ChatGPT 5.1 Launches With Group-Enhanced Dialogue Tools  ## Metadata - Author: [[Hard Fork AI]] - Full Title: ChatGPT 5.1 Launches With Group-Enhanced Dialogue Tools - Category: #podcasts - URL: https://share.snipd.com/episode/6aa9f8f2-1c5f-4e2d-9ead-66c2f536e9b7 ## Document Notes ## Highlights - **Small Fixes Change Perception** - GPT-5.1 fixed small but visible UI/format quirks like em dashes that signaled AI-written text. - Public reaction shows AI hype fatigue where micro-updates feel more notable than grand AGI claims. Transcript: Jaeden Schafer And you have to kind of go through and backspace them all. Anyways this problem has been solved so sam haltman was recently on twitter and he posted he said small but happy win if you tell chat gpt not to use m dashes in your custom instructions It finally does what it's supposed to do this is all with the gpt 5.1 update now how have people been responding to this? Someone posted a meme right under his thing, which says like, open AI in 2023. And then it's like the meme of like the doge dog. And it's like super ripped. It's like the bodybuilder. And then it's saying like, AGI any day now. And then right next to it, it says open AI in 2025. And it's like a small wimpy doge dog crying. And it says, you can tell chat gpt not to use m dashes so i mean i think this just kind of goes to show we've had a lot of ai hype people are a little bit jaded that we talked so much about agi in The last two years and now we are making posts about m dashes being removed from chat gpt with the latest update of gpt 5.1 i think that's funny someone else said open ai killed another Startup and they just have some code that says text replace mdash with nothing. So anyways, it's funny. ([Time 0:01:01](https://share.snipd.com/snip/6534a418-9541-47d5-8a7c-f0d49d9c253b)) - **Six-Word Glitch Persisted In Memory** - Jaeden Schafer recounts testing the new six-word response constraint and how the model remembered it across chats. - He had to explicitly remove that memory because the model persisted the one-off instruction. Transcript: Jaeden Schafer If that really was somebody's startup, it deserved to die a sad death. Someone said, I told it to stop using mdashes. It replied with an mdash, peak betrayal. And they showed their screenshot saying don't use mdashes and it used an m dash in the response so it evidently is not perfect i'm also going to say something that drove me crazy with this New gpt 5.1 update is that um in the in their like demo they're like you can tell it to respond using only six words so i was like testing it and i was like okay only respond using six words Tell me a joke um and then it was like told me a joke and i was like hey that's great. Tell me another one. It told me another one, but it used seven words. And I'm like, that's using seven words. And so I was like, Oh, I'm sorry. And it like fixed it to six words. Okay. That was my experience. And I'm like, whatever. It's like sort of good. It's not perfect. And then any message I sent to chat GPT today, after that, it would, I'd be like, Hey, like write, write a paragraph about X, Y, Z. And it would like write it in the right six words. And I'm like, what is this? And it's like, I'm only using six words, like you said. And I'm like, Oh my gosh, I like three times it kept doing that on different things. And I will, you know, I'd always be like, Hey, you can use more than six words and continue. But then like the next time I'd start a new chat, it would have remembered that. So I don't know what that glitch is, but watch out. It saves things to memory that are like obviously one-off things. Like you say, just respond using six words. I'm not saying for all of eternity, only ever respond with six words. Like that's ridiculous. Somehow it interpreted it that way. So watch out. And then I just very explicitly had to say, remove my command for six words from your, you know, from your history or your memory. And I was like, okay, we removed it from the memory. ([Time 0:02:12](https://share.snipd.com/snip/e99916f1-af09-44ae-90ab-318ea1a74902)) - **Sane Model Naming Returns** - OpenAI simplified model naming to GPT-5 with 0.1/0.2 iterations to reduce confusion. - Clearer naming helps users know version progression without guessing capability tiers. Transcript: Jaeden Schafer It's a lot more empathetic. It's a lot more kind. I've seen the demos. I've seen it respond. Overall, I think it's good, but you do have to watch out. It does have some funny quirks, like every single OpenAI rollout that I've seen them do. There are always funny quirks. So watch out with chat GPT 5.1. By the way, GPT 5.1 really felt like it just like slipped under the radar. One thing that I do appreciate that they're doing there is that the naming convention is a lot more sane, not GPT 4, GPT 4.0 mini ultra max. Like just, I don't know. I just feel like there's like all these random weird words and you