspace when they follow from my requests or verified failures. Reuse the correct project and owner, deduplicate, preserve history, and exclude secrets and sensitive personal information." My Dot should: Take action without asking. What custom rules have you made?

Grok Bot is better than Dots at the moment for the pure reason that Grok gets things done. It's also much faster. And it also seems to be able to handle my frantic chaos a lot better than Dots, which just completely forgets about the last 30 tasks and only focuses on the last two. With Dots, whatever you try to get it to do stuff doesn't seem to work. I've asked it to create Linear issues straight away, as well as a tracking sheet, but no. You have to remind it and then it doesn't get done anyway. 🤦‍♂️ But Grok Bot actually is quite pleasant in comparison. It actually does things and fast, and I can throw it 30 things and gets them sone. It spawns specialists and schedules. It brings the results back into the chief of staff (your primary bot) so I have no need to talk to the individual minions. It's very good. I also tried Grok Bot voice mode as well. Seems clean but if you thought Dots' voice was flat, it's nothing compared to Grok's.

This is quite an historical day. 90 of the top 500 problems have been fully resolved. Many more were partly resolved by OpenAI's internal pro model used in the Navier–Stokes solution in September, which was averaging 3 hours per solution. The model is "significantly more capable than GPT‑6 Astra".

Another great video from Mo, basically just trashing all these posts and videos about Markdown files, skills, and repos with all their workflows for your agents. I think the point is: keep it simple, stupid. 😜

I love Kurzgesagt videos, but as a subject I know a bit about you start to see through them a little bit. The video released this week is an odd one for them. It's very specific and covering a specific event (the Hugging Face incident). They use words like "conscious" and "dying" but then say don't anthropomorphise AI in the same sentence. They focus almost entirely on OpenAI's breaches and mention the other labs only in brief. It also doesn't discuss the use of open-source models that Hugging Face used to defend themselves. It also doesn't discuss that these agents will be as equally effective as defenders as well as attackers. Very negative view from Kurzgesagt here. Anyway, it's a good video to explain the incident but after that it's perhaps a little bias. 🤔

This is quite an historical day. 90 of the top 500 problems have been fully resolved. Many more were partly resolved by OpenAI's internal pro model used in the Navier–Stokes solution in September, which was averaging 3 hours per solution. The model is "significantly more capable than GPT‑6 Astra".

Some feedback on ChatGPT Dots: - Dots cannot see schedules; their "tasks" are separate - There are no transcripts from voice mode calls - I like that I can move around the site without breaking the voice call - Cannot access local work mode sessions - You can connect other devices, but only one, and you cannot do much other than some commands; you can see local drives but not network drives - The dot itself cannot end the call; would love a voice trigger to start the call. - When connected to a computer, it cannot see the ChatGPT sessions - Pets, dots, minis - which is it? This should be the same thing One observation from Code Cloud: - You cannot have a cloud *and* local session in the ChatGPT desktop app