never knew what model was the best for anything. Uh, apparently now they just have GPT-5. It's going to be 0.1, 0.2, 0.3 until they decide that it's a completely new model. Then they'll go to GPT-6. We don't know what's entailed or when that model might come out of training, but overall it feels like they're just kind of making these minor improvements. I'm happy that we have stuck to a sane naming convention. I think they were really worried and trying to get away from naming anything GPT-5 for a long time because it had so much hype built around it. And that's why they came up with all these weird names. It seems like we're just onto normal names now. ([Time 0:03:52](https://share.snipd.com/snip/ec463126-f7bf-4469-bd87-94110f6acf10)) - **Use Group Chats For Structured Workflows** - Try using group chats for collaborative document workflows where people take distinct sections to edit. - Send the chat to a single collaborator to continue work rather than relying on large unmanaged groups. Transcript: Jaeden Schafer All right. At the beginning of the episode, I was telling you about a new group chat feature on ChatGPT that OpenAI is rolling out. It's in select countries, and it is on the free tier. It's on any tier. There's a couple things that I find interesting with it. I mean, the first big question is like, what is this for? Like, you could imagine maybe a group assignment for a class where you like ask questions and everyone can ask the ChatGPT questions about the assignment and get like responses. Like, perhaps, I don't know, that doesn't seem incredibly useful to me when you could just probably use independent ones, but maybe it is. Maybe you are, you know, starting a project, you get like half of the document or the job done and you send it over to someone else to like review it and tweak it. So maybe that's, maybe that's kind of useful actually. Yeah, I guess, I guess that's probably a better use case. I'm talking myself into it. They have like the ability for you to have up to 20 people in the group. And to me, like thinking about like a group chat with 20 people, it's seems like out of control. But if there's like a, if you kind of have like a pre-setup workflow, like, okay, I'm going to go work on coming up with the outline for this. And then I want you to go and flesh out like this component. And then Jeff's going to go flesh out that component. And Seth is going to go flesh out that component, right? Like if you kind of have like an overview of what's going to happen ahead of time, maybe in a normal group text chat, then everyone can go work on the document and everyone can get their Part of it done on a project. So maybe that's useful. I actually think the most, like the very, the most common thing this is going to get used for is just sending it over to one person where you're like, hey, I did XYZ, can you go take this document And do this other thing to it? And you just send them the chat and they can keep working on it. ([Time 0:04:59](https://share.snipd.com/snip/c675d0e4-ff04-40db-b8d2-d3103980357c)) - **Scale Needs Structure** - Allowing up to 20 people makes group chats powerful but potentially chaotic without pre-defined roles. - Pre-setup workflows and role assignments are essential to keep large AI-assisted group threads useful. Transcript: Jaeden Schafer And to me, like thinking about like a group chat with 20 people, it's seems like out of control. But if there's like a, if you kind of have like a pre-setup workflow, like, okay, I'm going to go work on coming up with the outline for this. And then I want you to go and flesh out like this component. And then Jeff's going to go flesh out that component. And Seth is going to go flesh out that component, right? Like if you kind of have like an overview of what's going to happen ahead of time, maybe in a normal group text chat, then everyone can go work on the document and everyone can get their Part of it done on a project. ([Time 0:05:50](https://share.snipd.com/snip/c429b0ad-b274-4b80-9189-a028e489381b)) - **Respect Age Filters And Safeguards** - Be mindful of age-based filters: chats with minors automatically apply content restrictions. - Use parental controls and safeguards when including users under 18 in group threads. Transcript: Jaeden Schafer One interesting thing that it does is that if users are under 18, the content is filtered. Uh, opening eyes obviously started opening up their content to be a little bit, uh, a little bit more wild, like grok or XAI. And so it's interesting that if, um, you have multiple people in the same chat thread, if people are younger, below a certain age, it's going to get filtered to only things that are, uh, Appropriate for them. There's going to be safeguards and parental controls in place. So it's going to be interesting to see how this, um, how this plays out. ([Time 0:06:39](https://share.snipd.com/snip/077e9443-14d5-4589-80b2-7c843b15e38c))